Meta in Talks to Lease Computing Power to Anthropic in Potential $10 Billion Deal - The New York Times
Meta in Talks to Lease Computing Power to Anthropic in Potential $10 Billion Deal The New York Times
Publisher pages open in a focused popup so you keep Data Center Watch context. We never copy article bodies — only the headline link.
WATCH
power · compute · buying · dependenciesTrack facilities, current load, future requests, power-capacity gaps, component BOMs, buying intent, producer constraints, and the primary → secondary → tertiary dependencies that turn a greenfield site into an operating compute campus. Every number links to evidence.
Every headline datapoint is a card. Click any card to jump to the underlying detail.
Worldwide facility layer assembled from open sources plus the full ATLAS directory (unrestricted use granted to Data Center Watch by the ATLAS IP team, 2026-07), OpenStreetMap (ODbL), PeeringDB interconnection campuses, FracTracker US operating/proposed inventory, and Epoch AI large AI campuses (CC BY 4.0). ATLAS coordinates are city-level centroids (labelled as such), not rooftop GPS. DataCenterMap and Baxtel remain catalogued only (commercial no-scrape). MW and H100-eq appear only when a source states them. Gaps stay empty. The globe plots every geocoded row; fiber arcs remain illustrative corridors.
Facility count by country label · incomplete where source omits country
Provenance mix · dual-count possible pre-dedup across sources
| Facility | Operator | Country | City / region | MW | H100-eq | Status | Source |
|---|---|---|---|---|---|---|---|
| Project Jade | Crusoe | United States | Cheyenne | 10.0 GW | — | suspended | FracTracker Alliance National Da |
| SoftBank Data Center/PORTS Technology Campus | Softbank/Open AI (unconfirmed) | United States | Piketon | 9,200 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Stratos Project, eastern parcels | — | United States | Snowville | 9,000 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| New Era Energy & Digital Data Center | — | United States | Lea County | 7,000 MW | — | proposed | FracTracker Alliance National Da |
| GW Ranch (Pacifico Energy) | — | United States | Pecos County | 5,000 MW | — | proposed | FracTracker Alliance National Da |
| Homer City Energy Campus | Homer City Redevelopment (with Kiewit as contractor/partner) | United States | Homer City | 4,500 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Shippingport Bruce Mansfield Data Center/Project Phoenix | Aligned Data Centers | United States | Shippingport | 3,600 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Pittsylvania County Data Center | Balico LLC | United States | Chatham | 3,500 MW | — | cancelled | FracTracker Alliance National Da |
| MSB Global Services Matrix Data Center | — | United States | Sulphur Springs | 3,000 MW | — | proposed | FracTracker Alliance National Da |
| Stak Energy Data Center | Stak | United States | North Slope Borough | 3,000 MW | — | proposed | FracTracker Alliance National Da |
| Vermaland Data Center | Vermaland LLC | United States | Red Rock | 3,000 MW | — | proposed | FracTracker Alliance National Da |
| Hobart Data Center | Amazon | United States | Hobart | 2,400 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Tract Technology Campus | VALCO HANOVER COUNTY | United States | Ashland | 2,400 MW | — | proposed | FracTracker Alliance National Da |
| Adams Fork Harless Data Center | TransGas Development, LLC | United States | Holden | 2,398 MW | — | proposed | FracTracker Alliance National Da |
| Adams Fork Wharncliffe Data Center | TransGas Development, LLC | United States | Wharncliffe | 2,398 MW | — | proposed | FracTracker Alliance National Da |
| New Carlisle Data Center | Amazon | United States | New Carlisle | 2,200 MW | — | operating | FracTracker Alliance National Da |
| Eneus Energy | Eneus Energy | United States | Harlingen | 2,000 MW | — | proposed | FracTracker Alliance National Da |
| Green Data Center | — | United States | Socorro | 2,000 MW | — | cancelled | FracTracker Alliance National Da |
| PowerBridge/LandBridge | — | United States | Between Odessa and Pecos | 2,000 MW | — | proposed | FracTracker Alliance National Da |
| Project Horizon | Poolside/CoreWeave | United States | Fort Stockton | 2,000 MW | — | proposed | FracTracker Alliance National Da |
