# Chronology of Signals, Silos, and Sovereign Software - Research ## A Unified Historical Chronology of Military Computing, Intelligence Fusion, and Machine Language, from the Kammhuber Line to the Artificial Intelligence Platform This chronology merges five source transcripts into **one continuous timeline**. Every sentence has been rebuilt to stand on its own: material in square brackets restores the subject, object, date, place, or referent that the original sentence assumed from its surrounding context. Auto-transcription errors in the names of people, companies, programs, agencies, and places have been corrected from context, and a **Corrections Ledger** at the end records each correction and each place where a source's factual claim was adjusted or flagged. Each entry closes with a source key. **Source key.** _ABC_ — Matt Bevan, "How did Palantir get so powerful?", _If You're Listening_, ABC News In-depth (May 8, 2026). _PLTR_ — John Coogan, "Palantir: The App That Caught Bin Laden." _MSFT_ — John Coogan, "Microsoft's Military Empire, Explained." _VALLEY_ — John Coogan, "Old Military Tech Created Modern Smartphones." _LLM_ — "An Old Idea" (explainer on next-token prediction in GPT-2). **Source creators.** **John Coogan**, narrator of the _PLTR_, _MSFT_, and _VALLEY_ transcripts, identifies himself as the cofounder of **Soylent** (soylent.com) and **Lucy** (lucy.co), both funded by **Y Combinator** (Soylent in the Summer 2012 cohort, Lucy in the Winter 2018 cohort). **Matt Bevan** hosts _If You're Listening_ for ABC News In-depth. These are provenance facts about the narrators, recorded here rather than inserted into the timeline. --- ## I. Antecedents (1755–1939) **1755.** Benjamin Franklin wrote in a 1755 letter that "those who would give up essential liberty to purchase a little temporary safety deserve neither liberty nor safety." [Franklin's formulation] is the most famous statement of the centuries-old philosophical problem [of liberty versus security that Peter Thiel would later confront in founding Palantir]. — _PLTR_ **c. 1849–1855.** The California Gold Rush [of the mid-nineteenth century] brought legions of people out to the American West. California was always the place where men and women could come to make their fortunes [through risky business ventures]. Leland Stanford, before he established Stanford University, was a businessman who made his fortune selling picks and shovels to gold miners. — _VALLEY_ **c. 1910s–1920s.** Vannevar Bush, originally an engineer, received his PhD jointly from MIT and Harvard. [Vannevar] Bush's inventions in the field of **analog computing** led to his promotion to senior positions in science administration. Frederick [Fred] Terman said he had been interested in science and electrical engineering since he was ten, eleven, or twelve years old. Fred Terman completed his undergraduate and master's-level studies at Stanford [University] in the 1920s. [Fred Terman then earned his doctorate at MIT, where Vannevar Bush served as Terman's thesis advisor.] After graduating, Fred Terman went back to California to join Stanford's engineering faculty. [At Stanford, Fred Terman] designed a course, created a new laboratory focusing on research into **vacuum tubes**, and wrote one of the most important textbooks on electrical and radio engineering. Fred Terman could have carved out a great career in relative obscurity [at Stanford had World War II not intervened]. — _VALLEY_ **c. 1929–1934.** [Reginald Victor] Jones, [later a British physicist and military intelligence expert,] loved practical jokes. Once, when he was a student [at Oxford], Reginald Jones called a classmate and convinced him to drop his telephone into a bucket of water to clean out the bugs. A few years later, Reginald Jones convinced a reporter that he had invented a device which, when attached to a telephone, allowed Jones to see everything the reporter was doing. The reporter was skeptical until Reginald Jones described with perfect accuracy how the reporter was dressed. Reginald Jones knew how the reporter was dressed because Jones was peering down at the reporter from the next building. — _VALLEY_ **1930s.** Many people believe that if the Japanese had not attacked Pearl Harbor, the United States would never have entered [World War II], but America's leaders had already been preparing [for war] for years [before the attack]. The ominous storm clouds gathering during the 1930s were impossible to miss. In Germany, Adolf Hitler rose to power by exploiting resentments about the aftermath of World War I. In Spain, a group of fascist military officers staged a coup [in 1936] which led to a brutal civil war. In Asia, the Japanese had invaded China and were occupying Manchuria. Tensions were rising across the world, and every part of society had to respond. In the United States, the scientific community was led at that time by Vannevar Bush, an absolute giant in the history of American science. — _VALLEY_ **1937.** Two American brothers, [Russell and Sigurd Varian,] deeply alarmed by acts of unprovoked German aggression, collaborated with Stanford [University] to create a device [the **klystron**] which eventually proved to be the key to microwave-based radio detection. — _VALLEY_ **September 1, 1939.** When the Nazis invaded Poland on September 1, 1939, Vannevar Bush held the prestigious position of president of the **Carnegie Institution** [of Washington, now the Carnegie Institution for Science]. Vannevar Bush could see that the United States desperately needed closer cooperation between its civilian scientists and the military. All across the United States, thousands of talented men and women were working on technologies that could change the outcome of the war, but [those scientists] were not encouraged to share or commercialize their results. Vannevar Bush wanted [American civilian scientists] to be closer to the action. — _VALLEY_ --- ## II. The Electronic War (1940–1946) **1940 (Battle of the Beams).** During the Blitz, Reginald Jones was one of the first people to realize that the Germans were using **radio beams** to guide their bombing runs. Reginald Jones's love of tricks and deception filtered through to his [wartime intelligence] work. — _VALLEY_ **Early 1940.** Just a few months before President [Franklin] Roosevelt created the National Defense Research Committee, researchers at [the University of] Birmingham invented a device called the **cavity magnetron**. Microwaves had long been considered one of the most promising radar technologies because microwaves could carry more information using less power. Before [the cavity magnetron], there was no reliable way to actually generate [high-frequency] microwaves. The cavity magnetron was a huge step forward because it made high-frequency microwaves viable. — _VALLEY_ **June 12, 1940.** Vannevar Bush secured a meeting with President [Franklin] Roosevelt on June 12, 1940. Vannevar Bush brought a single piece of paper describing his idea for a new agency that would oversee research on the problems underlying the development, production, and use of mechanisms and devices of warfare. President Roosevelt approved [Vannevar Bush's proposal] in ten minutes. — _VALLEY_ **June 27, 1940.** Two weeks after [Vannevar Bush's meeting with President Roosevelt], the **National Defense Research Committee** (NDRC) was born. By enlisting some of America's best scientists and engineers directly in the war effort, Vannevar Bush knew he was going against the grain. The National Defense Research Committee focused on accelerating the development of two key military technologies: the **atomic bomb** and **radar**. The potential of radar to identify and track military targets was widely understood, and important [radar] work was already going on all over the world. — _VALLEY_ **September 7, 1940 – 1941 (The Blitz).** Adolf Hitler wanted to take Britain out of [World War II], but Hitler knew that a land invasion against a heavily defended island would fail. [Adolf Hitler] turned to his commander of the German air force, [Hermann Göring,] and decided to bomb British industrial targets. For months, German bombs fell on British cities in a campaign known as **the Blitz**. Starting on September 7, 1940, London was bombed by the Luftwaffe on 56 of the following 57 days and nights. [The Blitz] was terrifying, but thanks to a combination of British resilience and ingenuity, the Blitz failed. [The failure of the Blitz] opened the door for an Allied counterattack. — _VALLEY_ **September 1940 (Tizard Mission).** Shortly after the [cavity magnetron] breakthrough, President [Franklin] Roosevelt and Prime Minister [Winston] Churchill agreed that their countries should share their top-secret research. British scientists showed the cavity magnetron to their impressed American counterparts. American military leaders wanted to manufacture [cavity magnetrons] at home, and [American] scientists wanted to work on further improvements [to the magnetron]. The Americans and the British agreed to establish a new jointly administered research facility [for microwave radar]. — _VALLEY_ **October 1940.** [The jointly administered Anglo-American microwave research facility] opened at MIT in October 1940 and was called the **Radiation Laboratory**, or the **Rad Lab** for short. The name [Radiation Laboratory] was deliberately deceptive, chosen to make people think that the laboratory focused on nuclear physics research at a time when almost no one believed nuclear physics had viable military applications. With money from the National Defense Research Committee, the Rad Lab grew rapidly. At its peak, the Rad Lab employed more than 4,000 people. The Rad Lab was the perfect example of what the physicist Ernest Lawrence called **"big science."** — _VALLEY_ **1940–1941 (Big Science).** The idea behind big science was simple but revolutionary: fields like physics and electronics had advanced to the point where most cutting-edge work was too complex for small academic laboratories. The era of individual physicists doing an experiment at their desk was over [by the 1940s]. In place [of the lone desk physicist], [cutting-edge science] needed big, multidisciplinary research teams working in laboratories full of sophisticated equipment. Running [big-science] laboratories required scientists, engineers, managers, accountants, and support staff such as secretaries and janitors. Big science was too expensive for the private sector, which left just one source of funding: the [United States] government. Proponents of big science saw [government-funded big science] as the perfect marriage of America's scientific prowess and its prodigious industrial manufacturing capabilities. [Proponents of big science] convinced the government that science could solve the country's biggest problems. — _VALLEY_ **1941 (Kammhuber Line).** [Following the Blitz,] Allied air commanders knew the Germans had overcommitted, but [the British] were still cut off from the massive land battles raging on the European mainland. [The Allies] launched a major strategic bombing campaign of their own, aimed not at sowing terror but at crippling the Nazis' ability to wage war. Executing [the Allied strategic bombing] plan required taking back control of the skies. Thousands of fighters and bombers took off from British and American airfields, piloted by men barely older than boys, flying straight into the belly of the beast. Even by the standards of World War II, aerial missions were insanely dangerous. Over the course of the war, half of all Allied aircrews were killed while on operations. Of the many threats Allied pilots faced, one of the most formidable was the **German air defense network**, which the Germans brought online in 1941. British pilots named [the German air defense network] the **Kammhuber Line** after the colonel, [Josef Kammhuber,] who devised it. The first version of the Kammhuber Line consisted of several overlapping radar stations covering a zone about 20 miles long and 12 miles wide. [The Kammhuber Line's] radar stations were linked by **Himmelbett** zones. _Himmelbett_ means "canopy bed" in German, but [the Himmelbett zones] were much less inviting [than the name suggests]. Each Himmelbett zone consisted of a **Freya** radar with a range of more than 60 miles. [Each Freya radar] controlled a master searchlight. Each Himmelbett zone was also assigned one primary and one backup fighter plane. Royal Air Force pilots flying into Germany had to cross the Kammhuber Line. When [RAF pilots crossed the Kammhuber Line], the Freya radar operators used the master searchlight to spot [the RAF aircraft] and then sent night fighters to shoot them down. [The German] night fighters were some of the best planes the Luftwaffe had, and they were flown by some of Germany's top pilots. The Germans soon developed a more accurate tracking radar called the **Würzburg**, which they added to each Himmelbett zone. One Würzburg [radar] would lock onto the German interceptor while a second [Würzburg radar] would lock onto the Allied plane as soon as the Freya picked it up. Each Würzburg radar had a 45-mile range and was fully mobile. When the Germans suspected more bombers were incoming or that their radars could come under attack, [the Germans] would simply put [the Würzburg radars] on a truck and move them somewhere else. Initially, the Kammhuber Line was devastatingly effective. British commanders just did not have an effective counter [to the Kammhuber Line]. [British] pilots were guided by their instincts rather than by tactics, flew their own paths toward German airspace, and were easily picked off. Something had to change [in British bombing tactics]. — _VALLEY_ **June 28, 1941.** In June 1941, the United States government established the **Office of Scientific Research and Development** (OSRD). Vannevar Bush was appointed [the Office of Scientific Research and Development's] inaugural director. Before World War II and before Vannevar Bush, military research and development was performed exclusively in military laboratories. Vannevar Bush's fundamental insight was to enlist America's universities and scientists in the war effort. The Office of Scientific Research and Development funded universities directly, in sums of tens or even hundreds of millions of dollars, which was huge money for the 1940s. Vannevar Bush's time at the Office of Scientific Research and Development transformed the relationship between academic science and the United States government. — _VALLEY_ **Early 1942.** One day in early 1942, Fred Terman's telephone rang, and the caller was Terman's old thesis advisor, Vannevar Bush. Vannevar Bush explained that [the Office of Scientific Research and Development] was about to open a new laboratory to conduct top-secret research on **electronic countermeasures**, an emerging field of technologies that blocked enemy radars and communications. Vannevar Bush believed that electronic countermeasures could provide the Allies with the decisive advantage they needed to win the European air war once and for all. Vannevar Bush wanted Fred Terman to run the [electronic countermeasures] laboratory. Fred Terman did not have deep expertise in electronic countermeasures because the field was so new that no one did. Fred Terman must have had doubts, but he understood that he was being asked to do something important for his country. [Fred Terman] moved [from California] to Massachusetts to run the **Harvard Radio Research Laboratory**, helping the Allies win World War II and, although he did not know it at the time, taking a huge step toward the creation of Silicon Valley. — _VALLEY_ **May 30, 1942 (Operation Millennium).** When faced with the problem of how to safely cross the Kammhuber Line, Reginald Jones told British air force commanders to reorganize their attacks into **streams** and fly in a tight cluster down the middle of a Himmelbett zone. The new [bomber stream] tactic made its debut on the night of May 30, 1942, when about 1,100 RAF bombers attacked the German city of Cologne. [The Cologne raid,] code-named **Operation Millennium**, was the biggest bombing raid of the war so far. British commanders hoped that [Operation Millennium] would be so devastating that it might even knock Germany out of the war. [Operation Millennium] did not knock Germany out of the war, and in truth its results were slightly underwhelming, because Cologne was still standing. Operation Millennium nonetheless proved that Reginald Jones was right: fewer than 4 percent of the bombers sent on [the Cologne] raid were shot down. The German air defenses, which had struck fear into British hearts, were overwhelmed [by the bomber stream]. — _VALLEY_ **1942–1943 (German adaptation).** The Germans were not about to give up so easily [after Operation Millennium] and immediately started making important adjustments. Luftwaffe commanders began directing all available fighters within range to intercept the British bomber stream. The German war machine got to work producing hundreds of additional Würzburg radars. At the height of the war, more than 5,000 [Würzburg radars] were deployed in a line stretching from Denmark to the middle of France. Further technical improvements meant that the Himmelbett zones expanded until they reached 100 miles across. Germany's biggest cities and industrial areas were soon shielded by concentric layers [of radar coverage]. The Kammhuber Line once again became highly effective at deterring Allied bombing raids. The bomber stream worked until the Germans responded [with these adjustments]. — _VALLEY_ **July 1942 (Radio Research Laboratory).** In July 1942, the United States was fighting tooth and nail to beat the Nazis in a nationwide effort to defeat the greatest evil the world had ever known. [In July 1942,] the head of a secret government agency, [Vannevar Bush of the Office of Scientific Research and Development,] directed Harvard to create a research laboratory to build advanced anti-radar technology. [John Coogan opens his documentary "Old Military Tech Created Modern Smartphones" with this laboratory, presenting its wartime secrecy as the first causal link between World War II and modern consumer technology.] [John Coogan states that the Harvard anti-radar laboratory] was so secret that no photographs of it exist. [The Harvard anti-radar laboratory] was called the **Radio Research Laboratory**. [The Radio Research Laboratory] ultimately helped the Allies win the war, but [helping win the war] was not even its biggest legacy, because [the Radio Research Laboratory] marked the beginning of Silicon Valley as it is known today. Apple, Google, Microsoft, and Amazon owe their entire existence to an incredible series of events that unfolded during World War II. — _VALLEY_ **1942–1945 (Radio Research Laboratory operations).