United States's data centres draw 176 TWh of electricity a year, which places it 1st of the 33 countries this atlas carries a figure for.
That is 3.9% of everything the country generates — well above the 1.2% median across those countries.
Set against 53.7 GW of installed capacity, that implies a 37% load factor.
14 metros are tracked here, totalling 15,659 MW, of which Northern Virginia alone is 4,496 MW — 29% of the national total.
6,567 MW of that is not yet operating: it is under construction or still an announcement.
Its strongest showing anywhere in the atlas is Silicon on the ownership basis, at #1 of 9.
56 suppliers are domiciled here, booking roughly $1,000bn between them; the largest is Nvidia.
Its single strongest grip on the buildout is Switch silicon, where firms domiciled here hold 85% of global supply in a sub-layer weighted 0.90 for how hard it is to substitute.
About $500bn of AI infrastructure investment has been announced for United States, though announced is not built — that figure is graded A.
National context
Electricity generation4,520 TWh A
Data-centre electricity176 TWh A
— as a share of generation3.9% B
Installed data-centre capacity53.7 GW A
Tracked metro capacity15,659 MW B
Announced investment$500bn A
Population342m A
GDP$30,767bn A
Grid friction
Queue-constrained
~2,061 GW of generation and storage actively seeking interconnection at end-2025; gas requests up 86% year on year.
~2,061 GW of generation and storage actively seeking interconnection at end-2025; gas requests up 86% year on year.
Where United States ranks
The same tables read three ways. A country can lead one basis and be invisible on another, and that gap is the whole point of the atlas.
| Tier | Physicalassets here | Ownershiprevenue booked | Controlcould stop you |
|---|---|---|---|
| Silicon | #5of 14 | #1of 9 | #2of 10 |
| Memory & Storage | #3of 7 | #2of 5 | #2of 5 |
| Networking | #2of 9 | #1of 4 | #1of 5 |
| Systems & Facility | #1of 54 | #2of 9 | #1of 9 |
| Power & Grid | #1of 77 | #1of 17 | #2of 12 |
Data centre metros (14)
| Northern Virginia — OperatingLargest market on earth and still the tightest: 4,496.5 MW of inventory at 0.2% vacancy in H1 2026, with just 10.8 MW actually available to lease and 2,420.2 MW under construction. It was 3,046.1 MW in Q1 2025. | 4,496 MW A |
| Richland Parish, LA — AnnouncedAnnounced multi-GW campus with dedicated new gas generation. Announced only — treat with caution. | 2,200 MW C |
| Atlanta — Operating1,465.2 MW at 1% vacancy. Grew from 1,279.4 MW with a 2,159.3 MW construction pipeline behind it — the fastest large-market ramp in North America. | 1,465 MW A |
| Dallas–Fort Worth — Operating1,249.4 MW, plus 716.7 MW under construction that is already 88% preleased. ERCOT interconnection is faster than most US markets, which is much of the attraction. | 1,249 MW A |
| Abilene, TX — Under constructionLarge announced AI campus with on-site generation. Named by CBRE alongside Reno as a tertiary market outpacing primaries. | 1,200 MW C |
| Chicago — Operating910.6 MW at 2.2% vacancy; overtook Phoenix as the fourth-largest North American market in Q1 2026. | 911 MW A |
| Memphis — Under constructionOn-site gas turbines used to bridge grid connection timelines. | 700 MW C |
| Columbus — Under constructionOhio has become a primary hyperscale corridor. | 620 MW C |
| Phoenix — Under construction617.0 MW at 1.7% vacancy (Q1 2025). Water and summer peak demand are the live constraints. | 617 MW A |
| Santa Clara — OperatingMunicipal utility capacity effectively capped growth. | 520 MW C |
| Cheyenne, WY — Announced | 480 MW C |
| Portland / Hillsboro — OperatingTranspacific cable landings and cheap hydro. CBRE flags its pipeline being outpaced by tertiary markets. | 450 MW C |
| Reno — Under constructionCBRE expects its pipeline to outpace primary markets like Hillsboro or Silicon Valley in 2026. | 400 MW C |
| Des Moines — Under constructionCBRE names it an emerging market with multiple large-scale campuses under construction. | 350 MW C |
Supply positions
Share of each global sub-layer held by firms domiciled here, and how hard that sub-layer would be to substitute. Those two multiplied are what the control basis ranks on.