| Sterling Data Center | Granite Renewables | United States | Sterling | 2,000 MW | — | proposed | FracTracker Alliance National Da |
| Sweetwater 1 and 2 | IREN | United States | Sweetwater | 2,000 MW | — | proposed | FracTracker Alliance National Da |
| Project Bunkhouse | Gaines Family Land LLC | United States | Cartersville | 1,830 MW | — | proposed | FracTracker Alliance National Da |
| AREP PowerHouse Grand Prairie | PowerHouse Data Centers | United States | Midlothian | 1,800 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| PowerHouse Data Centers | — | United States | Jackson Twp/ Joliet | 1,800 MW | — | proposed | FracTracker Alliance National Da |
| PowerHouse Hillwood Holding | PowerHouse Hillwood Holding | United States | Joliet | 1,800 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Ridgeline Microgrid | Fundamental Data LLC | United States | Davis | 1,656 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Tract Data Center | — | United States | Silver Springs | 1,600 MW | — | proposed | FracTracker Alliance National Da |
| Antelope Data Center | — | United States | Cedar City | 1,500 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Arizona Land Consulting Co. | — | United States | Tonopah | 1,500 MW | — | proposed | FracTracker Alliance National Da |
| Blakely Data Center | — | United States | Blakely | 1,500 MW | — | cancelled | FracTracker Alliance National Da |
| Crusoe/ Blue Energy Data Center | Crusoe | United States | Victoria | 1,500 MW | — | proposed | FracTracker Alliance National Da |
| NE Edge Groton Data Center | — | United States | Waterford | 1,500 MW | — | cancelled | FracTracker Alliance National Da |
| Open AI/ SoftBank "Freebird" | — | United States | Burlington | 1,500 MW | — | proposed | FracTracker Alliance National Da |
| Project Gold Data Center | — | United States | Clifton Twship | 1,500 MW | — | proposed | FracTracker Alliance National Da |
| Stargate Data Center | Bristolville 25 Developer LLC | United States | Warren | 1,500 MW | — | proposed | FracTracker Alliance National Da |
| Frontier Vantage Data Center | — | United States | Abilene | 1,400 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Related Digital/The Barn | Related Digital/Open AI/Oracle | United States | Saline | 1,400 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| xAI Colossus 2 | xAI | United States | Memphis | 1,400 MW | — | expanding | FracTracker Alliance National Da |
| Project Arrowhead | — | United States | Ocilla | 1,250 MW | — | proposed | FracTracker Alliance National Da |
| Wyoming Hyperscale Whitebox | Prometheus Hyperscale | United States | Evanston | 1,250 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Zenith Volts Data Center | — | United States | Roswell | 1,240 MW | — | proposed | FracTracker Alliance National Da |
| Project Aquilla: Abbott Holding Data Center | RCM Hill | United States | Hill County | 1,235 MW | — | proposed | FracTracker Alliance National Da |
| Amazon Data Center | Amazon | United States | Heppner | 1,200 MW | — | proposed | FracTracker Alliance National Da |
| CloudBurst Data Center | — | United States | New Braunfels | 1,200 MW | — | proposed | FracTracker Alliance National Da |
| Kentucky Industrial Alliance LLC Data Center | Kentucky Industrial Alliance LLC | United States | Cave City | 1,200 MW | — | suspended | FracTracker Alliance National Da |
| Oasis Data Center | — | United States | Fredericksburg | 1,200 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Open AI/ SoftBank Stargate Data Center (Lancium Clean Campus) | Microsoft (previously Open AI/Oracle) | United States | Abilene | 1,200 MW | — | operating | FracTracker Alliance National Da |
| Project Marvel | Logistic Land Investments LLC | United States | Bessemer | 1,200 MW | — | proposed | FracTracker Alliance National Da |
| Project Spade | Related Digital/Google | United States | New Florence | 1,200 MW | — | proposed | FracTracker Alliance National Da |
| South Valley Technology Park Data Center | Tract | United States | Sparks | 1,200 MW | — | proposed | FracTracker Alliance National Da |
| Starwood Digital Ventures | — | United States | New Castle | 1,200 MW | — | suspended | FracTracker Alliance National Da |
| Tract Data Center | Fleet | United States | Silver Springs | 1,200 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Amazon Jeffersonville Data Center | Amazon | United States | Jeffersonville | 1,184 MW | — | proposed | FracTracker Alliance National Da |