** Fred Terman was the general of a very new kind of army. The 850 men and women Fred Terman led at the Radio Research Laboratory were not soldiers, and they fought their battles in the laboratory rather than in the trenches, but they were nonetheless fighting. Fred Terman and his [Radio Research Laboratory] staff had one mission: to systematically dismantle every part of the German air defense system. Over the course of the war, [the Radio Research Laboratory] created more than 150 anti-radar devices. The [Harvard] Radio Research Laboratory was the dark twin of what was happening across town at the [MIT] Rad Lab. Sometimes the MIT scientists would notice their radar equipment was not working. What [the MIT scientists] did not realize was that their colleagues at the Harvard Radio Research Laboratory were turning on their jammers and interfering with [the MIT radar equipment]. Reginald Jones, the fun-loving British scientist who came up with the idea of the bomber stream, would have been proud [of the Radio Research Laboratory's pranks on the Rad Lab]. — _VALLEY_ **1942–1945 (Doctrine of radar countermeasures).** Radar countermeasures have two main uses. The first use [of radar countermeasures] is blinding the enemy's radar or tricking it into reporting wrong information. The second use [of radar countermeasures] is getting the enemy's radar to unintentionally reveal its location, effectively turning the master's creation against him. Whatever the specific task, the first step in countering enemy radar is locating where [the radar signal] is coming from and identifying its wavelengths and frequencies. During the air war over Europe, [radar] reconnaissance was done by small **ferret** planes, so called because they used search receivers to ferret out enemy radar signals. Anti-radar detectors had a big advantage over radar itself. Radar was limited to the distance at which it could pick up the echo of its own signal, whereas anti-radar devices did not need to broadcast and could simply listen in near silence. Once the enemy's radar was located, [the enemy radar] had to be jammed. Two main radar-jamming techniques were developed in the field and perfected at the Radio Research Laboratory. [The Radio Research Laboratory's work ran as a recursive cycle: ferret reconnaissance located an enemy radar, jamming blinded it, German counter-jamming answered, and the Allies reconnoitered and jammed again, the same move-and-countermove loop that governed the radar war from the Kammhuber Line through the bomber stream to Naxos.] — _VALLEY_ **1943 (Pathfinding radar and Naxos).** The next major innovation [in the air war] came when the British and then the Americans developed air-to-ground **pathfinding radar** [H2S and H2X] to help their bombers see through the clouds. The Germans realized that [Allied pathfinding radar] effectively turned the Allied bombers into big beacons in the sky. In response, the Germans invented a mobile radar receiver code-named **Naxos** that picked up [the Allied bombers'] locations. [The radar war] was the world's most dangerous game of cat and mouse: the Allies would devise a solution to increase the efficiency of their bombing raids, the Germans would respond, the Allies would respond back, and on and on it went. Control of the skies was the difference between the German war machine being stopped in its tracks or rolling across Europe. Allied air commanders knew the German air defense systems were devastating their bombers, but [Allied commanders] did not even know what radars the Germans were using or how the Germans were controlling their night fighters. To defeat the Kammhuber Line, the Allies needed to understand it, and they needed to do so fast. The same old incremental improvements would not cut it, and [the Allies] needed a complete paradigm shift, a technological innovation so big that the Germans simply could not keep up. Finding [that paradigm shift] would require transforming the relationship between scientists and the military. — _VALLEY_ **July 1943 (Operation Gomorrah and Window).** The first [radar-jamming] method involved aluminum foil. [Allied] scientists had discovered that a strip of any metallic object cut to half the wavelength of a radar signal would bounce the signal back, confusing the radar and the person operating it. Foil was chosen because it was cheap and light, so it floated in the sky. The Allies called [the foil strips] **Window**, and [Window] is known today as **chaff**. Suddenly, planes escorting Allied bombers started dumping thousands of chaff packets out of their holds, with each packet containing thousands of individual strips. [Chaff] was first used in **Operation Gomorrah**, a bombing run on Hamburg in July 1943. Clouds of aluminum foil strips dropped by Pathfinder planes and the initial bomber stream completely clouded the German radar defenses [over Hamburg]. The result [of Operation Gomorrah] was the city's total destruction. As many as 40,000 people perished [in Operation Gomorrah]. One-third of Hamburg's large factories and almost half of its small factories were destroyed [in Operation Gomorrah]. Adolf Hitler himself is believed to have been alarmed that several more bombing raids like the one inflicted on Hamburg could push Germany out of the war. — _VALLEY_ **1943–1945 (Carpet jammers and the collapse of German air defense).** The second [radar-jamming] method used **electronic jammers**, high-frequency transmitters which overwhelmed enemy radar installations with radio waves broadcast at the same frequency. [These electronic jammers] became known as **carpet jammers**. Thanks mainly to the work of Fred Terman and his [Radio Research Laboratory] team, 24,000 carpet jammers were installed on Allied bombers. There were entire factories in the United States making only carpet jammers. The mighty German radars that had been a thorn in the side of Allied bombers for years, [namely] the Freya, the Mammut, and the Würzburg, simply could not handle [the carpet jammers]. German anti-aircraft efficiency was reduced by 75 percent [by Allied jamming]. The Radio Research Laboratory team led by Fred Terman even developed a powerful ground-based jammer to use against enemy airborne radar. The Germans managed to devise some partly effective counter-jamming measures, but [the Germans] were never able to fully answer the questions that the Allies had started asking. By the end of the war, 90 percent of Germany's high-frequency radio experts, about 7,000 men, were diverted from other urgent work to try to prevent the jamming of German radar. The battle for supremacy of European skies was over, and soon so was the war. — _VALLEY_ **1945–1946.** Fred Terman built the Harvard Radio Research Laboratory from nothing into the world leader in every type of advanced microwave device. The work done by [the Radio Research Laboratory] was credited with preventing the loss of 450 bombers and more than 4,500 lives in the United States' main bomber squadron, the Eighth Air Force. With the war over, the Radio Research Laboratory had fulfilled its role and was disbanded in 1946 as silently as it had started. Fred Terman was faced with the question of what to do next. — _VALLEY_ --- ## III. Stanford, the Cold War, and the Making of the Valley (1945–1970s) **1945–1946.** Fred Terman decided to return to the place he knew best: Stanford. Fred Terman already had tenure at Stanford. After his work running the Radio Research Laboratory, Fred Terman had three more things: an address book full of high-powered military contacts, access to some of the country's best scientists, and massive amounts of funding from the United States government. Newly installed as dean of the School of Engineering, Fred Terman set about turning Stanford into one of the best places in the entire world to be an engineer. In 1945, Fred Terman was influential in the creation of Stanford's first independent research laboratory, which looked into further applications for microwave technology. Fred Terman used his military connections to help found a new electronics research laboratory [at Stanford]. — _VALLEY_ **1948.** Claude Shannon, the Bell Labs engineer behind **information theory**, tried writing by guessing the next word in a paper from 1948 [titled "A Mathematical Theory of Communication"]. Claude Shannon did not have a computer [for his next-word experiment], so he used a book. Claude Shannon would open [the book] at random and write down a word. Then Claude Shannon would flip to another page, read until he hit that same word, and write down the word after it. Claude Shannon's result begins "the head and in frontal attack on an English writer that the character of this point." [Claude Shannon's book procedure] is the same loop a chatbot runs. Claude Shannon's book trick looked back only one word. Claude Shannon's [1948] paper even says that meaning is irrelevant to the engineering problem. Claude Shannon meant [that meaning is irrelevant to the engineering of] communication systems, but [Shannon's remark] fits [the next-token prediction] loop too, because [a language model's] odds measure what usually comes next rather than what is true. The idea [of writing by guessing the next word] is as old as information theory. — _LLM_ **Late 1940s–1950s (Cold War).** Fred Terman's mission [at Stanford] was helped massively by the emergence of the next major global conflict, [the Cold War]. The United States and the Soviet Union never met each other on a physical battlefield and never fought each other with guns. Instead, [the United States and the Soviet Union] fought with radar, spy planes, ICBMs, deterrence, and intrigue, [and in that contest] intrigue mattered more than raw military force. [The Cold War] meant that someone of Fred Terman's reputation could always receive the funding he needed to do the research he wanted in areas like electronics and signals intelligence. — _VALLEY_ **c. 1951.** The [United States] military approached Fred Terman and asked him to establish an **Applied Electronics Laboratory** [at Stanford] which would perform classified military work. [The military and Fred Terman] doubled the size of Stanford's electronics program and, in effect, made Stanford a full partner with the United States military and government. [The Stanford–military partnership] was big science at its best. Fred Terman transformed Stanford into the advanced research and development arm of America's military-industrial complex. Going from massive contracts with the United States military to the entrepreneurial culture of Silicon Valley required breaking some long-established rules. — _VALLEY_ **Early 1950s (Stanford Industrial Park).** By the early 1950s, Fred Terman realized that the research he was overseeing in Stanford's engineering department and its growing number of independent research laboratories could create huge amounts of private-sector value. Fred Terman also knew that Stanford was in some financial trouble. Fred Terman did something that had never been done before: he spearheaded the creation of the world's first **university research park**. [An interviewee described] much of [the Stanford research park] as based on research with practical applications that led to [companies such as] Cisco and Sun [Microsystems]. Jointly administered by Stanford and the City of Palo Alto, **Stanford Industrial Park** generated income for Stanford and tax revenue for Palo Alto. More importantly, [Stanford Industrial Park] brought together the area's most promising high-tech companies to share facilities and knowledge with each other and with the broader Stanford community. [An interviewee said that] the degree to which Silicon Valley has its roots at Stanford is essentially the degree to which Fred Terman was able to conceptualize [the research park] and then persuade the [Stanford] trustees that his concept was meritorious. [The Stanford trustees were the decisive institutional audience, because the research park required the university's governing board to commit Stanford land to long-term leases with private companies; Fred Terman's persuasion of the trustees is the mechanism that converted a wartime research network and a federal funding stream into a permanent commercial institution.] Whether he knew it or not, Fred Terman was tapping into California's centuries-long history of risky business ventures. In a way, Fred Terman did with Stanford Industrial Park what Leland Stanford did [during the Gold Rush]: [Terman] gave entrepreneurs the equipment, the facilities, and the talent, [which is to say] the tools to make fortunes of their own. — _VALLEY_ **1953.** In 1953, [Stanford Industrial Park] got its first tenant. [The first tenants] were the two brothers who had worked with Stanford researchers to create a device [the klystron] that became a crucial part of microwave-based radio detection during World War II: Russell and Sigurd Varian. One of **Varian Associates'** first employees was a payroll clerk named Clara Jobs, whose son, [Steve Jobs,] went on to become a major figure in technology. — _VALLEY_ **1956.** In 1956, Varian Associates became the first Silicon Valley company ever to go public. — _VALLEY_ **1950s–1960s (Industrial Park tenants and the Terman doctrine).** Other notable early tenants [of Stanford Industrial Park] included **Hewlett-Packard**, a company started by two of Fred Terman's engineering students [Bill Hewlett and David Packard]; **General Electric**, [described by the source as] one of the world's biggest semiconductor manufacturers; and **Lockheed** [then Lockheed Corporation, later Lockheed Martin], still one of the world's biggest defense companies. Instead of encouraging his graduate students to get their PhDs, which had been the way things worked for decades if not centuries, Fred Terman encouraged [his graduate students] to leave [the university] and start companies. Fred Terman himself sat on a number of corporate boards and suggested that his colleagues do the same. [Terman's encouragement of entrepreneurship among academics] was heretical, but by this stage Fred Terman had years of experience doing things that had not been done before. — _VALLEY_ **1958.** [Lawrence Preston Gise,] Jeff Bezos's grandfather, was a founding member of the Pentagon's [Advanced Research Projects Agency, later the Defense Advanced Research Projects Agency,] **DARPA**. — _VALLEY_ **Late 1950s–1960s.** By the late 1950s and early 1960s, [Silicon Valley] was teeming with companies expanding the frontiers in fields like microwaves, vacuum tubes, and early semiconductors. [Silicon Valley in the late 1950s and early 1960s] was also crawling with Soviet spies. Many [Silicon Valley] companies were still funded by the United States government to perform research and make devices used as part of the Cold War. **Fairchild Semiconductor's** [integrated circuits] helped send American astronauts to the Moon. After setting up shop in Sunnyvale, Lockheed [Missiles and Space] got a contract to build all the missiles for the United States Navy [the Polaris program]. [Lockheed] quickly became [Silicon Valley's] biggest employer. Eventually, public funding [in Silicon Valley] was eclipsed by private capital. Without that critical early help from the United States military, Silicon Valley would never have become what it is today. Since [the founding of Stanford Industrial Park], companies like Tesla, Facebook, and Rivian have all been tenants of the [renamed] **Stanford Research Park**. — _VALLEY_ **1960.** J. C. R. Licklider was a pioneer in the field of **man-computer symbiosis** and an important figure in the development of the internet. — _PLTR_ **1960s.** [FBI agent] John O'Neill had wanted to be in the FBI ever since he watched the imaginatively titled television show [_The F.B.I._, a QM Production,] back in the 1960s. — _ABC_ **1970s (The personal-computer insight).** Bill Gates today is a laid-back philanthropist working on problems like global health, but Bill Gates ran Microsoft like a drill sergeant. Back in the 1970s, computers were really only used in laboratories or offices. [Restricting computers to laboratories and offices] made sense, because [1970s computers] were massive, expensive machines designed to speed up highly specific tasks. The companies that dominated the [computer] market in the 1970s believed that no average person would ever want a personal computer. Bill Gates saw things differently, [predicting a computer] in every office and in every home. Build-your-own computer kits were growing in popularity [in the 1970s]. Independent hackers were starting to use computers at home, and the technology was getting cheaper. A massive avalanche of technological progress was clearly about to begin. — _MSFT_ --- ## IV. The Personal Computer and the Enterprise State (1975–1999) **1975 (Altair 8800 and Microsoft BASIC).** The snowball that triggered [the personal-computer] avalanche was the **Altair 8800**, one of the earliest personal computers. To modern eyes, the Altair 8800 looks like an alien artifact. The Altair 8800 did not have a screen or a keyboard, and users had to operate it using a series of switches and lights on the front panel. The Altair 8800 did not even come fully assembled, but [the lack of assembly] did not stop people from buying it, and the Altair flew off the shelves. Bill Gates and his old friend from high school, Paul Allen, started writing software for the Altair 8800. [Bill Gates and Paul Allen's] first release was a version of the **BASIC** programming language that let Altair users write and run their own simple programs. [Microsoft BASIC] became, for the first generation of personal computers, the thing that unlocked the power that was there. Bill Gates was obsessed with details and ran Microsoft like a general in wartime. For Bill Gates, business was war, and competitors had to be destroyed. — _MSFT_ **c. 1980–1981 (IBM and MS-DOS).