# Capital Intensity A compute campus is among the most capital-intensive structures private capital builds, and the distribution of that capital is counterintuitive: the building is the cheap part. Understanding the ratio explains the industry's behavior almost completely, including its urgency, its appetite for on-site generation, and its tolerance for paying whatever an early energization date costs. > [!note] The one-sentence version > The shell and the plant are a minority of the money; the computers inside are the majority, and they begin losing value the day they arrive. ## The shape of the spending Spending divides into three classes with very different lifetimes. Land, shell, and sitework last forty years or more. Electrical and mechanical infrastructure, the substation, transformers, switchgear, uninterruptible power, generators, chillers, and piping, lasts fifteen to thirty. Computing equipment, the accelerators, servers, switches, and storage, is replaced on a cycle measured in a few years. In a dense accelerator facility the third class dominates the total, and by a wide margin. That ratio inverts the intuitions of ordinary real estate. The building exists to serve the equipment, the equipment's clock is short, and therefore every decision is subordinated to getting equipment into production early and keeping it busy. A month of delay is not a month of deferred rent; it is a month of a depreciating asset earning nothing; see [[wiki/Depreciation of Compute Assets|Depreciation of Compute Assets]] and [[wiki/Utilization|Utilization]]. ## What follows from it **Schedule is worth almost any reasonable premium.** Hence phased energization, prefabrication, and willingness to build private generation rather than wait in a queue; see [[wiki/Construction Sequencing|Construction Sequencing]] and [[wiki/Behind-the-Meter Generation|Behind-the-Meter Generation]]. **Utilization is the central operating metric.** Capital this large, depreciating this fast, must be kept working; an idle accelerator is the most expensive object in the building. **Power cost compounds enormously.** A facility drawing hundreds of megawatts continuously spends very large sums on electricity over a long contract, which is why long-term power agreements and efficiency are financial rather than cosmetic concerns; see [[wiki/Power Purchase Agreement|Power Purchase Agreement]] and [[wiki/Power Usage Effectiveness|Power Usage Effectiveness]]. **Financing is structured against contracts.** Projects of this scale are funded against committed demand, which is why offtake agreements, prepayments, and long leases precede construction, and why capital at this scale follows contracted revenue rather than speculation about it. **Infrastructure is built to outlast several equipment generations.** Since shell and plant are the cheaper and longer-lived classes, overbuilding them modestly, with higher floor loading, greater clear height, spare electrical room, and reserved routes, is inexpensive insurance against the next generation's requirements; see [[wiki/Data Center Shell|Data Center Shell]]. ## The industrial scale of it The aggregate is historically large. Several major firms have each indicated spending on the order of hundreds of billions of dollars in a single year across the compute and energy stack, and credible projections place total infrastructure spending for this transition in the trillions. Figures of that size are easier to read as what they physically are: substations, transmission, turbines, fabs, packaging plants, buildings, and the wages of the people who build them. This is an industrial buildout of the kind a country undertakes a few times a century, and it is being financed largely by private capital against contracted demand. ## Relationships - **driven by:** [[wiki/Depreciation of Compute Assets|Depreciation of Compute Assets]] · [[wiki/Accelerator|Accelerator]] cost - **managed through:** [[wiki/Utilization|Utilization]] · [[wiki/Construction Sequencing|Construction Sequencing]] - **exposed to:** [[wiki/Supply Chain Chokepoint|Supply Chain Chokepoint]] · [[wiki/Interconnection Queue|Interconnection Queue]] - **financed against:** [[wiki/Power Purchase Agreement|Power Purchase Agreement]] · [[wiki/Bare Metal Compute|Bare Metal Compute]] contracts - **civic dimension:** [[wiki/The American Buildout|The American Buildout]] · [[wiki/National Compute Capacity|National Compute Capacity]] - **source articles:** [[articles/Vertically Integrating an AI Superpower from AI Factory to Citizen|Vertically Integrating an AI Superpower from AI Factory to Citizen]] · [[articles/The Gigawatt Is the New Measure of War|The Gigawatt Is the New Measure of War]] - **collection route:** [[collections/American Infrastructure|American Infrastructure]] · [[collections/American|American Collection]] ## Evidence posture - **Established:** the three asset classes and their differing service lives; the dominance of computing equipment in total project capital at dense accelerator facilities; the consequent primacy of schedule and utilization; project financing against committed offtake. - **Established (datestamped):** as of October 2026, several major firms have each indicated spending on the order of hundreds of billions of dollars within a single year across compute and energy infrastructure. - **Plausible:** total transition infrastructure spending in the trillions of dollars; these are projections by industry analysts rather than measured outcomes. ## Sources - [How the Electrical Grid Is Being Rebuilt for AI](https://www.bloomberg.com/) — Bloomberg Primer, October 2026; four companies intending to spend more than three hundred billion dollars in a single year. - [How AI Is Pushing the Semiconductor Supply Chain to the Limit](https://www.bloomberg.com/) — Bloomberg Originals, October 2026; projected infrastructure spending crossing a trillion dollars and projected chip industry revenue reaching a trillion dollars in 2026.