| EdgeCore | EdgeCore | United States | Kents Store | 1,100 MW | — | suspended | FracTracker Alliance National Da |
| Stack Data Center | — | United States | Stafford | 1,100 MW | — | proposed | FracTracker Alliance National Da |
| Goodnight Campus | Crusoe | United States | Claude | 1,030 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Wilsonville Data Center | Northpoint Development | United States | Wilsonville | 1,008 MW | — | proposed | FracTracker Alliance National Da |
| Amazon Ridgeland | Amazon | United States | Ridgeland | 1,000 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Eagle Rock Data Center | Eagle Rock | United States | Palmer | 1,000 MW | — | proposed | FracTracker Alliance National Da |
| Flint Hills Digital Campus | — | United States | Emporia | 1,000 MW | — | proposed | FracTracker Alliance National Da |
| Lacy Lakeview Data Center | InfraKey Capital | United States | Lacy Lakeview | 1,000 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Lower Peninsula Data Center | — | United States | Lower Peninsula | 1,000 MW | — | proposed | FracTracker Alliance National Da |
| Meta Project Seafox | Wurldwide LLC/ Meta | United States | El Paso | 1,000 MW | — | proposed | FracTracker Alliance National Da |
| Nixxy-Tachyon 9 Data Center | — | United States | Epping | 1,000 MW | — | proposed | FracTracker Alliance National Da |
| Panattoni Project Cannoli | United States | Van Buren Township | 1,000 MW | — | proposed | FracTracker Alliance National Da | |
| Project Domino | Meta | United States | Lebanon | 1,000 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Quantica LLC Big Sky Campus | Meta, Google, Amazon | United States | Broadview | 1,000 MW | — | proposed | FracTracker Alliance National Da |
| Rocklocker LLC | — | United States | Kalkaska Township | 1,000 MW | — | cancelled | FracTracker Alliance National Da |
| Sentinel Grove Technology Park/ Project Jarvis | — | United States | Fort Pierce | 1,000 MW | — | suspended | FracTracker Alliance National Da |
| Stafford Technology Campus | — | United States | Stafford | 1,000 MW | — | proposed | FracTracker Alliance National Da |
| Tract Morris Data Center | Tract Data Centers | United States | Morris | 1,000 MW | — | proposed | FracTracker Alliance National Da |
| Treetop Development Project | onovan Drive Holdings LLC | United States | Hopewell Junction | 1,000 MW | — | suspended | FracTracker Alliance National Da |
| Walterboro Data Center | Eagle Rock Partners | United States | Walterboro | 1,000 MW | — | proposed | FracTracker Alliance National Da |
| Salem Township Amazon Data Center | Amazon | United States | Wapwallopen | 960 MW | — | proposed | FracTracker Alliance National Da |
| Colossus 2 | SpaceXAI | United States | — | 946 MW | 1,390k | ai_campus | Epoch AI Data Centers (CC BY 4.0 |
| Project Sail | Atlas Development | United States | Newnan | 936 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Anthropic-Amazon New Carlisle | Amazon | United States | — | 910 MW | 687k | ai_campus | Epoch AI Data Centers (CC BY 4.0 |
| Robena Mine Data Center/ Project Hummingbird | Essential Utilities/ International Electric Power of Pittsburgh | United States | Monongahela Township | 910 MW | — | approved/permitted/under con | FracTracker Alliance National Da |
| Bolingbroke Technology Campus | — | United States | Macon | 900 MW | — | cancelled | FracTracker Alliance National Da |
Coverage model: what we already observe, what should be ingested next, and where a power story becomes a land, water, equipment, labor, capital, or community story. Green dots are tracked; violet dots are catalogued for ingestion.
site control · zoning · water rights
MW request · ramp · utilization
queue · study · milestone
service date · tariff · curtailment
gas · nuclear · renewables · BESS
transformers · lines · upgrades
PUE · WUE · source · discharge
backhaul · latency · route diversity
air · noise · tax · local votes
GPU · switchgear · generators
design · crews · commissioning
capex · incentives · revenue
uptime · emissions · waste heat
The ticker is fed by an automated RSS crawl across grid, energy, data-center, HPC, and large-load sources. Headlines link to the original publisher; Data Center Watch does not copy article bodies.