** Bill Gates was going up against an absolute giant, IBM, but giants cannot move fast, and IBM was a decades-old bureaucracy at this point. Eventually, the success of the Altair 8800 made IBM realize that it needed a personal-computer strategy. IBM's executives knew that they could not make a breakthrough product by doing things the old way, so [IBM] created a **skunkworks**-style research-and-development laboratory in [Boca Raton,] Florida. The IBM [Boca Raton] team began work on [IBM's] first personal computer, and [the team] needed software. As soon as [the IBM team] met Bill Gates, they were convinced that he was the best option. Bill Gates was still really young, but he knew more about software development than basically anyone else at the time. [IBM and Microsoft] formed an alliance: Bill Gates and his team would provide IBM with an operating system, but crucially, it would not be an exclusive deal, [and Microsoft would be able to sell the operating system to other companies]. The deadline to deliver [the IBM operating system] was tight, so instead of trying to build everything from scratch, Bill Gates bought an existing operating system [86-DOS] from a small company named **Seattle Computer Products**. Microsoft engineers modified [86-DOS] and renamed it **MS-DOS**, short for the **Microsoft Disk Operating System**. [Buying 86-DOS] was a clever move, but like a general pressing home his advantage, Bill Gates did not stop there. IBM wanted to pay a fixed price for the software license, but [a fixed price] would have limited Microsoft's upside, and Bill Gates knew that. [Bill Gates] went back [to IBM] and demanded a royalty arrangement. [The royalty arrangement] gave Microsoft a guaranteed income stream and set Microsoft up perfectly to dominate the rapidly growing personal-computer software market. — _MSFT_ **1980s–1990s.** The personal-computer growth wave of the 1980s and 1990s meant that software became a key part of how large organizations worked. Eventually, one of the biggest [organizations] of them all, the United States military, gave Bill Gates a call. — _MSFT_ **Late 1980s (The Soldier's Computer).** [The United States military's] obsession with augmented reality dates back to the 1980s. The [United States] Army had identified that technology was transforming battlefield intelligence. Satellites and drones were generating so much data that old technologies just were not keeping up, but with the right gear, raw [battlefield] data could be transformed into useful information. [The Army's] early thinking became a prototype Army project called **the Soldier's Computer**. [The Soldier's Computer] was far from perfect. The screen used for the Soldier's Computer's heads-up display [was the Reflection Technology Private Eye, the same red-LED display technology later used] in Nintendo's [1995] **Virtual Boy**. [The Soldier's Computer display] was low resolution and could only display things in red. The Virtual Boy was a big flop with consumers and critics. For some reason, military officials were determined to succeed [with augmented reality], so they doubled down and moved the Soldier's Computer project to the United States Army's **Natick Soldier Center** in Massachusetts. There was no better place for the augmented-reality technology to be put through its paces [than Natick]. The [Soldier's Computer] project's budget was increased. With more money on the line, [the late 1980s and early 1990s] looked like the moment when [military augmented-reality] technology would either prove its viability or be canceled entirely by the Department of Defense. — _MSFT_ **1991.** By 1991, Microsoft was making more than 200 million dollars a year from MS-DOS. Bill Gates had executed flawlessly, but [MS-DOS] was just the first stage of his quest for dominance. — _MSFT_ **1992 (Soldier's Computer field testing).** In 1992, the Soldier's Computer went into field testing. At first, everything was going smoothly. Soldiers were really happy about how [the Soldier's Computer's] thermal imaging increased their sight range, and the ability to fire around corners while staying in cover was truly novel. The drawbacks [of the Soldier's Computer] quickly became apparent. First, [the Soldier's Computer's] head-mounted display was extremely clunky. Data transfer [on the Soldier's Computer] lagged by up to 75 seconds, if the system did not crash entirely. The United States Army Research Institute published a report which included some very honest opinions from soldiers. [Soldiers in the Army Research Institute report] called [the Soldier's Computer] junk, said it was terrible, and all mentioned how painful it was to wear. One soldier said that whoever put [the Soldier's Computer] on a soldier's back had never tried walking the street, hopping a wall, going through a door, or coming out of a hatch with it, and that [the soldiers] just wanted it to go away. The [Soldier's Computer] package added about 40 pounds to a soldier's pack. For comparison, the **Meta Oculus Quest 2** [a 2020 consumer virtual-reality headset, which John Coogan cites as a present-day reference point rather than a 1992 product] weighs just over one pound, and people do not really like wearing even the Quest 2 that much. If people will not wear a one-pound VR headset, soldiers are definitely not going to be comfortable wearing a 40-pound VR rig. Despite [the Soldier's Computer's failures], enough people high up in the military still believed in the program, so it survived and work continued. — _MSFT_ **February 26, 1993 (First World Trade Center bombing).** [On February 26, 1993,] a van packed with 550 kilograms of explosives was driven into an underground car park of the World Trade Center [in New York City]. [The 1993 World Trade Center bomb] blasted a hole through three floors, killing six people instantly and injuring hundreds more. As New Yorkers came to terms with a possible act of terrorism on United States soil, investigators released the first pictures of the explosion site, leaving little doubt that a powerful bomb was the most likely cause. For some time, it was not clear who was behind the [1993 World Trade Center] attack. As many as 19 groups claimed responsibility for the [1993 World Trade Center] explosion. One of the FBI agents involved in the [1993 World Trade Center] investigation was John O'Neill. The people behind the 1993 attack were religious extremists [whose operational leader, Ramzi Yousef, was of Pakistani descent and was captured in Pakistan in 1995], and John O'Neill was instrumental in their capture. — _ABC_ **Mid-1990s (O'Neill and "ordered liberty").** John O'Neill became obsessed with researching Islamic extremism and quickly became an expert on the growing threat of terrorism on American soil. As John O'Neill continued rising through the ranks of the FBI's counterterrorism department, he was cognizant of the importance of balancing the need to keep people safe with the infringement on their personal liberty. John O'Neill said that if you have a lot of order, there is very little liberty, and if you have a lot of liberty, there tends to be little order. John O'Neill said that in the great experiment called the United States of America, there is a perfect blend of **ordered liberty**. [Ordered liberty, in O'Neill's sense, is] a balance between government surveillance and individual freedom. In the United States, the government-surveillance side of the [ordered-liberty] seesaw is populated by many different intelligence agencies: local police and the FBI keep an eye on people domestically, and the CIA keeps tabs on potential threats abroad. In the 1990s, [the American intelligence] system had a serious flaw: each agency gathered its own data, but sharing data between agencies was complicated. — _ABC_ **Mid-1990s (The marble-jar problem).** [ABC's Matt Bevan illustrated the 1990s intelligence data-sharing problem by imagining] the CIA, the FBI, and local police as individual jars and each piece of information they gather as a marble. [In Bevan's illustration,] each agency handled its own marbles. If the FBI wanted to check out the CIA's marbles, the FBI had to go through a complicated process of requesting specific ones and had to know what it was looking for. [In Bevan's illustration,] the opening of the CIA bottle was very narrow, and it was not easy to get an individual marble out. As John O'Neill probed the World Trade Center attack of 1993, he became interested in one particular marble [Osama bin Laden] that most counterterrorism experts at the time were not paying attention to. If you asked most terrorism experts in the mid-1990s about Osama bin Laden, most would have said that bin Laden was a terrorist financier. — _ABC_ **Mid-1990s (Smart Ship and Windows for Warships).** In the mid-1990s, the [United States] Navy had a big debate over what operating system should power its warships. The Navy's ultimate goal was to develop a **Smart Ship** system that used off-the-shelf personal computers to automate tasks that sailors had traditionally performed manually. If [Smart Ship] worked, the Navy stood to reduce crew workload, improve combat readiness, and ultimately create a safer operating environment. Unix-based systems like Linux were open source, stable, and flexible, but everyone in the Navy was already using Windows at home. Sensing his opportunity, Bill Gates pitched the Navy on something he called **Windows for Warships**, a bespoke version of **Windows NT 4.0** that would serve as the standard operating system on Navy vessels. Windows for Warships is still in use today, and not just in the United States. The British Royal Navy still uses a variant of Windows for Warships to provide tactical information and torpedo controls across all of its active submarines. Upgrading the United States Navy's fleet [to Windows for Warships] was not easy in practice. — _MSFT_ **1994–1996 (Owens and network-centric warfare).** The main person driving the [military augmented-reality] vision forward was William Owens. William Owens was a Navy admiral who served as vice chairman of the Joint Chiefs of Staff during the Clinton administration, [which meant] he had the ear of the president. William Owens was the biggest advocate in the armed forces for what he called **network-centric warfare**. William Owens laid out his entire vision in an interview with Charlie Rose in the mid-1990s. William Owens said he genuinely believed that within five years there would be a revolution in the way the United States fights wars. William Owens said [the revolution] would come primarily from the smart front end of warfighting: the ability to use high-technology surveillance of a very large battlefield, perhaps the size of North Korea or Iraq, and to see that battlefield 24 hours a day, in real time, in all weather. William Owens said that never in the history of warfighting had anyone seen the battlefield that way. [William Owens's] vision was highly compatible with the Army's new post–Cold War strategy of a smaller, better-equipped fighting force. — _MSFT_ **1996 (Land Warrior).** Eager to move past earlier failures, the Department of Defense rebranded [the Soldier's Computer] project and started calling it **Land Warrior** in 1996. [The Department of Defense] even approved a billion-dollar budget for Land Warrior. The Department of Defense appointed a colonel to act as an interface between the engineers building [Land Warrior] and the soldiers who would actually use it. Even with all of [the Land Warrior] structure in place, the project still had one big flaw it could not get over. To take advantage of the latest technology, the military had switched to using **consumer hardware** as a foundation for the Land Warrior project. [Land Warrior's consumer-grade] gear did not work in the rain and could not withstand electromagnetic interference. [Vulnerability to rain and interference] is not a problem for VR gaming on a couch, but it is a huge problem for a soldier in the middle of a battlefield. [Land Warrior thus carried two failure modes at once: the wearability burden documented in the Soldier's Computer field tests, a roughly 40-pound rig set against a consumer headset weighing about one pound, and the environmental fragility of consumer-grade electronics under rain and electromagnetic interference.] — _MSFT_ **Late 1990s (DCGS).** [In the late 1990s, the United States military services created a program] called the **Distributed Common Ground System**, known as **DCGS** [pronounced "D-sigs"]. [The Distributed Common Ground System] was designed to be a massive data-sharing system and a major weapon for the war on terror. — _PLTR_ **September 1997 (USS Yorktown).** In [September 1997], the [guided-missile cruiser] **USS Yorktown** [(CG-48)] suffered a system failure that left it dead in the water for about two and a half hours, and the new Windows operating system was squarely to blame. [The USS Yorktown failure] was the kind of thing that was embarrassing in peacetime but potentially fatal in wartime. Following the Yorktown debacle, it became easy to question the decision to go with Windows. Critics said [the Navy's decision to adopt Windows] was rushed and based more on politics than on engineering requirements. The source of the [USS Yorktown] problem was simple: one extra zero fed into just one application on the [ship's] network created a **divide-by-zero error** that crashed everything. [The USS Yorktown failure] was a big setback for the Windows for Warships program. At the time, Microsoft was still riding high as the biggest company in the world, and military contracts were just a small share of its overall business. Struggling to deliver high-quality products to the military was, however, just part of [Microsoft's] problem. — _MSFT_ **May 1998 (The bin Laden interview).** In 1998, ABC News reporter Chris Isham, a personal friend of John O'Neill, trekked for ten days across Pakistan to interview the leader of a little-known extremist group called al-Qaeda: Osama bin Laden. Chris Isham said [Osama bin Laden] sat down on a kind of bench covered in red fabric and put a blanket over his knee, and that [the interview] was like sitting at storytime with an old uncle. Chris Isham and his friend from the FBI, [John O'Neill,] shared a growing interest in this "old uncle" in the Pakistani wilderness, who seemed pretty harmless during the interview. Osama bin Laden's handlers would not allow anyone to translate the sheikh's answers [during the interview]. [ABC correspondent John] Miller did not know what Osama bin Laden was saying, and the al-Qaeda leader's monotonously measured delivery was deceptively calm. Only when the [bin Laden] interview was finished and the producers sent the tapes back to be translated did [ABC] learn what this seemingly harmless man was saying. Osama bin Laden said [in the 1998 ABC interview] that al-Qaeda did not differentiate between those dressed in military uniforms and civilians and that all were targets. Osama bin Laden predicted [in the 1998 ABC interview] a black day for America and the end of the United States. — _ABC_ **1998–2001 (O'Neill's fixation).** When John O'Neill saw [the 1998 ABC] interview, he became fixated on Osama bin Laden. John O'Neill was sure that the United States was due for another domestic attack and that Osama bin Laden would be the person to carry it out. John O'Neill tried hard to raise the alarm within the FBI, but [overseas al-Qaeda] was outside of his remit. Chris Isham says that John O'Neill started getting frustrated that he was not being taken seriously enough. [According to Chris Isham,] John O'Neill felt that the Saudis were definitely playing games and that senior officials in the United States government just did not get [the al-Qaeda threat]. A colleague said that John O'Neill's aggressive posture, aggressive nature, and willingness to go forward when it might not be politically correct made a few people uncomfortable. The problem was that John O'Neill did not have all the information: [in Bevan's metaphor,] O'Neill could not see all the marbles. John O'Neill could only look at the intelligence the FBI was gathering domestically, not the intelligence the CIA was gathering in the Middle East. John O'Neill's frustration began to compound. — _ABC_ **1999 (Microsoft's peak).** [The late 1990s and early 2000s were] like the world's worst roller coaster for Microsoft. [United States government antitrust litigators argued that] Microsoft had the power to exercise predatory and exclusionary control, [while Microsoft maintained] that courts would ultimately find all of Microsoft's actions completely legal and good for consumers. At its peak in 1999, Bill Gates presided over a 620-billion-dollar company, by far the biggest in the world. — _MSFT_ **1999 (Land Warrior cost overruns).** Trying to solve [Land Warrior's hardware] problems blew out [the program's] budgets. By 1999, spending on Land Warrior had increased by 50 percent. If a private company were burning [that much] cash with nothing to show for it, the project would be scrapped, and investors would start asking tough questions. Too many of the military's leaders had staked their reputations on Land Warrior, so failure just was not an option. — _MSFT_ --- ## V. Silos and September (2000–2001) **January 2000 (Kuala Lumpur).** Unbeknownst to John O'Neill, the CIA had more [relevant intelligence]. The CIA knew that a number of high-profile al-Qaeda operatives were attending planning meetings in [Kuala Lumpur,] Malaysia, and intended to travel to the United States. — _ABC_ **2000–2001 (Ballmer era begins).** Just two years after [its 1999 peak], Microsoft had lost 60 percent of its value, and, worse still, the company was floundering. Microsoft's arrogant dominance in the computing industry turned into uncertain anxiety. Part of [Microsoft's] problem was that the all-conquering general, [Bill Gates,] had handed over control of the company to **Steve Ballmer**. What was once a laser-focused strategy [at Microsoft] quickly became fragmented. [Microsoft's fragmentation] was not just because of Steve Ballmer, [who] was trying to make Microsoft succeed after the battlefield conditions had changed. Microsoft's original mission was to put a computer with Windows into every home, and Microsoft had basically done it, but every company needs a second act, and it took Microsoft years to find new focus. Under Steve Ballmer's leadership, Microsoft lost its core identity. Between Windows, Office, the Xbox, and various developer tools, Microsoft went from not really diversifying at all to diversifying far too much. None of Microsoft's major initiatives were at the forefront of the next major technology waves. — _MSFT_ **c. 2000–2001 (PayPal's fraud crisis and Igor).