Silicon
| Sub-layer | Share | Substitutability |
|---|---|---|
| Design & IP | 88% B | ×0.85 |
| Foundry | 5% A | ×0.98 |
| Advanced packaging | 8% C | ×0.92 |
| Semicap equipment | 38% B | ×0.96 |
| Test & assembly | 8% C | ×0.45 |
Strongest single grip: Design & IP — 88% × 0.85 substitutability weight.
Memory & Storage
| Sub-layer | Share | Substitutability |
|---|---|---|
| HBM | 25% B | ×0.95 |
| Conventional DRAM | 25% B | ×0.90 |
| NAND | 24% C | ×0.75 |
| Enterprise SSD | 30% C | ×0.55 |
| HDD | 80% C | ×0.60 |
| Controllers | 35% C | ×0.40 |
Strongest single grip: HDD — 80% × 0.60 substitutability weight.
Networking
| Sub-layer | Share | Substitutability |
|---|---|---|
| Optics & transceivers | 25% B | ×0.70 |
| Switch silicon | 85% C | ×0.90 |
| Switch systems | 70% C | ×0.60 |
| Fibre plant | 30% C | ×0.50 |
| Submarine cable | 30% C | ×0.85 |
Strongest single grip: Switch silicon — 85% × 0.90 substitutability weight.
Systems & Facility
| Sub-layer | Share | Substitutability |
|---|---|---|
| Rack integration & ODM | 14% B | ×0.80 |
| DC development | 55% C | ×0.50 |
| Colocation & neocloud | 62% B | ×0.45 |
| Cooling & thermal | 40% C | ×0.65 |
| Construction & EPC | 45% C | ×0.35 |
Strongest single grip: Colocation & neocloud — 62% × 0.45 substitutability weight.
Power & Grid
| Sub-layer | Share | Substitutability |
|---|---|---|
| Turbines & prime movers | 30% B | ×0.95 |
| Transformers & switchgear | 10% C | ×0.80 |
| On-site power & backup | 40% C | ×0.45 |
| Nuclear & SMR | 20% C | ×0.70 |
Strongest single grip: Turbines & prime movers — 30% × 0.95 substitutability weight.
Suppliers domiciled here (56)
| Nvidia — Accelerators, interconnectSilicon~86% of accelerator revenue in late 2025, easing as custom silicon scales. | $130bn |
| Nvidia — InfiniBand, Spectrum EthernetNetworkingNetworking is bundled with the accelerator platform. | $130bn |
| Broadcom — Custom AI ASICsSiliconAI ASIC backlog reported at $73bn, giving visibility into 2027. | $64bn |
| Broadcom — Switch silicon, opticsNetworkingTomahawk-class switching silicon underpins most large AI fabrics. | $64bn |
| Cisco — Switching, routingNetworkingOnly a fraction sits in the AI fabric perimeter. | $54bn |
| Intel — CPUs, foundrySiliconOnly US-domiciled leading-edge logic fab operator. | $53bn |
| Qualcomm — SoCs, edge inferenceSilicon | $39bn |
| GE Vernova — Gas turbines, grid equipmentPower & GridLargest single position in heavy-duty gas turbines. | $38bn |
| Micron — DRAM, HBM, NANDMemory & StorageOnly US-domiciled DRAM maker; expanding HBM aggressively. | $37bn |
| Applied Materials — Deposition, etch, metrologySilicon | $27bn |
| Dell Technologies — AI server systemsSystems & FacilityInfrastructure Solutions Group only. | $24bn |
| Turner Construction — Data-centre constructionSystems & Facility | $20bn |
| Lam Research — Etch, depositionSilicon | $18bn |
| Eaton — Electrical systems, switchgearPower & GridIrish-incorporated historically; classified here by principal operating domicile. | $17bn |
| AMD — Accelerators, CPUsSiliconCredible second source; multi-gigawatt agreements signed in 2025. | $16bn |
