Build. Power. Scale.
DatacentersX follows AI compute through the whole stack: the server and rack, the network fabric that couples chips into one machine, the power that feeds it and the cooling that carries its heat away. Every scale gets the same four questions, from a humanoid robot to a gigawatt campus to a satellite constellation, and every layer is read for how much of it the United States can build.
Hubs
The stack runs from the server to the campus. Power comes in, heat goes out, and the same machine shows up at eight scales. The deployments, the supply chain and the operating disciplines each have a hub of their own.
Server, rack, cluster, facility and campus, with the network fabric that turns thousands of accelerators into one machine. Each rack-scale platform carries a supplier ring and a US-made share. Chips and modules hand off to SemiconductorX.
Power in: grid ties, behind-the-meter gas, nuclear, microgrids and battery storage.
Heat out: air handling, direct-to-chip liquid, immersion, cooling towers and the water behind them.
The scale ladder, from vehicles and humanoids as mobile compute nodes to edge sites, hyperscale regions, AI factories and orbital compute.
Training, inference, HPC, cloud, content and the regulated industries, and where each one runs.
Exemplar campuses chosen for the model they define, by energy model, cooling regime, coherence and deal structure. A campus outside the US earns a page when it shows a model no US campus does.
The bottlenecks that set the pace of the buildout, and how much of each layer is moving back to the US.
Running the machine once it is built: facility monitoring, compute operations, physical and cyber security, and governance, risk and compliance.
The public argument over the buildout: moratoriums, power bills, the bubble question, public opinion, the compute race and the pace of AI.
The scale ladder
Every data center, at any size, takes power in, sheds heat, couples its chips to some degree, and has an owner. The answers change by scale; the questions do not. Coherence is the column the network fabric sets, so it shades the rows.
| Scale | Examples | Power in | Heat out | Coherence | Owner or operator |
|---|---|---|---|---|---|
| Humanoid | Tesla Optimus, Unitree, UBTech Walker | Onboard battery | Air or small liquid loop | Independent inference | Fleet operator |
| Vehicle | Tesla Model Y and Cybercab, Waymo, Baidu Apollo Go | Traction battery | Vehicle coolant loop | Independent, aggregated across the fleet | Fleet operator or owner |
| Edge and micro | 5G MEC sites, retail edge, modular pods | Grid, battery backup | Air | Independent or loosely coupled | Telco, colocation, enterprise |
| Enterprise and colocation | Equinix, Digital Realty | Grid | Air, rear-door heat exchangers | Loosely coupled | Tenant in a colocation hall |
| Hyperscale cloud | AWS, Azure and Google regions | Grid plus power purchase agreements | Air and liquid hybrid | Loosely coupled | First-party |
| AI factory campus | SpaceX Colossus, Stargate Abilene | Behind-the-meter gas, nuclear | Direct-to-chip liquid | One coherent fabric, 100,000+ GPUs | Joint venture, merchant, first-party |
| Multi-campus training | Microsoft Fairwater, Atlanta to Wisconsin | Mixed | Direct-to-chip liquid | Coherent across sites over long-haul fiber | First-party |
| Orbital | SpaceX Starmind, built from AI1 satellites at the Gigasat factory; Starcloud, Google Suncatcher, China's Three-Body Computing Constellation | Solar | Liquid loops to deployable radiators | Clusters of about 10 satellites on laser links | Constellation operator |
Exemplars
Each exemplar campus anchors a way of powering, cooling, coupling or financing an AI data center that the others do not. The largest-campus lists stay as entry points: largest in the US and largest in the world.
Grid tie, behind-the-meter gas, nuclear, renewables with firming, fuel cells.
Heat outAir, direct-to-chip, immersion, free-air and district heat reuse, radiative in orbit.
CoherenceHow many accelerators train as one machine, and over what distance.
OwnerFirst-party, cloud joint venture, merchant AI factory, multi-party stacks.
Platform supplier rings
A rack-scale platform is the same machine wherever it is installed, so its supplier ring is drawn once: chips, server and rack integration, network and optics, cooling, and power. Each ring shows who makes every component of nine leading AI platforms, where it is made, and how much of it is made in the US.
Supply chain
Global facilities enter DatacentersX through the supply chain: where the transformers, optics, rack integration and turbines come from, and how much of that is coming home.
Transformers, turbines, HBM, optics, water and interconnection queues.
ReshoringWhich layers of the stack are moving to the US, and which are not.
RiskHardware provenance, firmware trust and single-source exposure.
LaborElectricians, technicians and engineers the buildout runs short of.
Across the network
DatacentersX starts at the server. The chip below it, the Texas campuses around it, the vehicles and robots that carry it, and the constellations above it each have a sister site.
Accelerators, HBM, packaging and the fabs that make them.
Texas campusesTexas AI campuses, their supplier rings and the power behind them.
Mobile computeVehicles, robots and fleets as compute nodes on wheels and legs.
In orbitOrbital compute and the space-industrial buildout.
The book
Daniel M. Sisson
The AI-Industrial Revolution From Austin to Alpha Centauri. The book follows the buildout from Texas fabs and AI factories to compute in orbit. DatacentersX keeps the data center layer behind it current: the stack, the power and cooling, the supplier rings and the public argument over the pace.