** During the PayPal years, Peter Thiel had struggled to detect fraudulent transactions on the payment network. Even though PayPal had signed up millions of users, [PayPal] was losing as much as 10 million dollars every month to credit card fraud. The [PayPal fraud] issue was so dire that Visa and Mastercard threatened to cut ties with PayPal unless the company got on top of the problem. PayPal's chief technology officer, Max Levchin, stepped up [to solve the fraud problem]. Max Levchin built a program to scan through transactions and flag suspicious behavior for manual review. Max Levchin named the [fraud-detection] program **Igor** in honor of the Russian hacker who had been the biggest thorn in his side. Igor was so good that it not only solved PayPal's fraud problem but actually exposed the identities of two Russian cybercriminals who were promptly arrested by the FBI. The experience of searching for a few small criminal operations in the massive dataset of PayPal transactions would prove very useful for Palantir. — _PLTR_ **July–August 2001 (Phoenix and Minnesota).** In the FBI's [metaphorical] jar, which John O'Neill had access to, there were a couple of concerning marbles. One [concerning FBI] marble was that a man with known connections to extremist groups, [Zacarias Moussaoui,] had recently been arrested by the FBI while attending a flight training school in Minnesota. Another [concerning FBI] marble suggested that Osama bin Laden might be coordinating an effort to send people to the United States to obtain flight training, after a number of suspicious people were seen attending flight schools in Arizona. [The Minnesota and Arizona leads] were two very interesting marbles, but not enough to give [the FBI] the full picture. John O'Neill never saw all the marbles. — _ABC_ **Mid-2001.** In mid-2001, John O'Neill left the FBI, partly out of frustration, and took a high-paying job in the private sector as head of security for the World Trade Center. — _ABC_ **September 11, 2001.** John O'Neill died in the World Trade Center's South Tower on September 11, 2001, just three weeks into [his job as head of World Trade Center security]. Chris Isham said that the coincidence was just unbelievable: the man who identified Osama bin Laden as the villain he was had been killed by Osama bin Laden. — _ABC_ **After September 11, 2001.** People who were around at the time will remember the understandable concern that [an attack like September 11] might happen again. Many changes were made to try to make people feel safe from terror attacks, and many of those changes were quite annoying. Peter Thiel said that the way [the United States] was going, with ridiculous airport security checks and super-intrusive surveillance all the time, was not really making [Americans] safer. Peter Thiel, [a technology billionaire,] has complicated opinions about how the world should be run. At [the post-9/11] stage, Peter Thiel was not a fan of the government trying to intrude into people's lives in the name of stopping terrorism. Peter Thiel was concerned that if there were another attack along the lines of 9/11, the government would try to become even more intrusive. Peter Thiel said that if the World Trade Center [attack] eroded civil liberties as much as it did in 2001, he did not even want to think what would happen if there were another terrorist attack, and that [America] therefore had to prevent [another attack] to stop more erosion. Peter Thiel was more in favor of the liberty side of John O'Neill's ordered-liberty seesaw. Peter Thiel asked whether one could do something from a libertarian or civil-liberties point of view that would still be tough on terrorism. Peter Thiel said that Silicon Valley ought to be involved in fighting terrorism and protecting [Americans'] civil liberties. — _ABC_; _PLTR_ **October 2001.** Apple introduced the **iPod** in October 2001. It took Microsoft five years to release [a competing music player,] the **Zune**. — _MSFT_ --- ## VI. The Founding of Palantir and the Long War (2002–2011) **Late 1980s–2002 (Karp's formation).** Understanding Palantir really requires understanding its CEO, Alex Karp. Today, Alex Karp runs a multibillion-dollar public company and is referred to as "Dr. Karp" by everyone who reports to him, but Karp took a long and winding road to success. Alex Karp was born in 1967 and grew up in Philadelphia. Both of Alex Karp's parents were forward-thinking progressives who took him to labor-rights protests on weekends. Despite struggling with dyslexia from an early age, Alex Karp graduated from Haverford College in 1989. [Alex Karp] went on to study law at Stanford. While at Stanford, Alex Karp roomed with Peter Thiel, the same Peter Thiel who would go on to build PayPal and Founders Fund. During their time at Stanford, [Alex Karp and Peter Thiel] often sparred intellectually. Peter Thiel had established himself as an outspoken libertarian as an undergraduate, even founding an on-campus newspaper called the **Stanford Review**. Alex Karp, on the other hand, leaned more socialist. Instead of driving [Karp and Thiel] apart, [their political disagreement] actually drew them together. [Alex Karp and Peter Thiel] spent many nights arguing about politics and philosophy and became close friends. After graduation, [Karp and Thiel] went separate ways: Peter Thiel went to the prestigious law firm **Sullivan & Cromwell**, and Alex Karp went off to continue his studies in Europe. Alex Karp formally became Dr. Karp at [Goethe University] Frankfurt in Germany. [At Frankfurt,] Alex Karp studied under **Jürgen Habermas**, one of the world's most famous philosophers, earning a PhD in neoclassical social theory in the process. [Earning the Frankfurt PhD] wound up being a turning point for the young Dr. Karp, but not in the way he thought: even though he had earned his stripes under a world-renowned academic, Karp realized that he did not want [an academic] life, so he pivoted. — _PLTR_ **c. 2000–2003 (Caedmon Group).** [Around] the turn of the millennium, the dot-com boom was in full swing, and Dr. Karp had developed a network of high-net-worth individuals in Europe who were looking for new investment strategies. Alex Karp had a knack for investing, but at this stage his aspirations were still fairly modest: he mostly wanted to build a small nest egg and then settle into a casual life in Berlin. To start building toward [his modest financial] goal, Alex Karp created an investment company in London called the **Caedmon Group**. For a few years, Alex Karp had been helping introduce high-net-worth individuals to [Peter Thiel's hedge fund,] **Clarium Capital**. Many of [the individuals Karp introduced to Clarium] were connected to global intelligence and defense institutions. Alex Karp's dream of a modest life was soon interrupted. — _PLTR_ **2002 (PayPal exit).** While Dr. Karp was off in Europe, Peter Thiel had been building an empire back in the United States. Peter Thiel had started an internet payments company, [Confinity,] merged it with a similar project Elon Musk was building, [X.com,] and then taken the combined entity public under the name **PayPal**. In 2002, PayPal was acquired by eBay for 1.5 billion dollars. [The eBay acquisition of PayPal] made Peter Thiel a multimillionaire, and his ambitions were just getting started. — _PLTR_ **2002–2003 (Total Information Awareness).** The United States national security establishment made many mistakes [in the hunt for bin Laden]. The first big [mistake] came just a few months after 9/11, when the United States government established a program called **Total Information Awareness** (TIA). The idea [of Total Information Awareness] was to correlate detailed information about people in order to anticipate and prevent terrorist attacks before they took place. The man leading [Total Information Awareness] was Admiral John Poindexter, a highly controversial figure. John Poindexter had retired from the Navy in 1987 and was convicted of multiple felonies stemming from the **Iran-Contra** affair. Although [John Poindexter's] conviction was later overturned on appeal, Poindexter returned to government during the George W. Bush administration to head the **Information Awareness Office**, which was run out of DARPA, the Pentagon's [Defense Advanced Research Projects Agency]. On paper, the goal [of the Information Awareness Office] was simple: to bring together as much information as possible about terrorist networks in order to identify the most pressing threats. John Poindexter referred to [Total Information Awareness] as a **Manhattan Project for counterterrorism**. [Total Information Awareness] was short-lived. [Total Information Awareness] was heavily criticized for sounding like a program that allowed the United States government to spy on all of its citizens. John Poindexter resigned in August 2003, and the Information Awareness Office was disbanded. The United States government nonetheless continued much of [Total Information Awareness's] core architecture [within classified programs]. [Losing his post] did not matter to John Poindexter, who was on the outside but still a useful asset for anyone who wanted to sell security products to the United States government; in other words, [Poindexter] was a useful asset to Peter Thiel and Alex Karp. — _PLTR_ **2003 (Palantir founded).** Peter Thiel had seen the damage wreaked by the attacks on 9/11 and realized that America was about to change. Peter Thiel became increasingly fixated on using technology to stop terrorist attacks before they happened. [Using technology to stop terrorist attacks] would require working with the notoriously closed United States intelligence community. Peter Thiel had been a libertarian for years, and he wanted to stop terrorism without sacrificing civil liberties to do it. Peter Thiel was facing a modern version of [the] centuries-old philosophical problem [of liberty and safety] most famously stated by Benjamin Franklin in 1755. On top of the ethical conundrum, Peter Thiel also knew that achieving his vision would be a formidable technical challenge. Peter Thiel decided to assemble a team of people he had worked with before, both at Stanford and at PayPal. Nathan Gettings, Joe Lonsdale, and Stephen Cohen had all met Peter Thiel through Stanford, PayPal, or Thiel's hedge fund, Clarium Capital. Peter Thiel named the startup **Palantir** after the seeing stones in J. R. R. Tolkien's _The Lord of the Rings_. [ABC noted that the palantír was] Saruman's all-seeing glass ball, a dangerous tool, [and that] naming a nascent technology company after a tool used by the villain trying to take over Middle-earth is an interesting decision. Peter Thiel and a friend named Alex Karp started focusing on the marbles. The [United States] government already had much of the information it needed to prevent terror attacks, but [that information] was spread across many different jars, and the government was not using the information efficiently. As far as Alex Karp and Peter Thiel were concerned, Palantir was designed to break open all the jars of marbles, spill them onto a flat surface, and make sense of whatever came out. [Palantir's founders said Palantir] would allow humans to find needles in haystacks and make data intelligible to ordinary people. By [making data intelligible], [Palantir's founders said,] Palantir would allow [analysts] to find bad people trying to destroy [Western] society and could also be used to protect civil liberties by making datasets transparent. — _ABC_; _PLTR_ **2003 (Choosing a CEO).** The people [Peter Thiel] brought together [to found Palantir] were all great, but none of them were quite right for the job of CEO, and [the lack of a suitable CEO] was a big problem. Peter Thiel was able to provide initial funding for Palantir, and the team of engineers was able to build a functional prototype, but [Palantir] was struggling to get potential investors and customers to take it seriously. Palantir was not some viral consumer technology company; Palantir wanted to sell to the biggest customer in the world, the United States government. [The need to sell to the United States government] is where Dr. Karp came up as an interesting option. Peter Thiel knew that building a software business that could work effectively with the intelligence community would require a CEO who could work with big-government people. Peter Thiel needed someone smart who could bridge the gap between Palantir and the intelligence community. Peter Thiel thought about it and then decided to call his old friend, Dr. Alex Karp. Peter Thiel later said that at the time he was not sure Alex Karp was the perfect person for the job, but that Karp possessed the right qualities. Even though Alex Karp had never been a CEO, he clearly understood the mission and could build a solid team. The team Alex Karp put together had all the skills to build a good product, but actually selling [the product] to the United States government was a different challenge entirely. If [Palantir's team] were going to solve [the government-sales problem], they would need to be patient, because in counterterrorism, progress was measured in years, not days. — _PLTR_ **2001–2011 (The decade-long hunt).** Despite [Osama bin Laden] being their global number-one target, it took the United States military almost a decade after 9/11 to catch bin Laden. The process of finding [Osama bin Laden] was long, winding, and full of setbacks and near misses. Effective counterterrorism requires synthesizing many different sources of intelligence: intercepting communications between militants, talking to people on the ground, and making clever use of informants. Just gathering [counterterrorism] data was not nearly enough, [because analysts] needed different agencies to collaborate with each other in order to detect the underlying patterns. — _PLTR_ **2004 (The Shire and the prototype).** Alex Karp said that terrorism is asymmetric, that asymmetry presupposes software because [counterterrorism] is finding needles in haystacks, and that [Palantir's approach] was not exactly data mining but a critique of some of the data-mining approaches that had been used at PayPal, [an approach Palantir's founders] thought would be very effective [in counterterrorism]. Peter Thiel said that what Palantir does is a way for patterns and data to be visualized through a combination of computers and human analysts, in a way that does not simply scour the planet but looks some more when something suspicious appears, which in that sense [makes Palantir] far less intrusive. In 2004, Peter Thiel and Alex Karp had assembled a growing team of hardcore engineers focused on shipping at breakneck speed. The Palantir team moved into an office in Palo Alto which, in keeping with the _Lord of the Rings_ theme, they nicknamed **the Shire**. After months of arduous days, long programming sessions, and too many energy drinks to count, the Palantir team cobbled together a prototype. The Palantir prototype was designed to link disparate databases and then let anyone, even a non-programmer, search and visualize the information within them. Most importantly, everything that happened within Palantir's software would be **logged** so that it could be **audited** by authorities. Classified data [within Palantir] could not be accessed by just anyone; [a user] needed proper clearance and proper justification for an investigation. [Logging, auditing, and clearance controls] were critical if Palantir was going to get the United States government on its side. [Palantir's architecture thus joined two properties in one design: correlation across previously separated datasets, and access control enforced through clearance, investigative justification, and an audit log of every action, so that the system could fuse what agencies held while governing who could see the fused result.] The [Palantir] team had already put a huge amount of work into building the prototype, but they knew the hardest part of their job was just about to begin. An early Palantir engineer said that [Palantir] had a crucial [government] meeting only two months away and did not have much product at the time. [The early Palantir engineer said that] the initial prototype was admittedly light on the back-end elements and more focused on the lights, the fireworks, and the front end. — _PLTR_ **c. 2004–2005 (Early rejection).** The [Palantir] prototype demonstrated the value Palantir could provide to the government. Because [the Palantir prototype] had been built by engineers with experience delivering products people actually used regularly, like PayPal, it was intuitive. Palantir clearly believed its product could make a difference in the war on terror. Selling to the government is more than just building a functional product, and often even having the best product is not enough. [Selling to the government requires] convincing dozens of different decision makers and wading through miles of red tape. In hindsight, the Palantir team seems to have believed that its product would sell itself, but that is not what happened. [A Palantir founder] said that [Palantir] did a very bad job making its case, that [selling to government] was very difficult, and that [Palantir] did not understand what [government officials] were saying and [government officials] did not understand what [Palantir] was saying. [Palantir's] team of engineers, who were used to moving extremely fast, came up against the immovable object of the United States government and were left disappointed. [The early rejection exposed a gap between technical merit and procurement literacy: Palantir's engineers could build a working, intuitive product, but they did not yet speak the language of requirements, clearances, and contracting through which Washington buys software.] [The government setback] was compounded by the fact that, without any customers, Silicon Valley investors were not lining up to pour money into Palantir. Peter Thiel had invested millions of dollars of his own money in Palantir, but other venture capitalists were not interested in [a company] selling to the government. One venture capitalist from **Sequoia** spent the entire [Palantir] pitch meeting doodling, while another lectured the Palantir founders for over an hour and a half about how their company was going to fail. [Silicon Valley's indifference to Palantir] is deeply ironic, because Silicon Valley was built on strong partnerships with the American military. Early semiconductor companies were spun out of government-funded research laboratories, and many early technology companies received funding from the Department of Defense. By the mid-2000s, it was as though Silicon Valley and Washington spoke different languages. — _PLTR_ **Mid-2000s (Steve Blank on the defense-industrial base).