| AECOM — Engineering, constructionSystems & Facility | $16bn |
| Supermicro — AI systemsSystems & Facility | $15bn |
| Hewlett Packard Enterprise — AI server systemsSystems & Facility | $15bn |
| Corning — Optical fibreNetworking | $14bn |
| Google — TPUSiliconCaptive silicon; revenue is an imputed internal value. | $12bn |
| KLA — Process controlSilicon | $12bn |
| Jacobs — EngineeringSystems & Facility | $12bn |
| Vertiv — Thermal, powerSystems & FacilityAlso appears in the Power & Grid tier. | $10bn |
| Vertiv — UPS, power distribution, thermalPower & GridDirect exposure to data-centre electrical fit-out. | $10bn |
| Seagate — Nearline HDDMemory & StorageBulk training-data storage economics still rest on HDD. | $9bn |
| Western Digital — Nearline HDDMemory & Storage | $9bn |
| Equinix — ColocationSystems & FacilityPart of the ~30% of colocation revenue held by the top three. | $9bn |
| Amazon — Trainium, InferentiaSiliconCaptive silicon. | $8bn |
| SanDisk — NAND, SSDMemory & Storage | $8bn |
| Arista Networks — Data-centre switchingNetworkingLargest merchant AI-fabric switch vendor. | $8bn |
| GlobalFoundries — Mature-node foundrySilicon3.3% of foundry revenue. | $7bn |
| Cummins — Standby generationPower & GridPower Systems segment only. | $7bn |
| Caterpillar — Standby and prime generationPower & GridEnergy & Transportation segment. | $7bn |
| Baker Hughes — Industrial gas turbinesPower & GridIndustrial & Energy Technology segment. | $7bn |
| Marvell — Custom silicon, DPUsSilicon | $6bn |
| Amkor — Packaging and testSilicon | $6bn |
| Marvell — DSPs, optical siliconNetworking | $6bn |
| Coherent — Transceivers, lasersNetworking~16% of transceiver shipments; leading Western supplier. | $6bn |
| Digital Realty — ColocationSystems & Facility | $6bn |
| Westinghouse — Reactors, servicesPower & Grid | $6bn |
| CoreWeave — GPU cloudSystems & FacilityLargest of the neoclouds; heavily debt-financed against GPU collateral. | $5bn |
| Ciena — Optical transportNetworking | $4bn |
| Generac — Generation setsPower & Grid | $4bn |
| Vantage Data Centers — Hyperscale developmentSystems & Facility | $3bn |
| QTS — Hyperscale developmentSystems & FacilityBlackstone-owned; financials not separately disclosed. | $3bn |
| Lumentum — Lasers, transceiversNetworking | $2bn |
| SubCom — Submarine cable systemsNetworkingOne of four suppliers that build effectively all intercontinental capacity. | $2bn |
| STACK Infrastructure — Hyperscale developmentSystems & Facility | $2bn |
| Aligned Data Centers — Hyperscale developmentSystems & Facility | $2bn |
| CyrusOne — Hyperscale developmentSystems & Facility | $2bn |
| Crusoe Energy — GPU cloud, on-site powerSystems & FacilityBuilds generation alongside compute. | $2bn |
| Boyd Corporation — Liquid coolingSystems & Facility | $2bn |
| Credo — SerDes, active cablesNetworking | $1bn |
| Astera Labs — Connectivity siliconNetworking | $1bn |
| Lambda — GPU cloudSystems & Facility | $1bn |
| Applied Digital — Hyperscale developmentSystems & Facility | $1bn |
Revenue is the group or segment figure closest to each tier's perimeter. Segment boundaries differ between issuers, so treat the split as indicative.