Meta in Talks to Lease Computing Power to Anthropic in Potential $10 Billion Deal The New York Times
Huge AI data center rejected in Florida. Why are they so controversial? Yahoo
Brookhaven Town OKs 18-month AI data center moratorium Newsday
Anthropic in early talks with Meta to acquire compute power CNBC
'Prefer an outright ban': Small Texas town weighs data center restrictions amid industry growth Austin American-Statesman
SpaceXAI Provides Near IMMEDIATE Gigawatt Scale AI Data Center Capacity NextBigFuture.com
Earthjustice set to take the issue to court if applications keep getting granted
AI data center growth could force US utilities to rethink generation plans, BofA says Utility Dive
Duos Edge AI to provide 10MW of capacity for five years to “investment-grade hyperscaler” Data Center Dynamics
Pennsylvania families sold 1,700 acres for $586 million as AI data center land rush mints new millionaires Yahoo! Finance Canada
Kentucky powered America once. It can help power the AI era too | Opinion Lexington Herald Leader
Pennsylvania pig farmers sell 89-acre farm for $22 million as AI data center boom mints dozens of new millionaires moneywise.com
3 Utility Stocks Built for the Coming AI Power Crunch The Motley Fool
Crusoe, Lancium announce partnership for 1-gigawatt AI data center campus in Childress MyHighPlains.com
Deal could be worth up to $1bn
3 Utility Stocks Built for the Coming AI Power Crunch The Globe and Mail
PJM capacity auction results compound ‘alarm bells’: FERC Chairman Swett Utility Dive
Investor has withdrawn from the project in Babenhausen
Project Tango Denial Challenges Wall Street's AI Power Bet Data Center Knowledge
AI-data center protests set for Saturday in Texas Audacy
The company sees long-term revenue opportunities in connecting data centers with fiber
Jefferies says Palm Beach County's rejection of a proposed 600 MW AI campus shows local permitting could determine whether AI load materializes.
Data Center Troubles Stoke Industry’s Fear of Coming Distress Bloomberg.com
AI data center 'grid'lock: Former Southern Company CEO on paying for data center power surge CNBC
Few details known about facility being planned for Fairfield
Sadeh was CEO from 2009
Taxpayer group warns New York’s AI data center moratorium could cost jobs, investment Baltimore Sun
The AI Data Center Boom Is Bigger Than One Stock—These ETFs Spread the Bet MarketBeat
Data Center Connectors Market Size, Forecasts Report 2035 Global Market Insights Inc.
Data Center Ethernet Market Size, Forecasts Report 2026-2035 Global Market Insights Inc.
EuroHPC JU puts out call for tender for Italian industrial supercomputer dubbed Innovate Data Center Dynamics
Tender has an estimated value of €11.6m
AI data centers must produce as much power as they use, Australia PM says — new national AI framework will also ensure water efficiency and protect intellectual property rights Tom's Hardware
What Intel Stock Was Signaling About The AI Data Center's Real Engine Trefis
Sharon AI signs $1.32bn cloud computing service agreement with unnamed AI lab Data Center Dynamics
Why New York’s Data Center Ban Could Rewrite the Future of the AI Revolution 24/7 Wall St.
Population growth in Haines limiting available water for Cielo
Targets under-construction data center near Sloterdijk Station in Amsterdam
Targets under-construction data center near Sloterdijk Station in Amsterdam
Black and white Louisiana churches split over $250B Meta data center thelensnola.org
Partnership to reduce data center deployment times… and dust
Why Modularization is Becoming the Blueprint for Modern Data Centers Data Center Frontier
As gas plants rise to power AI, renewable energy allies are fighting for cleaner alternatives constructconnect.com
Datavault AI to tokenize Available Infrastructure edge data center network Investing.com
New York’s Data Center Freeze Adds a New Risk for AI Investors: State Regulation Morningstar
The Social Skinny: Apple sues OpenAI and New York imposes AI data center moratorium Campaign US
Dominion Energy to pilot LōD's grid-responsive AI inference technology in Ashland, Virginia Data Center Dynamics
As part of Dominion's Accelerate program
Data Center CPU Market Size, Global Report 2026-2035 Global Market Insights Inc.