** Steve Blank, a Stanford professor who has studied how technology companies have worked with the military over the past century, became particularly frustrated by [the estrangement between Silicon Valley and the Pentagon]. Steve Blank said that getting companies like Anduril, Palantir, or SpaceX to become major defense contractors required a billionaire [backer], which he called insane. Steve Blank told [John Coogan at Stanford] that it is not that innovators, whether startups, scale-ups, or venture firms, do not want to work with the Department of Defense. [Steve Blank said] the Department of Defense has no conception of what innovation at scale could do for it and is fundamentally not organized to [use such innovation]. [Steve Blank said the Department of Defense's] organization chart is wrong and it has the wrong people in place. Over the course of the 1990s and early 2000s, the United States defense industry went through a massive wave of consolidation until it was left with just a handful of **prime contractors**. [The defense primes] won the overwhelming majority of government defense contracts. If the United States government wanted a new software system, [the government] would not ask a small Silicon Valley startup but would simply call Lockheed Martin, Booz Allen, or IBM. If Palantir wanted to break through to the United States government, something would need to change. — _PLTR_ **c. 2005 (In-Q-Tel).** Dr. Karp was not going to sit around and hope to get lucky. One of the venture capital funds that had turned Palantir down for investment was willing to make an introduction to **In-Q-Tel**, a nonprofit venture capital firm run by the CIA to fund innovation that could help the intelligence community. The Palantir team met with the head of In-Q-Tel, **Gilman Louie**, and pitched him on their idea for an integrated data platform. Gilman Louie said that Palantir got started when Peter Thiel and Alex Karp met with him and said they had algorithms for detecting fraudulent transactions that might be valuable to the intelligence community. Gilman Louie said that [In-Q-Tel] immediately put [the Palantir] team inside the building [at the CIA] to work on what [the agency] called a counterterrorism problem. [In 2005,] George W. Bush was president, the United States military was still in Iraq and Afghanistan, and Osama bin Laden was still at large. Palantir's luck was finally starting to improve. After struggling to get his foot in the door for months, Dr. Karp was able to meet with many government officials and start learning what it would take for them to buy Palantir software. [Alex Karp's government] meetings did not always go well, because the cultural divide [between Silicon Valley and Washington] seemed as big as ever. The agency officials Dr. Karp met with were often skeptical of his relaxed style and had many questions for him. [Agency officials asked] whether [Palantir's] team was cleared to work with classified documents, whether they had ever worked with intelligence agencies before, whether they had senior advisers with government experience, and whether their sales force was built to work with classified information. The answer to all of [the agency officials'] questions was no. [Those negative answers] did not discourage Dr. Karp and the Palantir team, for whom "no" did not mean "never" but only "not yet." Everyone [Palantir] pitched could tell that, even though the team lacked government credentials and experience, it could deliver. On one visit to the Palantir offices, a senior intelligence official saw sleeping bags under the engineers' desks. Clearly, the pace of play at Palantir was completely unlike anywhere else. — _PLTR_ **c. 2005–2006 (Langley iteration and forward deployment).** Gilman Louie, the head of In-Q-Tel, acted as Palantir's anchor at the CIA. In-Q-Tel invested 2 million dollars [in Palantir] and helped connect Palantir to CIA analysts in Langley, Virginia. [ABC described the CIA, through In-Q-Tel, as] Palantir's first [institutional] investor. Every two weeks, Palantir engineers would fly from Palo Alto to Langley with an updated version of the software they were building. The CIA analysts would test [each Palantir build] and offer feedback, and then the [Palantir] engineers would fly back to California to implement the suggested changes. [In-Q-Tel functioned as the translation layer between the two cultures: it supplied the capital, the introductions, physical access to CIA analysts, and a two-week feedback cadence that converted analysts' operational needs into engineering requirements Palantir could build against.] Today, [the practice of embedding engineers with customers] is standard practice at Palantir. Palantir even has a name for employees who embed themselves within the customer they are building software for: **forward-deployed engineers**. [Forward deployment] has been a critical part of Palantir's success. The company embedded technical experts inside classified United States intelligence operations and facilities, trying to access as many different information sources as possible and figure out how those sources relate to each other. An early Palantir engineer said that one point demarcating success was when his government sponsor introduced him at a meeting as "Mr. Two Weeks," [because] he could build anything anyone wanted in two weeks. [The early Palantir engineer said that being called "Mr. Two Weeks"] definitely increased his Red Bull consumption. Word was spreading within the United States government about the intensity the Palantir team brought to problem solving. [The Palantir team] was building and iterating extremely quickly and, most importantly, knew where it should be focused. One In-Q-Tel executive said the most impressive thing about the [Palantir] team was how focused it was on the problem of how humans would talk with data. Things were beginning to go in the right direction [for Palantir], but solving the problem of human-computer interaction in a way that would scale past a single organization would take years. — _ABC_; _PLTR_ **2006 (Shyam Sankar and man-computer symbiosis).** [Alex Karp said that] the basic idea [behind Palantir] was how to extend the human mind into an enterprise to get the full power of the human mind, and that the problem with that idea is that there is not one dataset but five, ten, or fifty. Palantir needed to build a system that allowed humans and computers to work together effectively. Peter Thiel said there is a tremendous amount that can be done by combining computers with humans, [where the prevailing approaches offered] either all-human solutions or all-computer solutions. Peter Thiel said there is a lot to be done with these **intermediate hybrid solutions**, and that if the division of labor [between human and machine] is right, [hybrid systems] can do a lot. Dr. Karp had always been focused on building a world-class team, and not just in engineering. In 2006, [Alex Karp] hired **Shyam Sankar**, [who is described by the source as having served as Palantir's chief operating officer]. Shyam Sankar is still with Palantir today, although he switched his title to chief technology officer. Shyam Sankar was convinced that real-world problems could only be solved by combining the abilities of humans and computers. Shyam Sankar had extensively studied the work of J. C. R. Licklider. Shyam Sankar believed that the key to making good decisions in complex situations was not depending on predetermined programs but having man and machine cooperate to test hypotheses. [Sankar's symbiosis thesis] had real-world applications to Palantir's work. [Alex Karp said that] computers do not detect novel patterns or new behaviors, but humans do, by using technology and asking machines to do things for them. [A Palantir-affiliated speaker said that] Osama bin Laden was not caught by artificial intelligence but by dedicated, resourceful, brilliant people in partnership with various technologies. After months of trying, Dr. Karp had won over the CIA, and Palantir software was finally starting to have an impact in the fight against terrorism. — _PLTR_ **2006 (Microsoft and the cloud).** The consumer-internet and mobile-device trends were both underpinned by a major shift to the **cloud**. When a phone or laptop performs operations, [those operations are] usually not performed locally but on a server in the cloud. This time, Microsoft was determined to ride the new [cloud] wave. Microsoft's history with cloud computing dates back to 2006 and an experimental project to put Windows into the cloud. Businesses did not want Windows in the cloud; [businesses] wanted flexibility, and flexibility is exactly what **Amazon Web Services** (AWS) delivered. — _MSFT_ **2006–2010 (Microsoft's lost decade).** The personal-computer growth wave that turned Microsoft into the world's biggest company was over. The 2000s were an era dominated by the consumer internet and mobile devices, and Microsoft did not have a winner in either category. [Microsoft's] Zune [music player] flopped and was discontinued in 2011. The iPhone launched in June 2007. It took Microsoft three years to release **Windows Phone**, which also completely flopped. Google's search engine launched in 1997. It took Microsoft twelve years to put out a competitive [search] product, and when [Microsoft's search engine] did arrive, it was **Bing**. Today, Google has 92 percent market share in search, and Bing has just 2 percent. All of [Microsoft's 2000s] launches were reactions, and none of them played to Microsoft's strengths. The company that had been such a fearsome leader became a meek follower. The incredible and ongoing success of Windows had left Microsoft with one key strength: **distribution to enterprise customers**. Microsoft was still trusted by the world's largest corporations, many of which ran on Windows and Word. [Microsoft] needed to leverage those [enterprise] relationships the next time there was a major technology trend. — _MSFT_ **2005–2007 (Land Warrior in Iraq).** By the time the United States invaded Iraq, Land Warrior was ready to be deployed. In 2005, [according to the source,] 229 United States infantry soldiers were fitted with Land Warrior headsets. It had been almost twenty years since the [Soldier's Computer and Land Warrior] project began, but it was finally getting traction and had found a use case. Soldiers deployed in Iraq used their Land Warrior gear to digitally mark safe regrouping points. Soldiers still did not really love [Land Warrior], which still needed a lot of work. — _MSFT_ **2007–2010 (Palantir enters the war zones).** Terrorism was a threat in American cities, but [terrorism] also posed a threat to members of the United States military stationed abroad. The [United States] Army needed to identify and track potential attackers just as urgently as the CIA [did]. Palantir saw an opportunity to expand its product offering [to the Army]. If [Palantir] thought selling software to the CIA was tough, it was about to face an even bigger challenge [in the Department of Defense]. Part of the reason selling to the Department of Defense was so tough was the defense primes. Rather than take on multibillion-dollar leviathans, Dr. Karp and his team adopted an unorthodox strategy. Instead of focusing on lobbying the Pentagon like other major defense contractors, [Palantir] introduced its product to soldiers directly. Palantir would contact soldiers and offer free training on its software. [Palantir] would also give out funding to support the work of soldiers who would be working with data when they deployed. [Palantir's direct-to-soldier] approach helped create internal demand for Palantir software inside the military, because soldiers already knew how well the software worked and could start advocating for it internally. Getting troops comfortable with complicated software would not be easy. Matt Grimm, the chief operating officer of **Anduril**, who worked at Palantir around the time [Palantir] started deploying software to troops in the Middle East, remembered how difficult [the deployment] was. Matt Grimm said that Palantir had just deployed to its first customers and thought its product was so intuitive and so easy that [soldiers] would have an immediate "aha" moment, which is not what happened. Matt Grimm said that [the early Palantir deployment] became a litany of help-desk [calls]. [Deploying to troops] was yet another culture shock for Palantir. After spending ages trying to learn the language of the government, Palantir now had to learn how to deal with the military. The first thing [Palantir] had to learn was that getting soldiers comfortable with the software was just as important as building the software. Over time, [Palantir's direct-to-soldier] plan worked. Soon, high-ranking military officers were meeting with the Palantir team to discuss using [Palantir] software in Afghanistan. — _PLTR_ **2007–2012 (Palantir versus DCGS-A).** [Palantir's Afghanistan expansion] could have been completely smooth sailing except for one thing: the United States military had already signed up for a 10-billion-dollar project to build a similar system internally, [the Distributed Common Ground System]. All of the biggest defense contractors had signed on to build the Distributed Common Ground System. Basically, the Distributed Common Ground System was too big to fail, but it was still failing. [The Distributed Common Ground System] was clunky and prone to crashing. Military officials had spent years developing the requirements for [the Distributed Common Ground System] and wanted to see their internal programs see the light of day. Palantir was not built by the military but by a team of Silicon Valley engineers with no prior experience working with the government, yet remarkably, [Palantir] was still winning the battle for hearts and minds. [A military officer said that] a unit came to [Palantir's advocates] and said that [Palantir] actually saved lives because [the unit] was able to plot where all the IEDs were. [The officer said that the unit] saw a huge increase in IEDs found, which was a big deal because IEDs were the number-one killing and wounding mechanism used by the enemy in Afghanistan. Palantir was winning over both ground troops and the people who led them. General Jim Mattis supported the [Palantir] program, as did Lieutenant General H. R. McMaster. Not every senior officer liked Palantir. [The Palantir situation] became even more complicated when Congress started asking questions about how much the [Distributed Common Ground System] program was costing American taxpayers. [An Army leader defending the program, apparently Army Chief of Staff General Raymond Odierno,] said that the Army had more intelligence capability than it had ever had and that a company commander with **DCGS-A** [the Distributed Common Ground System–Army] had twenty times the capability [the speaker] had as a division commander in 2003. — _PLTR_ **c. 2008–2010 (Mattis and PowerPoint).** United States Marine Corps General James Mattis was fond of saying that "PowerPoint makes us dumb." [James Mattis] once added that he had a friend who ran Microsoft who did not like it when Mattis said [that PowerPoint makes us dumb]. At the time, the United States military's overuse of PowerPoint was an easy punchline for commanders giving public speeches. [One commander joked that] it was the inalienable right of every four-star Army general to use PowerPoint slides when communicating. [Another commander spoke of] relying only on the commander's intent rather than detailed orders in multicolored PowerPoint format. — _ABC_ **2010 (Xbox Kinect, Azure, and the McChrystal slide).** In the early 2010s, virtual- and augmented-reality technology looked like the next big thing, so Microsoft got to work. Microsoft introduced the **Xbox Kinect** in 2010. By the time Microsoft released the **Azure** cloud computing platform in 2010, [Microsoft was] a long way behind Amazon, but [Microsoft] had one key advantage: distributing the products it made to big customers. In 2010, the United States military produced [a PowerPoint slide] to illustrate the difficulties it was facing during the war in Afghanistan. When the public got hold of [the Afghanistan strategy slide], it became a case study in how not to do PowerPoint, [and ABC called it] the worst PowerPoint slide in human history. The [Afghanistan strategy slide] was prepared by the staff of General Stanley McChrystal. A _New York Times_ article [about the slide] was headlined "We Have Met the Enemy and He Is PowerPoint." Even the head of United States forces at the time, General Stanley McChrystal, knew immediately how bad the slide was. When [McChrystal's] staff brought him the [Afghanistan strategy] slide, McChrystal said that when [the military] understood that slide, it would have won the war. — _ABC_; _MSFT_ **June 2010 (Nett Warrior).** By June 2010, under the Obama administration, Land Warrior was rebranded again as **Nett Warrior**. [Under Nett Warrior,] the bulky pack and headgear were gone, and in their place was a ruggedized Android phone secured to the soldier's chest. Mobile technology had become good enough to fulfill most of William Owens's vision, but the Army still thought it needed more to show for all its work and money spent. — _MSFT_ **May 2, 2011 (Operation Neptune Spear).