# Neocloud A neocloud is a cloud operator specializing in accelerator infrastructure rather than offering a general catalogue of services. The category emerged because accelerator computing needs different things from a provider than web hosting does, and because demand arrived faster than general-purpose providers could serve it. > [!note] The one-sentence version > A cloud that does one thing: accelerators, their fabric, their storage and their power, sold as capacity rather than as a service catalogue. ## What distinguishes the model **Specialization rather than breadth.** The offering is accelerator capacity with the fabric, parallel storage, and tooling that make it usable, instead of hundreds of managed services. The customer wants devices and a network between them, delivered quickly. **Bare metal by default.** Physical machines with visible topology and tenant-chosen driver stacks, for the reasons accelerator work requires them; see [[wiki/Bare Metal Compute|Bare Metal Compute]]. **Accelerator diversity.** Several vendors' silicon offered side by side, so a customer can choose hardware per workload rather than committing to one manufacturer or one hyperscale platform. Making a mixed fleet usable depends on the routing and selection layer above it; see [[wiki/Inference Request Routing|Inference Request Routing]]. **Capacity on contract.** Reserved capacity over months or years, which matches how model developers plan and how the operator finances equipment; see [[wiki/Capital Intensity|Capital Intensity]]. **Integration as the core competence.** The operator's skill is assembling accelerators, fabric, storage, power, and cooling into clusters that perform, which is an engineering discipline rather than a reselling one; see [[wiki/Rack and Cluster Integration|Rack and Cluster Integration]]. ## Where the category sits Above it are the hyperscale providers, with vast service catalogues, their own custom silicon, and global footprints; see [[wiki/Hyperscaler|Hyperscaler]]. Below it are colocation operators supplying space, power, and cooling to customers who bring their own hardware; see [[wiki/Colocation|Colocation]]. Specialist operators occupy the layer between: they own and operate the compute, they do not attempt the full service catalogue, and they frequently deploy inside third-party colocation space rather than building every facility themselves, which lets capacity follow available power. ## Why the layer matters nationally Model development, research, and enterprise deployment all require large blocks of accelerator capacity, and the number of organizations able to build their own facilities is small. A competitive layer of specialist operators distributes access to that capacity across many more institutions: universities, hospitals, state agencies, manufacturers, and companies of ordinary size. That distribution is how a national capability becomes a national asset rather than the private advantage of a handful of firms, and it is one of the clearer ways in which the market structure of this industry serves the country's interest; see [[wiki/National Compute Capacity|National Compute Capacity]]. ## Relationships - **offers:** [[wiki/Bare Metal Compute|Bare Metal Compute]] · [[wiki/Inference Endpoint|Inference Endpoint]] · [[wiki/Private Model Hosting|Private Model Hosting]] - **deploys within:** [[wiki/Colocation|Colocation]] - **compared with:** [[wiki/Hyperscaler|Hyperscaler]] - **depends on:** [[wiki/Rack and Cluster Integration|Rack and Cluster Integration]] · [[wiki/Utility Interconnection|Utility Interconnection]] - **civic dimension:** [[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 specialist operator model offering accelerator capacity with fabric, parallel storage, and tooling rather than a broad service catalogue; bare-metal default; multi-vendor accelerator portfolios; reserved multi-month and multi-year capacity contracts; integration as the operator's core competence; deployment inside third-party colocation space alongside self-operated facilities. - **Established (datestamped):** as of October 2026, specialist operators offering several vendors' accelerators within one infrastructure platform are an established industry category. - **Strongly indicated:** the specialist layer broadening institutional access to large accelerator blocks; market structure supports it and distributional measurement is limited. ## Sources - Specialist accelerator cloud market practice and operator service descriptions, as reviewed October 2026.