Quinnipiac University Poll Finds 74% Of PA Voters Would Oppose Building An A.I. Data Center In Their Community; 76% Are Following Data Center News Closely PA Environment Digest Blog
Bucks County's first data center will open soon. What you need to know PhillyBurbs
Would be located south of Lake Graham
Would be located south of Lake Graham
Right-Sizing Data Centers: How to Match Footprint to Real-World Demand Data Center Knowledge
Sizing Data Centers: Match Footprint to Workload Data Center Knowledge
Data centers span from single-rack edge sites to multi-building campuses. Here’s how to choose the right scale for your workload, power, and growth profile.
Plan at very early stages
After Months of Debate, Virginia Fails to Pass Data Center Clean Energy Requirements Inside Climate News
As data center backlash grows in Florida, here’s why Miami projects are different Miami Herald
Intel's AI-Driven Data Center Growth Set To Power Second Quarter Earnings Seeking Alpha
This Utility Stock Could Benefit From the AI Power Bottleneck The Motley Fool
Kevin O’Leary sued for defamation after he claimed China fueled opposition to his Utah data center Fortune
Is SLB (SLB) Cheap Following Its AI Data Center Alliance? simplywall.st
North Sea gas to power offshore AI data centre AOL.com
Deep Dive: Understanding Data Center Perils Carrier Management
Pocatello City Council Upholds Hearing Examiner's Decision, says "no" to Proposed AI Data Center Local News 8
Pocatello City Council upholds decision to deny AI data center East Idaho News
Microsoft partners with 3M to enhance AI data center infrastructure Indiatimes
Brookhaven Votes On 18 Month Ai Data Center Moratorium News12 | Long Island
AMD (AMD) Wins Major AI GPU Deals As Data Center Becomes Over Half Of Sales simplywall.st
New York State just hit pause on the AI data center boom Network World
AMD Stock Search Is Rising With 3 AI Data Center Names Worth Watching simplywall.st
When can a power company take your land for a data center? The Washington Post
Palm Beach County commissioners vote against controversial Project Tango data center proposal WPBF
When can a power company take your land for a data center? The Washington Post
City to appeal BLM’s data center plan Boulder City Review
New Mexico regulators reject Oracle's pipeline application again, threatening AI data center plans Crypto Briefing
Inside look at AI data center in Lowell NBC Boston
Actual system demand — not interconnection requests. Crosshair cursors, ±1σ band, mean reference, scientific axes. Gaps stay empty — nothing is interpolated.
μ / σ / range from observed hours only · brush + scroll to zoom · EIA-930
Daily peak by balancing authority · fixed categorical colors · nulls not filled
Step series · queue jumps only at published decks · no interpolation between months
Approved vs observed · realization % = observed ÷ approved when both exist
| Region | Total queue | As of |
|---|---|---|
| ERCOT | 438.0 GW | 2026-03-31 |
Nameplate capacity by fuel — what is built, not what is requested. Complements the queue charts above.
Nightingale rose · nameplate MW share · EIA-860/860M
Lollipop fleet map · operating nameplate by region
| Region | MW | As of |
|---|---|---|
| ERCOT | 174.9 GW | 2026-04 |
| ERCOT fuel | MW | Share |
|---|---|---|
| Natural Gas | 67.2 GW | 38% |
| Wind | 40.0 GW | 23% |
| Solar | 31.6 GW | 18% |
| Battery Storage | 15.9 GW | 9% |
| Coal | 13.4 GW | 8% |
| Nuclear | 5,139 MW | 3% |
| Oil/Diesel | 798 MW | 0% |
| Hydro | 549 MW | 0% |
| Other/Biomass | 374 MW | 0% |
Lower is better · reference line at industry ~1.5 PUE · source-linked table below
| Entity | $/W | As of | Source |
|---|---|---|---|
| 1x1 combined-cycle gas turbine plant (generation side, not the data center building) | $0.92/W | 2023 | GridLab / Energy Futures Group / Halcyon, citing EIA (Sargent & Lundy cost study) |