** The time was "zero dark thirty," and the location was Abbottabad, less than 100 miles from Pakistan's capital [Islamabad]. Nestled [in Abbottabad] was a fortress of secrets and a haven for the world's most wanted man, [Osama bin Laden]. [The Abbottabad raid] began thousands of miles away, deep in the halls of power of the most technologically advanced nation on Earth, America. [The bin Laden hunt] is not just a story of bravery under cover of night but one of years of relentless pursuit, intelligence gathering, and geopolitical maneuvering. **Operation Neptune Spear** killed Osama bin Laden and highlighted the incredible prowess of the Navy SEALs. [Operation Neptune Spear] also demonstrated the power of American technology to decipher coded messages and connect millions of invisible threads of intelligence. Palantir needed something big to happen involving its technology, and in the early hours of May 2, 2011, in the Abbottabad compound, [Palantir] got just what it needed. It is hard to say for sure how big a role Palantir software played in finding Osama bin Laden, because the United States [military] does not share those kinds of details, especially concerning the killing of the world's most notorious terrorist. It has been widely reported, though never officially confirmed, that in 2011 Palantir played a key role in processing information which led to the killing of Osama bin Laden in his compound in Pakistan. [One commentator said that] Palantir effectively vectored Osama bin Laden's location. The whispers [about Palantir's role] started right away, but no one at Palantir ever said anything about it. [The bin Laden question,] much like Palantir itself, is shrouded in secrecy. What is clear is that, in the years following bin Laden's killing, Palantir became the [United States] government's go-to company when it needed to make sense of big datasets for military and law-enforcement purposes. — _ABC_; _PLTR_ --- ## VII. The Platform Era (2012–2022) **c. 2012 (Big data).** Part of [Palantir's post-2011 rise] reflects the rise of **big data** in general. Around 2012, there was a major turning point, as big corporations started storing increasingly large amounts of data. The realization that companies and governments were sitting on valuable troves of data opened up big opportunities for Palantir. Dr. Karp and his team started expanding Palantir's offerings to solve a variety of problems for non-governmental organizations. [Palantir] partnered with JPMorgan to build financial software. [Palantir] worked with airlines to help optimize flight routes. [Palantir] landed a big deal with **Airbus** to improve nearly every aspect of how Airbus built airplanes. — _PLTR_ **2012 (The Army study).** Despite intense scrutiny from some military officials, Palantir was still overwhelmingly popular with soldiers. According to a 2012 study commissioned by the Army, 96 percent of military personnel deemed Palantir's software effective. [The favorable 2012 Army study] should have been amazing news for Palantir, but it was not. [Palantir] had all this support from soldiers on the ground and still was not getting anywhere [with Army procurement]. The military basically ignored [the 2012] report and kept doing business as usual. [The 2012 Army study formalized at institutional scale what field units had already reported directly, such as the unit that credited Palantir's IED mapping with saving lives; the study converted individual soldier testimony into an official 96 percent effectiveness figure, and procurement still did not move.] For two years, everyone [in the Palantir–Army dispute] was treading water. — _PLTR_ **2013.** Amazon had just won a 600-million-dollar contract to provide cloud services for the United States intelligence community. — _MSFT_ **Through 2015 (The quiet decade).** For the first decade of Palantir's existence, basically nobody knew who [Palantir was] or what it did. [Palantir] did not even start posting on social media until 2015. For most of its existence, Palantir flew under the radar as an extremely obscure United States defense contractor that few people outside the industry had ever heard of. [A Palantir presenter said that] a person can see a pattern in a hundred things but finds it very hard to see a pattern in a million. [Palantir's product is] really about creating a unified view over many disparate data sources which do not otherwise make sense together. The reason [Palantir] had so little public presence is that [Palantir] really had only one customer: the United States government. [Palantir's government work included] fighting ISIS, stopping human trafficking, supporting money-laundering investigations, working with Special Forces, law enforcement, and the Department of Defense. According to [Palantir's] marketing, Palantir was all about organizing information in a way that humans would find more intuitive. [A Palantir presenter said that] humans are incredible at insight and that [Palantir's task] is just surfacing the right information for humans to see. [In ABC's characterization,] Palantir was effectively creating an enormous, self-updating PowerPoint slide which humans could use to find patterns, circle them, connect them, and figure out what was really going on. Using Palantir's software was like looking down on the world from above, with information displayed on a large, flat, interactive map. [A commentator described Palantir's interface as] basically like Google Maps, where [a user] can look around the United States, zoom in on targets, and click on an individual person to bring up a name and photograph. [The analyst workflow ran from the map downward: pan across a geography, zoom on a target area, select an individual entity, and pull that person's identifying record, including name and photograph, out of the fused data.] — _ABC_ **c. 2004–2022 (Building the Ontology).** Programming [Palantir's] maps, which Palantir calls **the Ontology**, was slow, painstaking work. [Building the Ontology] involved writing incredibly complicated software and enormous amounts of code. [Building the Ontology also involved] talking to many different people so that [Palantir] did not overemphasize the importance of any one data source or ignore another. [The Ontology] did not entirely solve the marble-jar problem, because each industry needed information laid out in different ways, and [Palantir's] systems did not talk to each other perfectly. For instance, **Gotham**, Palantir's military and law-enforcement software, was different from [Palantir's] logistics software, which was called **Foundry**. — _ABC_ **2014–2016 (Azure's ascent).** Microsoft studied Amazon's [cloud] strategy very closely and improved Azure. Once [Microsoft] had a solid [Azure] product in place, it started including Azure credits in its enterprise deals. [Bundling Azure credits] drove rapid adoption from large customers who needed thousands of Office licenses. A company already buying many Windows licenses for its employees and suddenly holding free cloud-computing credits would naturally choose Microsoft Azure over AWS, which would cost it far more. [Bundled credits turned Microsoft's existing Office and Windows licensing relationships into a procurement channel for cloud: a customer that had already budgeted for Microsoft licenses faced almost no marginal cost and almost no switching friction in adopting Azure, while choosing AWS meant a separate purchase, a separate vendor relationship, and a separate budget line.] It was not long before Azure hit a 10-billion-dollar run rate and was growing faster than AWS. Azure was the biggest reason Microsoft's stock price started climbing after being flat for years. Microsoft was finally riding a major new technology wave. Large organizations everywhere were closing their data centers and moving their hosting to the cloud. — _MSFT_ **2016 (HoloLens).** Microsoft introduced the first **HoloLens** in 2016. Microsoft ultimately believed that augmented reality was too cumbersome for everyday consumers. Unlike other headsets such as the Oculus VR, which were aimed mostly at gamers, Microsoft positioned the HoloLens as a professional device for large corporations. Selling [the HoloLens] to large enterprises would leverage Microsoft's key strengths, and enterprises do not come much bigger than the United States military. — _MSFT_ **2016 (Airbus A350).** In 2016, Airbus was ramping up production of its new **A350** jet but quickly ran into a problem. Going from manufacturing one single [A350] plane to building dozens would increase the complexity of all [Airbus's] processes exponentially. Missing parts, production mistakes, and communication glitches would slow down the [A350] assembly process and cause millions of dollars of cost overruns. Airlines do not like waiting for delivery of the planes they have ordered and often demand compensation for any delays. To solve [the A350 production problems], five forward-deployed Palantir engineers flew to the Airbus factory in **Toulouse**, France, and set up shop. While on site, [the Palantir engineers] merged 25 different data silos related to the production of the A350. [Merging the A350 data silos] produced results immediately, reducing delays to fix production mistakes and realizing hundreds of millions of dollars in savings. Palantir was eventually rolled out to thousands of Airbus employees. Today, information from nearly every Airbus plane flows through Palantir for analysis to improve reliability and safety. — _PLTR_ **2016 (Palantir sues the Army).** The [United States] Army finally decided that enough was enough. [The Army] realized that the Distributed Common Ground System would never work the way it wanted. The Department of Defense began soliciting bids to develop a replacement for the Distributed Common Ground System. [The DCGS replacement solicitation] seemed to be the breakthrough Dr. Karp and the Palantir team had been waiting for, except that it was not quite like that. The Army refused to allow Palantir to participate in the new bidding process because Palantir's software was an **off-the-shelf product**. [The Army] only wanted to hear from defense contractors who would promise to build an entirely new system from scratch. To some degree, [the Army's preference for a custom build] made sense, because everyone wants something tailor-made to their specific needs, and the Army wanted a system that would be perfect. [Excluding Palantir] would have been a disaster if it had gone forward, because 96 percent of soldiers already thought Palantir software was good enough for the Army and the software was getting better every day. When the Palantir team made improvements for one government agency, it was easy to make those improvements available to the Army. The Army did not have much experience building and running successful software programs, so it rejected Palantir. Yet again, a government agency and a Silicon Valley startup did not speak the same language. Dr. Karp and the [Palantir] team were backed into a corner. [Palantir] knew it could make a difference on the battlefield, but government bureaucracy was getting in the way. As the old saying goes, desperate times call for desperate measures, so Palantir sued the United States government. The federal court [the United States Court of Federal Claims] took only three months to rule in Palantir's favor. The judge said that the Army had acted in an **"arbitrary and capricious"** manner and ordered that Palantir be included in the bidding process. [The court ruling] was a big win for Palantir, which would finally be a first-class citizen in the world of military contractors, rubbing shoulders with the primes. [The 2016 lawsuit closed a causal chain that ran from 2007 onward: Palantir's direct-to-soldier strategy created demand from below; the Army's build-from-scratch procurement rules and the primes' incumbency on the Distributed Common Ground System blocked that demand from above; the 2012 Army study documented the demand without changing the rules; and litigation became the instrument that forced the procurement system to admit a commercial off-the-shelf competitor.] — _PLTR_ **Post-2016 (The software-budget ceiling).** Just as [Palantir] solved its Department of Defense problem, a new problem emerged. At the end of the day, the [United States] government just did not spend very much money on software. The United States spends trillions of dollars every year on the military broadly, but only a tiny amount of that goes toward software. The Distributed Common Ground System was a 10-billion-dollar program, but [that 10 billion dollars] was spread over decades. People were getting uncomfortable with even that amount of [software] spending. Palantir had ambitions to grow to the same size as other Silicon Valley technology giants, and getting a small part of a 10-billion-dollar contract simply would not be enough. [Palantir] needed to expand, and fast. Despite their early problems raising money, Dr. Karp and the rest of the Palantir team had become quite good at [raising money] over the years. [Palantir] now had a clear goal: to take the company public at a valuation in the tens of billions of dollars. [A public listing] looked a long way away, and things were getting somewhat desperate [at Palantir]. — _PLTR_ **June 12, 2017.** The trick behind how a chatbot knows what the word "its" is pointing to comes from a paper called **"Attention Is All You Need,"** published in 2017 by a team mostly at Google. ["Attention Is All You Need"] introduced the **transformer**, the design [OpenAI's] GPT-2 is built on. — _LLM_ **2017–2018 (JEDI launched).** [By the late 2010s,] years of small test projects had left the [United States] military with over 500 different external cloud-computing contracts. [The proliferation of cloud contracts] was a mess that needed tidying. The military had tons of data to sift through, and centralizing [that data] safely and securely would allow for better coordination between different areas of the military. [Better coordination] would ultimately lead to more effective performance on the battlefield. The Department of Defense went looking for a cloud service provider to [centralize its data]. [The Department of Defense] launched the **Joint Enterprise Defense Infrastructure** project, or **JEDI**. Microsoft immediately saw the opportunity [in JEDI]. JEDI was not as flashy as IVAS, but [JEDI] was arguably more important. [JEDI] was an up-front 10-billion-dollar commitment with potentially much more further down the road. Winning the JEDI contract would cement Microsoft as the military's most trusted vendor. Despite many advantages, Microsoft was still an underdog [for JEDI]. AWS was bigger [than Azure], and the JEDI contract looked tailor-made for AWS. Other cloud providers, sensing that Amazon was the strong favorite, argued that the Department of Defense should split up the [JEDI] contract so they would at least have a chance at some of the business. The whole point of JEDI was to centralize the military's cloud systems, and dealing with multiple vendors just did not make sense, so every major cloud provider would have to fight it out during the bidding process. Five companies made it to the [JEDI] starting line: Amazon, Microsoft, Google, IBM, and Oracle. Pretty quickly, three of [the JEDI bidders] crashed out. Google said it would not be working with the military on ethical grounds. IBM and Oracle were eliminated [from JEDI] because they simply did not have enough capabilities. Ultimately, [JEDI] came down to a battle between Microsoft and Amazon. Everyone thought that Amazon was going to win [JEDI], except, apparently, President [Donald] Trump. One of President Trump's least favorite people was Jeff Bezos. President Trump did not care much about Amazon, but Jeff Bezos was also the owner of the _Washington Post_, and President Trump hated the _Washington Post_. President Trump complained that reporters from the _Washington Post_ were calling every hour asking ridiculous questions and said that [the _Washington Post_] was owned as a toy by Jeff Bezos. — _MSFT_ **2018 (Project Maven).** Thousands of Google workers protested Google's involvement in a Pentagon drone program [**Project Maven**]. Google worked with the Pentagon on an AI project to improve the precision of drone operations on the battlefield. That collaboration between Google and the Pentagon was shut down after more than 3,000 Google employees signed an open letter. [The Google employees] got their way, but [according to John Coogan] what they did not understand is that Silicon Valley was created because the United States government needed access to the best scientists and engineers in order to defeat an existential threat. [Coogan argued that] the need [for that partnership] is as great as ever. Most technology workers [are described by the source as hating] the idea of working with the military. American technology companies had largely turned their backs on the United States military. — _MSFT_; _VALLEY_ **October 2018 (Bezos at Wired25).** When _Wired_ magazine turned 25 in 2018, it threw a party, and the party had a surprise speaker: Jeff Bezos. Jeff Bezos said something during his [Wired25] talk that ruffled some feathers. Jeff Bezos said [at Wired25] that [Amazon] would continue to support the Department of Defense and that it should. Jeff Bezos said that one of the jobs of a senior leadership team is to make the right decision even when it is unpopular. Jeff Bezos said that if big technology companies were going to turn their backs on the United States Department of Defense, the country was going to be in trouble. Jeff Bezos knew [his statement] would not be popular with the people in the room, but he knew what he was talking about, [given his grandfather's role at ARPA]. Jeff Bezos's comments have aged very well. — _VALLEY_ **2018–2019 (IVAS genesis).** In 2019, the Army invited Microsoft engineers to a military base to teach them what soldiers needed from an augmented-reality headset. That weekend marked the birth of the **Integrated Visual Augmentation System** (IVAS). — _MSFT_ **2019 (GPT-2, WebText, and tokenization).