| Traditional prime markets, full build (shell + M&E + cooling + fit-out) | $11.00/W | 2025-2026 | CBRE Investment Management, citing Turner & Townsend Data Center Cost Index |
| Global average (shell-and-core construction) | $11.30/W | 2026 (forecast) | JLL |
| Entity | PUE | As of | Source |
|---|---|---|---|
| Google (global fleet-wide) | 1.09 | 2025 | Google (datacenters.google) |
| AWS (global fleet-wide) | 1.14 | 2025 | Amazon.com Inc. (Amazon Sustainability) |
| US national average (all data center types) | 1.40 | 2023 | Lawrence Berkeley National Laboratory — 2024 United States Data Center Energy Usage Report (LBNL-2001637) |
| Global industry average (Uptime Institute Global Data Center Survey 2025, n=681) | 1.54 | 2025 | Uptime Institute (Uptime Intelligence) |
| Entity | Share | As of | Source |
|---|---|---|---|
| United States — all data centers (649 TWh reference case) | 11.8% of total U.S. electricity use | 2030 (reference forecast) | Lawrence Berkeley National Laboratory — United States Data Center Energy Usage Report: 2025 Update |
| Ireland — all grid-connected data centers (7.663 TWh metered) | 23.0% of total metered electricity consumption | 2025 | Central Statistics Office Ireland — Data Centres Metered Electricity Consumption 2025 |
| United States — all data centers (176 TWh estimated consumption) | 4.4% of total U.S. electricity use | 2023 | Lawrence Berkeley National Laboratory — 2024 United States Data Center Energy Usage Report |
| Metric | Entity | Value | As of | Source |
|---|---|---|---|---|
| Capacity auction clearing price | PJM RTO (2027/2028 Base Residual Auction; up from $329.17 in 2026/27) | 333.44 USD/MW-day | 2027/2028 delivery year (BRA report published 2025) | |
| Contracted large-load capacity | Ireland — data centers + other new-technology loads, transmission contracts | 2,000 MW | 2025-08 | |
| Data-center capacity | New Zealand — data-center capacity (median scenario) | 400 MW | 2035 (forecast) | |
| Data-center energy share | Ireland — data centers + new-technology loads (median scenario) | 30.0% of electrical energy requirements | 2032 (forecast) | |
| Projected data-center load growth | PJM — projected data-center load growth (up to) | 30.0 GW | 2030 (forecast) | |
| Observed data-center peak demand | Ireland — data centers (observed peak demand to date) | 800 MW | 2025-08 | |
| DC projects blocked or delayed (value) | United States (at least 75 projects blocked or delayed by local opposition) | 130.00 USD billions (approx.) | Q1 2026 (Jan-Mar) | |
| Direct liquid-cooling adoption | Data center operators globally (Uptime survey, n=512 for cooling-type question) | 22.00 percent of organizations using direct liquid cooling | 2025 survey (fielded Apr-Jun 2025) | |
| Electrical-equipment unfilled orders | United States (NAICS 335 electrical equipment, seasonally adjusted; +5.9% YoY NSA) | 54,578.00 USD millions | May 2026 (advance durable goods report) | |
| EU data-center electricity use | European Union data centres (all, EED reporting context) | 70.00 TWh/year | CY2024 (EC news item, 2025-11-17) | |
| Fleet water usage effectiveness (WUE) | Microsoft (owned datacenter fleet, global; PUE 1.17 same period) | 0.27 L/kWh | FY2025 (Jul 2024 - Jun 2025) | |
| Fleet water usage effectiveness (WUE) | Meta (owned data centers, global; five-year series 0.30 down to 0.19) | 0.19 L/kWh | CY2024 | |
| Gas-turbine backlog + slot reservations | GE Vernova (gas power equipment backlog + slot reservations; guiding to 110+ GW by end-2026) | 100.00 GW | 2026-03-31 (Q1 2026, reported 2026-04-22) | |
| Global DC electricity — 2030 projection | Global data centres (IEA base case; ~415 TWh in 2024, US 45% share) | 945.00 TWh/year projected by 2030 | 2025-04-10 (IEA Energy and AI report) | |
| Global semiconductor sales | Worldwide semiconductor sales (WSTS 3-month moving average; +104.1% YoY, Americas +132.2%) | 120.60 USD billions per month | May 2026 (SIA release, 2026-07-06) | |