** [In 2019, OpenAI released] **GPT-2**. When a user types a message, a chatbot reads neither letters nor quite words, but **tokens**. OpenAI's GPT-2 splits a sentence so that most pieces are whole words with the preceding space glued on. Rarer words get chopped up [by GPT-2's tokenizer]. The word "chatbot" arrives [in GPT-2] as two pieces: "chat" and then "bot." "ChatGPT," a name nobody had heard of when GPT-2 came out, falls apart [in GPT-2's tokenizer] into "Chat," "G," and "PT." OpenAI published the source of [GPT-2's token] pieces as a plain text file. [GPT-2's tokenizer file] is a list of **merge rules** learned from real text. The first [GPT-2] merge rule glues a space onto the letter "t." A few lines later [in GPT-2's merge file], that "space-t" joins "h" and "e" to make the word "the." [GPT-2's merge] file holds 50,000 merges. [GPT-2's merges] sit on top of the 256 raw bytes every computer already uses, plus a single special token at the end. Adding it all up gives GPT-2's whole menu of **50,257** pieces. GPT-2's menu is fussy about details: a capital "The" at the start of a message is one token, a capital "The" after a space is another, and a lowercase "the" after a space is a third. A chatbot reads chunks from a fixed menu, never the words a user thinks he typed. A chatbot turns every word into a number because the model is arithmetic all the way down, and arithmetic can do nothing with the letters B-E-S-T. Each token [in GPT-2] gets swapped for its position on GPT-2's menu, and from then on those numbers are all the model ever sees. A position on [GPT-2's] list is just an address, like a house number. ["Ability"] lives at [token number] 2,694, and ["best"] lives at [token number] 1,266, but that does not make "ability" bigger or better than "best." The model just uses the address to look something up. Each [GPT-2 token] address points to a row in a huge table. In the smallest GPT-2, every row holds 768 numbers. Nobody typed in [GPT-2's embedding] numbers by hand; training set them, nudging them over and over until tokens used in similar ways ended up with similar rows. The rows closest to "Tuesday" [in GPT-2's embedding table] belong to "Wednesday," "Monday," and "Thursday." Everything becomes numbers because a computer can compare, add, and multiply numbers, and meaning has to become something [a computer] can do math on. [The embedding row] is where every token starts before the words get to interact with each other. [In GPT-2,] attention lets each token look back at the tokens before it and decide how much each one matters. [Through attention, the token] "its" can lean hard on "AI," mostly ignore a word like "the," and blend in what it found until the numbers for "its" carry a bit of "AI." [GPT-2's attention] has one strict rule: a token can only look backward. The ["Attention Is All You Need"] design **masks out the future** so that no position can peek ahead, because the model's whole job is guessing what comes next. The smallest GPT-2 stacks twelve layers of attention, and each layer runs several **heads** side by side, each tracking its own kind of link. [GPT-2] can look back across as many as 1,024 tokens at once. Layer by layer, every token [in GPT-2] pulls in what it needs from the words before it. By the top of [GPT-2's] stack, the last word carries a summary of the whole sentence, and that [final token] is what gets scored. One final step checks [the final token's] summary against every entry on [GPT-2's] menu and gives each entry a raw score. A function called **softmax** squashes [GPT-2's raw] scores into odds, and all the odds together add up to a certainty. Why "learn" scores much higher than, say, "Tuesday" comes down to what the model saw in training. OpenAI built a training set it called **WebText** by following links people shared on Reddit, as long as the post received at least three karma. After cleanup, WebText came to a little over 8 million documents. Training is one game played over and over: hide the next token, let the model guess, then nudge [the model's] numbers so the right answer becomes a little more likely next time. The largest GPT-2 has 1.5 billion of those numbers to nudge. [A language model's] odds are learned from what tended to come next across millions of pages people chose to share. [Taken end to end, GPT-2's generation pipeline runs as one causal chain: text is split into tokens from the 50,257-piece menu; each token becomes an integer address; each address retrieves a 768-number embedding row; twelve layers of masked attention let each position draw on the positions before it; the final position's summary is scored against the whole menu; softmax converts those scores into odds; a token is drawn from the odds; and the drawn token is appended and fed back in until token 50,256 ends the text.] "Learn" scores well because in text like [the example sentence], "learn" is a word that often follows. A chatbot [such as GPT-2] picks [a token such as] "learn," adds it to the end, and feeds everything back in to get fresh odds. [GPT-2] stops when it draws token number **50,256**, the token that means the text is over. What changed [since Claude Shannon's 1948 experiment] is how far back [the model] can look: Shannon's book trick looked back one word, while GPT-2 looks back over a thousand tokens. [The longer look-back] is a big reason GPT-2 holds together while Shannon's [1948] text falls apart. — _LLM_ **2019 (Neural text degeneration).** A team at the University of Washington described the trap [of always choosing the most likely next word] and called it **degeneration**. [The University of Washington] paper shows a larger GPT-2, told to hunt for the most likely text, repeating the name of a Mexican university [the Universidad Nacional Autónoma de México] over and over. [Degeneration is the deterministic end of the sampling spectrum, fully random sampling from the raw odds is the opposite end where text dissolves into gibberish, and temperature-weighted sampling is the working regime between them; Stephen Wolfram's 2023 demonstration with GPT-2 reproduced the deterministic failure step by step.] — _LLM_ **2019 (Windows 10 contract).** [In 2019,] Microsoft signed a five-year contract worth nearly a billion dollars to provide Windows 10 and associated hardware to the [United States] military. — _MSFT_ **October 2019 (JEDI awarded to Microsoft).** Amazon said it was considering its options after the Pentagon's surprise decision to award the 10-billion-dollar [JEDI] cloud-computing contract to Microsoft. There was even a report that President Trump directly called Secretary of Defense [and retired] General [James] Mattis and ordered him to "screw Amazon." Amazon made what appeared to be an unprecedented request in its case against the government over the 10-billion-dollar JEDI contract. Amazon sought to depose President Trump, Secretary of Defense Mark Esper, and former Secretary of Defense James Mattis in [the JEDI] case. When Microsoft won the [JEDI] contract, many people, including Amazon, thought that President Trump had tipped the scales. Microsoft was nonetheless a solid choice [for JEDI]. By [2019], Microsoft had already been doing business with the Department of Defense for decades. Microsoft was making more money from the American defense establishment than Amazon was. Building products specifically tailored to enterprise clients was what Microsoft excelled at. The Department of Defense was a big, complex customer, but it was not looking for anything flashy; it just wanted something that worked. Microsoft CEO Satya Nadella said that if technology companies do not help the United States, [the country is] in a whole load of trouble. [The JEDI award rests on three separate layers: the accusation of presidential favoritism, advanced by Amazon and others; the reported executive pressure, namely the account that President Trump told James Mattis to "screw Amazon"; and Microsoft's own suitability, grounded in decades of Department of Defense business, larger existing defense revenue than Amazon, and enterprise-tailored product design.] — _MSFT_ **2020 (The viral Pentagon slide).** Ten years [after the McChrystal slide], things had not improved much. A different Pentagon PowerPoint slide showing how different parts of the military were interconnected went viral in 2020. [The 2020 Pentagon slide] was also an abomination of communication. There was a reason the United States military kept producing these awful diagrams: [the military] was in the midst of a decades-long battle to figure out how all the different parts of the armed forces could work together more efficiently. — _ABC_ **2020 (COVID-19).** Building out [Palantir's] air-travel business seemed like a good idea at the time, and it was a good idea at the time. At the start of 2020, the whole world got an important lesson in why air travel is a volatile business. Governments responded to the outbreak of COVID[-19] by introducing restrictive lockdowns, and planes everywhere were grounded. [ABC asked] why anyone would need planes if no one could fly. A crisis should never go to waste, and as the pandemic started to spread globally, Dr. Karp and his team started to see an opportunity to help. By [2020], Palantir had fifteen years of experience working with governments and handling sensitive data. There had to be a way for Palantir to make a positive impact [during the pandemic]. [Palantir decided] to offer its product for free to combat COVID[-19], rebuilding supply chains and figuring out who was becoming infected and under what conditions. [Palantir] wound up working closely with the United States Department of Health and Human Services to merge around **two billion data elements** related to the outbreak, to help government officials get a clearer picture of how the virus was spreading. [Palantir] even partnered with the United Nations **World Food Programme** to help get food and supplies distributed during the pandemic. [Palantir's humanitarian work] came as a surprise to Palantir's many critics, who thought that the company could only work with the military. — _PLTR_ **2020 (Remote forward deployment).** Prior to the outbreak of COVID[-19], Dr. Karp was extremely focused on getting engineers embedded with the organizations [Palantir] was helping. Alex Karp said it is better to have engineers figure out what the core issues are and then iterate against them. Alex Karp said [Palantir] was not planning to hire salespeople, still had not hired any, did not often hire non-technical people, had no marketing department, and was not planning to get one. The problem was that all of [Palantir's] forward-deployed engineers were extremely expensive. Palantir was spending millions of dollars flying [forward-deployed engineers] around the world and putting them up in hotels while they helped customers implement the software. COVID[-19] revealed that [forward-deployed engineers] might be just as effective working remotely. Without the option to travel, Palantir naturally cut costs significantly. Interestingly, [Palantir's] revenue kept climbing [during the pandemic]. — _PLTR_ **September 2020 (Palantir IPO).** Seventeen years after it was founded, Palantir went public in September 2020 at an initial valuation of [about] 15.2 billion dollars. [Palantir's September 2020 listing came on the back of the pandemic year's economics: remote forward deployment had cut Palantir's largest cost center while revenue kept climbing, and that widening margin made the long-sought public offering viable.] — _PLTR_ **March 2021 (IVAS contract).** Microsoft won a massive 22-billion-dollar contract for a project called the **Integrated Visual Augmentation System**, or IVAS. [IVAS] will see over 100,000 American soldiers equipped with augmented-reality headsets based on HoloLens technology. The IVAS project is actually just the latest installment of the military's obsession with augmented reality, [which] dates back to the 1980s. Given the decades of work that had gone into [the military augmented-reality] project and Microsoft's years of experience with augmented reality, one might think that progress [on IVAS] would be smooth. IVAS was plagued with problems from day one, and it felt as though everything was back at square one. Billions of dollars later, Microsoft and the Army are determined to make IVAS work. [IVAS] has huge potential to improve soldiers' awareness and effectiveness. Russia and China are working on their own IVAS-style programs, so the United States is determined not to fall behind. — _MSFT_ **July 2021 (JEDI canceled).** Ultimately, JEDI ended with a big anticlimax when the Pentagon canceled the 10-billion-dollar JEDI deal with Microsoft. — _MSFT_ **November 2022 (ChatGPT).** In 2022, people were starting to get excited about the potential uses of large language models like **ChatGPT**. [Commentators noted that ChatGPT] could create custom code, write entire books from scratch without plagiarism, and write essays. At Palantir, [executives] were quickly coming to a realization. [Palantir] had been building software to be used by humans, but the way [Palantir] had built it was also perfect for large language models. — _ABC_ **December 2022 (JWCC).** JEDI was replaced with the **Joint Warfighting Cloud Capability** (JWCC) project. [The JWCC] contract, worth 9 billion dollars, was split among four of the five companies originally in the JEDI bidding process. Microsoft got a piece [of JWCC], but it was not the big win [Microsoft] had hoped for. [JWCC reversed JEDI's founding premise: where JEDI was designed as a single-vendor, winner-take-all centralization of the military cloud, JWCC distributed the work across Amazon, Google, Microsoft, and Oracle, trading vendor concentration for multi-vendor competition and leaving no single company with the strategic position JEDI had promised.] — _MSFT_ --- ## VIII. The Artificial Intelligence Platform and Palantir's Ascent (2023–2026) **January 2023 (Roadmaps torn up).** [Palantir CTO Shyam Sankar said that Palantir] was pleasantly surprised to see how much the world [Palantir] had been building for met its moment with LLMs. Shyam Sankar said that LLMs were not just great at reading [Palantir's] data maps but needed [Palantir's] data maps in order to be reliable. Shyam Sankar said that one cannot unleash the value of an LLM without [Palantir's ontology structures]. [Palantir] had spent [about] twenty years building a system [the Ontology] that organized data in a way that could be intuitively read by the new wave of AI systems. [The Ontology] was laid out in an ideal way for an AI system to produce reliable information. [A Palantir executive] said that by January [2023] it was obvious that [Palantir] needed to tear up all its roadmaps and get excited about how it could incorporate LLMs into its software to provide a whole new series of experiences. — _ABC_ **February 2023 (Wolfram on next-token prediction).** [Stephen] Wolfram typed "The best thing about AI is its ability to" into GPT-2, [whose] favorite next word was "learn," but [GPT-2] was only 4.5 percent sure of it. Stephen Wolfram pulled [GPT-2's next-token] odds straight out of the model. A chatbot does not know what it is going to say. At every step, a chatbot puts odds on every piece of text it knows and then rolls a loaded die. The favorite [token] wins much of the time, but not every time, which is why the same question asked twice can get two different answers. A whole [chatbot] reply is a chain of small bets placed one piece at a time, and each bet can only see what is already on the page. For [Wolfram's example] sentence, [GPT-2] was not very sure of its next word, since even its best guess was a long shot. Stephen Wolfram tried having [GPT-2] always pick the most likely word with [the example] sentence. Always taking the top word, GPT-2 finished [Wolfram's] sentence with "learn from experience," which was fine at first. A few sentences later, [GPT-2] got stuck and kept repeating "it's a very good example of how to use AI to improve your life" again and again, word for word. Going fully random does not work either, because drawing straight from the raw odds gives every weird long shot its turn, and the text slides into gibberish. Chatbots split the difference by drawing at random but weighted by the odds. A setting called **temperature** decides how loaded [the chatbot's] die is. Turning [temperature] down makes the favorite win more often, and turning it up gives the long shots more turns. Stephen Wolfram found that [a temperature of] 0.8 worked best for essays. [Chatbots do not always pick the favorite] because the safest word at every step adds up to a reply that is dull and goes in circles. A little bit of chance is what makes [chatbot] writing sound as though a person wrote it. [The loop Stephen Wolfram describes is the same loop Claude Shannon ran by hand with a book in 1948: choose a next word conditioned on what came before, append it, and repeat; GPT-2 differs from Shannon's procedure by conditioning on up to 1,024 prior tokens through learned weights rather than on a single prior word through a page flip.] Anthropic describes pretraining its own models the same way: as learning to predict the next word from the text before it. Newer chatbots get extra training, but [newer chatbots run] the same loop underneath, which is how a chatbot can sound completely sure and still be wrong. — _LLM_ **April 2023 (AIP).** [Palantir] developed something called the **Palantir AIP**, which stands for **Artificial Intelligence Platform**. [Palantir describes AIP as an] AI operating system. [With AIP,] individual users do not need to look at the [Ontology's] marble map at all; [a user] can just use AIP, which sits on top of the marble map and looks at it for [the user] far faster than any human could. To use [AIP], all a user has to do is ask it a question. [In Palantir's AIP for Defense demonstration, an operator asked] what enemy units were in the region. The human operator can ask AIP in plain English to deploy surveillance drones. [In the AIP demonstration, the operator instructed AIP to] task the [MQ-9 Reaper] to capture video of a location. The [MQ-9] drone footage showed an enemy **T-80** main battle tank. The operator can then ask [AIP] to look at what options are available in the area to destroy [the T-80] tank. [The operator instructed AIP to] generate three courses of action to target the enemy equipment and send the three options to the commander for review. [In the AIP demonstration,] the commander selected and approved course of action three. Until recently, executing this kind of operation would have required big maps rolled out on tables, little wooden figures pushed around on top, and telephone calls up and down the chain of command. AIP allows [the same targeting workflow] to be done as easily as asking ChatGPT to plan a holiday to Queensland. [AIP changed the interface layer while keeping the data layer: the Ontology remained the structured substrate, but where Gotham and Foundry had required a trained analyst to navigate maps, graphs, and entity records by hand, AIP placed a natural-language control surface over that substrate, so that an operator could query it, task sensors, generate courses of action, and route decisions to a commander in plain English, with the commander's approval retained as the final step.] — _ABC_ **2023 (AI-accelerated deployment and commercial sprawl).