| US interconnection queues — active capacity | United States (all 7 ISOs/RTOs + ~50 non-ISO utilities, active queue) | 2,060.00 GW | 2025-12-31 (Queued Up 2026 edition, published June 2026) | |
| Micron quarterly revenue (DRAM/HBM proxy) | Micron Technology (vs $9.30B a year earlier; FQ4 guidance ~$50B) | 41.46 USD billions per quarter | Fiscal Q3 2026 (quarter ended 2026-05-28) | |
| North-America top-4 markets net absorption | North America top-4 data center markets (CBRE; +34% YoY, Northern Virginia vacancy 0.3%) | 2,236.20 MW net absorption per quarter | Q1 2026 (Global Data Center Trends 2026, published 2026-06-17) | |
| Nuclear PPA capacity contracted to DCs | Talen Energy-Amazon (Susquehanna nuclear plant, expanded PPA through 2042) | 1,920 MW | 2025-06-11 (8-K press release) | |
| Power-transformer supply deficit | United States power transformer market (distribution transformers 10%; imports cover 80% of supply) | 30.00 percent of 2025 demand unmet | 2025 (Wood Mackenzie release, 2025-08-14) | |
| PPI — power/distribution transformer mfg | United States (NAICS 335311 transformer manufacturing, BLS series PCU335311335311) | 457.06 index points, NSA (preliminary) | May 2026 | |
| US private DC construction spend (SAAR) | United States (private data center construction, Census VIP; +23% vs May 2025) | 59,307.00 USD million, seasonally adjusted annual rate | May 2026 (preliminary, released 2026-07-01) | |
| DC-REIT quarterly bookings (annualized rent) | Digital Realty Trust (100% share; backlog $1.8B, occupancy 90.1%) | 707.00 USD millions annualized GAAP base rent signed per quarter | Q1 2026 (supplement filed 2026-04-23) | |
| Quarterly capex (incl. finance leases) | Meta (capex incl. finance-lease principal, global; FY2026 guidance raised to $125-145B) | 19.84 USD billions per quarter | Q1 2026 (quarter ended 2026-03-31, reported 2026-04-29) | |
| Quarterly capex (PP&E) | Alphabet (purchases of property and equipment, global; TTM $109.9B) | 35.70 USD billions per quarter | Q1 2026 (quarter ended 2026-03-31, reported 2026-04-29) | |
| Quarterly capex (PP&E) | Amazon (purchases of property and equipment, global; TTM $151.0B) | 44.20 USD billions per quarter | Q1 2026 (quarter ended 2026-03-31, reported 2026-04-29) | |
| Quarterly capex (PP&E) | Microsoft (additions to property and equipment, global; commercial RPO $627B) | 30.90 USD billions per quarter | FY26 Q3 (quarter ended 2026-03-31, reported 2026-04-29) | |
| TSMC monthly revenue (AI-demand proxy) | TSMC (company-wide, unaudited; +30.1% YoY) | 416,975.00 NT$ millions per month | May 2026 | |
| US data-center electricity use | United States data centers (4.4% of US electricity; projected 325-580 TWh by 2028) | 176.00 TWh/year | CY2023 (LBNL report published Dec 2024) | |
| Virginia DC air permits issued | Virginia (statewide, all data centers; backup-generator air permit registry rows) | 198.00 issued air permits | 2026-07-06 |
Translate a compute-power target into a first-pass equipment envelope using primary manufacturer specifications. Reported facts and derived arithmetic are labelled separately; unknowns stay unknown.
Screening formulae · units = ceil(target MW × 1,000 ÷ unit kW) · facility MW = target × PUE · utility MVA = facility MW ÷ PF · loop gpm = target MW × 3,412,142 ÷ (500 × ΔT). N+1 is a user-selected 1.25× allowance, not an engineered topology. Water consumption is never inferred from loop flow.
The calculator starts the brief; these six packages keep scope gaps from becoming change orders.
Live buying-intent feeds · SAM.gov opportunities ↗ · EU TED search API ↗ · the source atlas adds certified model universes for servers, network, storage and UPS.
Cooling · Direct liquid cooling at the rack/compute-tray level
Memory / DIMMs · Grace uses soldered LPDDR5; 1 TB aggregate CPU main memory per compute tray, not field-replaceable DIMMs.
Network · Derived rack total: 72 x ConnectX-8 HCAs and 18 x BlueField-3 DPUs from the per-tray specification.