** Not only can AI be used to operate Palantir software, but [AI] can also be used to design [Palantir software]. What once took software engineers years of observing operations and writing code [to build Palantir ontologies] now takes weeks. [AI-accelerated ontology construction] meant that Palantir suddenly had much extra time and many extra resources. Palantir started expanding. Previously, pretty much only the United States government could afford Palantir's services, [but by 2023] Palantir was getting into everything. **ExxonMobil** added powerful new features by partnering with Palantir Technologies. [Palantir offered an] AIP customer-service engine. [The Australian supermarket chain] **Coles** and the CIA both use Palantir. [The **European Cricket Network** adopted] Palantir as the platform for making cricket the number-one team bat-and-ball sport in Europe, [which ABC called] the most out-of-pocket of Palantir's clients. [A European Cricket Network executive said that,] with entrepreneurial intuition, [the network] could now use Palantir to ask the right questions of the data, all in one place. [Palantir's marketing stated that] maintaining United States and Western military superiority requires deliberate action and technological innovation and that Palantir delivers secure, innovative, and scalable software solutions at speed. [Palantir's marketing stated that,] from the global war on terror to space-domain awareness, Palantir continues to be a mission partner for the United States and allied nations. [The client roster demonstrates the Ontology's portability across domains: the same architecture that fused CIA, FBI, and NSA holdings also structures an energy major's operations, a supermarket chain's business, a sports network's data, an aircraft manufacturer's production line, military logistics, and immigration enforcement, because the Ontology models objects, relationships, and actions generically rather than encoding any single industry.] The rapid expansion in [Palantir's] clients made markets extremely excited about Palantir's potential future profits. In 2023, Palantir stock jumped 167 percent. — _ABC_ **2023 (Microsoft's position).** [As of 2023,] Microsoft had used its traditional strengths to build a military empire. [John Coogan] mapped out each of Microsoft's major military projects and concluded that Microsoft is trying to completely change the nature of warfare, but that [Microsoft's military efforts have] not all gone exactly as planned. [While researching Google's work with military drones,] John Coogan noticed footage of an Air Force officer flying a 60-million-dollar drone using an **Xbox controller**. The current plan [as of 2023] was to get working HoloLens [IVAS] headsets into the field by the end of 2023, but ultimately IVAS might not even matter, because Microsoft's work with the military goes much deeper. Led by its general, Bill Gates, Microsoft grew into a behemoth that understood its allies and ruthlessly destroyed its enemies. Today, Microsoft's work with the military is based on the same principles. [Microsoft's military] setbacks do not matter very much in the end. What matters is that Microsoft has become one of the United States military's most trusted vendors because [Microsoft] is more like the military than any other big technology company. Today, the United States military runs on **Windows**, **Excel** spreadsheets, and the **Microsoft cloud**. [Microsoft's military stack] is not necessarily flashy, but winning wars is a grind, and Microsoft knows that better than almost anyone. As war continues to become more digital throughout the twenty-first century and the Defense Department looks for reliable private-sector partners, Microsoft has made itself indispensable. — _MSFT_ **2023 (The new defense entrants).** The world is entering an era of intense **strategic competition**, [according to John Coogan]. The threats [of strategic competition] are diverse: China, terrorism, rogue AI, and black-hat hackers in Russia. Managing [those threats] will require a close partnership between technology companies and the Pentagon. Some big technology companies understand the stakes [of strategic competition]. Microsoft has worked closely with the military for decades. The United States government is investing billions of dollars to meet the threats of today and tomorrow. [United States government investment] is enabling the emergence of a new type of defense company, [including] **Anduril**, **Palantir**, and **SpaceX**, that want to do totally new things instead of just doing the old thing better. [Anduril, Palantir, and SpaceX] all follow in the footsteps of men like Fred Terman. — _VALLEY_ **2023 (Palantir's AI posture).** Large language models have taken the world by storm, and every company that can claim even a tangential relationship to the technology has benefited. Palantir has actually been working on artificial intelligence for years. Alex Karp said that, unlike most people, [Palantir had] been involved in what people call AI for the previous five, six, or seven years in the classified environment, building systems that identify adversarial positions. [Alex Karp] seemed a little annoyed that everyone was jumping on the AI bandwagon, [since] it is easy to put Palantir in the same bucket as companies merely riding the hype wave, but [Palantir] is happy for its work to do the talking. Palantir has faced much criticism over the years. Some [criticism of Palantir] has come from other technology companies that turn their noses up at working with the military. Much [criticism of Palantir] has come from civil libertarians who believe Palantir helped build technology used to spy on American citizens. Some [criticism of Palantir] has come from people who believe Palantir is more of a consulting company than a software company. [The consulting critique holds] that Palantir builds too much custom software for each customer, which hurts the versatility of the core Palantir product. As AI becomes more useful for processing information and gleaning insights from large datasets, [Palantir's] product should naturally become more generalizable. Palantir is now a twenty-year-old company, but Dr. Karp, the former philosopher who went rogue, is still the CEO. Alex Karp believes that everything taken for granted in [American] democracy is underpinned by national security. America continues to need the support of technology companies, and Dr. Karp understands the importance of a stable world order. Big tech has had a tumultuous history with the Department of Defense: Palantir had to sue the government to get a fair shot at a contract, while Google dropped out of a major military AI program. Alex Karp said [of Palantir's people] that they are underpaid and do [the work] because they love it and continue doing it because they love it. [One commentator] called Palantir the Rudy Giuliani of technology. — _PLTR_ **2024.** In 2024, Palantir stock jumped by 340 percent. [A Wall Street analyst said] there was a golden path for [Palantir] to monetize what [the analyst's firm] viewed as potentially a 1-trillion-dollar market opportunity. Palantir has become one of the fastest-growing companies on Earth, with clients all over the world, including government contracts in many different countries. — _ABC_ **2025 (Top performer and a public CEO).** Palantir was the top-performing stock on the entire United States stock market in 2025. These days, Palantir is involved in everything from retail to military operations to immigration enforcement. Up until recently, [Palantir's] CEO tried to remain out of the spotlight. Since becoming more of a public figure, Alex Karp has said that [Palantir is] "at the core of making the obvious superiority of the West." Alex Karp has said that [Palantir's] product is sometimes used to kill people. Alex Karp has said that [Palantir has] dedicated the company to the service of the West and the United States of America. Alex Karp has said that Palantir is there to disrupt and, where necessary, to scare [America's] enemies and on occasion kill them, adding that he hoped [listeners] were in favor of that. [A sponsor message at one event read,] "Special thanks to our sponsor, Palantir!" — _ABC_ **2025–2026 (Public reaction).** [ABC described Palantir as] the multinational corporation of the moment [in public suspicion]. [Media coverage described Palantir as] largely closed and specializing in the practice of data mining. Despite its exponential growth in size and strategic significance, Palantir is still a bit of a mystery to most people. [By 2026,] many people were starting to think that Palantir was tipping [John O'Neill's ordered-liberty] balance toward order and away from liberty. [Public commentators compared Palantir to China's treatment of its citizens, called it a "surveillance state" and a "predictive policing firm," labeled it a CIA front company, and attacked its leadership in personal terms.] There are more hostile theories swirling around Palantir than around any other Silicon Valley company. [In ABC's analysis, the hostility persists] because, even though [Palantir] is in essence a very sophisticated PowerPoint slide that makes intelligence, military logistics, customer service, and cricketing operations more efficient, its leaders have decided to cultivate a deliberately provocative public image. — _ABC_ **May 8, 2026 (ABC's assessment).** In the aftermath of 9/11, the [American intelligence] problem was not just intelligence failure but information stuck in silos. The FBI and the CIA had pieces of the puzzle, but no shared picture. Palantir was a company built on the premise that data, if stitched together properly, could surface threats before they metastasized. Co-founded by Peter Thiel and Alex Karp, Palantir's early pitch was deceptively simple: give analysts the ability to see connections across messy datasets without compromising privacy. The arrival of large language models has supercharged what Palantir was already doing: ingesting, structuring, and interrogating enormous amounts of information. [In ABC's assessment,] the result is a shift from finding needles in haystacks to, arguably, predicting where the needles will land. In the last three years or so, one company, [Palantir,] appears to have finally figured out [how the different parts of the armed forces can work together efficiently]. [In ABC's assessment,] Palantir is basically a glorified version of the PowerPoint slide General [Stanley] McChrystal was considering during the war in Afghanistan. [Palantir] was designed to help keep the balance that FBI agent John O'Neill thought was so important. [ABC announced that it would spend] the next episode on Alex Karp and what he wants to do with the company that now controls so much of the world's data, and that a further episode would cover how [Palantir] is being used for immigration enforcement. — _ABC_ --- ## Wiki Integration The canonical relationship router is [[wiki/Signals, Silos, and Sovereign Software|Signals, Silos, and Sovereign Software]]. It connects this chronology to [[collections/The Civilian Intelligence Marketplace|The Civilian Intelligence Marketplace]] and [[collections/Terrorism, Counterterrorism, and the Intelligence Environment|Terrorism, Counterterrorism, and the Intelligence Environment]] and preserves the central progression **signal → detection → fusion → representation → prediction → command**. The Stanford–Palo Alto route is [[wiki/Radio Research Laboratory|Radio Research Laboratory]] → [[wiki/Fred Terman|Fred Terman]] → [[wiki/Applied Electronics Laboratory|Applied Electronics Laboratory]] → [[wiki/Stanford Industrial Park|Stanford Industrial Park]] / [[wiki/Stanford Research Park|Stanford Research Park]] → [[wiki/Varian Associates|Varian Associates]], [[wiki/Hewlett-Packard|Hewlett-Packard]], Lockheed, and [[wiki/Fairchild Semiconductor|Fairchild Semiconductor]] → [[wiki/Military Origins of Silicon Valley|Military Origins of Silicon Valley]]. ## Corrections Ledger **Transcription corrections (names, places, programs).** "Kamu bear / camhu bear / cam Huber line" → **Kammhuber Line** (after Josef Kammhuber). "Himel bet / hamilton zone / himobat" → **Himmelbett**. "Wartersburg / warsberg / wordersburg / wurzberg" → **Würzburg**. "Naxos" retained. "Vanavar / vanivar Bush" → **Vannevar Bush**. "Turman / Kerman" → **Frederick (Fred) Terman**. "Gamora" → **Gomorrah**. "Carnegie Institute of science" → **Carnegie Institution** (of Washington, now Carnegie Institution for Science). "Rivien" → **Rivian**. "Sigurd" retained for **Sigurd Varian**. "Lawrence Preston guys" → **Lawrence Preston Gise**. "Special research project agency DARPA" → **Defense Advanced Research Projects Agency** (founded as ARPA, 1958). "Anderill / Andrew" → **Anduril**. "Palunteer / palantir" → **Palantir**. "Peter Teal / Dio" → **Peter Thiel**. "Havford" → **Haverford**. "Jurgen Hobbermas" → **Jürgen Habermas**. "Cadmin Group" → **Caedmon Group**. "Joe Lansdale" → **Joe Lonsdale**. "Steven Cohen" → **Stephen Cohen**. "Po Dexter" → **John Poindexter**. "Eigor" → **Igor**. "Inqel / Incel / Inutel" → **In-Q-Tel**. "Cheyam / Chem / Jam Sankar" → **Shyam Sankar**. "JCR Licklider" retained as **J. C. R. Licklider**. "Matt Grim" → **Matt Grimm**. "DIGs / distributed common ground surface system / D6A" → **Distributed Common Ground System (DCGS)** and **DCGS-A**. "Loheed" → **Lockheed**. "Booze Allen" → **Booz Allen**. "To France" → **Toulouse, France**. "Badabad / Ibadabad" → **Abbottabad**. "0 dark30" → **zero dark thirty**. "Chris Isham" retained (Christopher Isham, ABC News); "ABC America" → **ABC News**. "Miller" identified as ABC correspondent **John Miller**. "net Warrior" → **Nett Warrior**. "ivas / ibas" → **IVAS (Integrated Visual Augmentation System)**. "Connect" → **Kinect**. "Balmer" → **Ballmer**. "Bay" (search share) → **Bing**. "Mark esper" → **Mark Esper**. "joint Warfighter Cloud capability" → **Joint Warfighting Cloud Capability (JWCC)**. "Steven Wolfrram / Wolram" → **Stephen Wolfram**. "Open AAI" → **OpenAI**. "the head and infernal attack" → Shannon's actual output, **"the head and in frontal attack."** **Factual adjustments and flags.** The USS Yorktown was a **Ticonderoga-class guided-missile cruiser (CG-48)**, not an aircraft carrier, and the divide-by-zero failure occurred in **September 1997** (the source says 1998, the year it was widely reported); the outage is generally recorded at roughly two and three-quarter hours rather than two and a half. The Soldier's Computer (late 1980s–1992) predates the 1995 Nintendo Virtual Boy, so the display is rendered as the **Reflection Technology Private Eye**, the technology the Virtual Boy later adopted. Lockheed in the 1950s was **Lockheed Corporation**; Lockheed Martin formed in 1995. Army records place Land Warrior's operational deployment to Iraq with the **4th Battalion, 9th Infantry Regiment in 2007**; the source's "2005, 229 soldiers" is retained but flagged as unverified. Fairchild Semiconductor's first order came from IBM (B-70 avionics transistors), so the entry renders Fairchild's Apollo role as its **integrated circuits in the Apollo Guidance Computer** rather than its "first business." The 1993 World Trade Center attackers are bracketed with **Ramzi Yousef**, captured in Pakistan in 1995, since the source's "religious extremists from Pakistan" compresses a more mixed group. Operation Millennium launched **1,047** bombers (rendered "about 1,100"). R. V. Jones's student pranks date to his Oxford years (c. 1929–1934) rather than strictly the 1920s. Palantir's initial funding came from **Peter Thiel personally**; ABC's "first investor was the CIA" is rendered as the CIA's In-Q-Tel being the first institutional investor. The 2012 DCGS-A quotation is attributed, by inference from its content and context, to **General Raymond Odierno**; this identification is strongly indicated but unconfirmed in the transcript. The source's characterization that Palantir's LLM-ready Ontology arose "purely by luck" has been replaced with Palantir's own executive statements about the fit between twenty years of ontology construction and LLM reliability, since the transcript offers no evidence for non-intention and the company's own account describes a deliberate architecture that met its moment. Profane, defamatory, and slur-bearing lines quoted from online commentators in the ABC episode are summarized in brackets rather than reproduced verbatim. **Revision of October 13, 2026 (insertion audit).** This revision adds the narrators' provenance (Soylent, Lucy, Y Combinator, Matt Bevan) as a _Source creators_ note rather than as timeline events. It names the Meta Oculus Quest 2 in bold and marks it as Coogan's present-day comparator. It attributes the claim that no photographs exist of the Radio Research Laboratory to Coogan rather than asserting it categorically. It adds bracketed synthesis sentences that restore compressed mechanisms: the recursive reconnaissance–jamming–counter-jamming loop; the Stanford trustees as the institutional gate on the research park; Land Warrior's combined wearability and environmental failure modes; Palantir's access-controlled correlation design; the gap between technical merit and procurement literacy; In-Q-Tel as translation layer; Thiel's "intermediate hybrid solutions" quoted in full; field IED testimony set apart from the formal 2012 Army study; the causal chain from soldier demand to the 2016 lawsuit; the map-to-person drill-down workflow; Azure credit bundling as a switching-cost channel; the three layers of the JEDI award; JWCC's reversal of JEDI's single-vendor premise; the pandemic cost-and-revenue path to the IPO; the end-to-end GPT-2 pipeline; the sampling spectrum from degeneration to gibberish; the Shannon-to-Wolfram cross-reference; AIP as a natural-language control layer over the Ontology; and the Ontology's cross-domain portability. Several items the audit flagged as absent were already present in the prior version, namely the Quest 2 comparison, the Stanford trustees, the no-photographs claim, the front-end versus back-end prototype testimony, the help-desk testimony, the IED testimony, and the name-and-photograph drill-down; for those, this revision sharpens the existing material rather than duplicating it. --- ## Sources [How did Palantir get so powerful? — _If You're Listening_, ABC News In-depth](https://www.youtube.com/watch?v=PpEg0XIeFtA) (Matt Bevan, May 8, 2026). [Palantir: The App That Caught Bin Laden](https://www.youtube.com/watch?v=EIX7hnkJiuY) (John Coogan). [Old Military Tech Created Modern Smartphones](https://www.youtube.com/watch?v=hVwHFKtEKMw) (John Coogan). Microsoft's Military Empire, Explained (John Coogan, cofounder of Soylent and Lucy, both Y Combinator companies; transcript supplied, URL not provided). An Old Idea (explainer on GPT-2 next-token prediction; transcript supplied, URL not provided).