142 kW is a published maximum for the full rack, not a measured workload average. Memory uses decimal GB. Storage count excludes management nodes and external storage. Installed PSU nameplate capacity must not be treated as rack draw.
NVIDIA Enterprise Reference Architecture: GB300 NVL72 Components ↗Cooling · Direct liquid cooling
Memory / DIMMs · 1 TB aggregate LPDDR5 CPU main memory across two Grace CPUs; no replaceable DIMM population is specified.
Network · 4 x ConnectX-8 HCAs and 1 x BlueField-3 DPU per compute tray.
A tray is a component of the NVL72 system rather than a standalone server bill of materials. The five-device count is 1 M.2 OS device plus 4 E1.S NVMe cache devices and does not express capacity.
NVIDIA Enterprise Reference Architecture: GB300 NVL72 Components ↗Cooling · Not prescribed; depends on the selected OEM control node
Memory / DIMMs · 8 x 64 GB DDR5 DIMMs = 512 GB.
Network · 4 x ConnectX-7 200 Gb/s adapters.
The architecture recommends 12 management/control nodes, but this profile describes one node. CPU wording follows the reference configuration; it is not an endorsement of a particular CPU vendor or SKU.
NVIDIA Enterprise Reference Architecture: GB300 NVL72 Components ↗Cooling · Platform implementation may use passive or active cooling
Memory / DIMMs · 288 GB HBM3E accelerator memory; no host DIMM requirement is defined by the accelerator page.
Network · PCIe 5.0 x16 host interface; fabric and NIC design are platform-specific.
Do not use 1.4 kW as server or rack input power. AMD also reports 8 TB/s peak memory bandwidth and 185 billion transistors; neither is a facility-load metric.
AMD Instinct MI355X accelerator specifications ↗Cooling · OEM/platform-specific; both passive and active accelerator implementations are supported
Memory / DIMMs · 2,304 GB aggregate HBM3E is derived; host CPU memory and DIMM population are unspecified.
Network · Host, scale-up fabric, NIC, DPU, and switch requirements are not included.
This is transparent screening arithmetic, not an AMD server BOM. It omits CPUs, DIMMs, storage, networking, cooling auxiliaries, PSUs, conversion losses, and redundancy; it cannot be compared directly with the complete 142 kW NVL72 rack rating.
AMD Instinct MI355X accelerator specifications ↗Cooling · Hybrid cooling: GPUs and CPUs are direct liquid-cooled via cold plates fed by rack manifolds; networking and storage devices are air-cooled (per the same user guide).
Memory / DIMMs · Not stated in the reviewed NVIDIA GB200 documentation at rack-spec level (no CPU DIMM/memory-channel table published); accelerator memory is separately stated as 'Up to 13.4 TB HBM3e | 576 TB/s' on NVIDIA's DGX GB200 product page.
Network · Per compute tray (18 trays/rack): 4x NVIDIA ConnectX-7 single-port 400G OSFP NIC for cross-rack compute networking; 2x NVIDIA BlueField-3 DPU, dual-port 400G InfiniBand/Ethernet, for storage/management. Out-of-band: 1x 1GbE RJ45 from the compute tray BMC plus 2x 1GbE RJ45 from the BlueField-3 BMC interface. Rack also includes 9 NVLink switch trays providing the 72-GPU NVLink domain.
Covers only the DGX GB200 NVL72 rack itself (compute trays, NVLink switch trays, power shelves). Does not include external InfiniBand/Ethernet leaf-spine switches, external high-performance storage appliances, management nodes, or facility cooling (CDU/pump) overhead -- these are shown as separate racks in NVIDIA's own SuperPOD Scalable Unit diagrams and are folded into the separate, not-directly-comparable 1.2MW per-SU TDP figure.
NVIDIA DGX GB Rack Scale Systems User Guide -- Hardware (power, storage, cooling, networking); NVIDIA DGX GB200 product page (GPU memory figure); NVIDIA DGX SuperPOD Reference Architecture Featuring NVIDIA DGX GB200, RA-11338-001 (compute tray CPU/GPU composition) ↗Cooling · Air-cooled, 10 RU rack-mount chassis (per the datasheet's Rack Units spec).
Memory / DIMMs · 2TB system memory, configurable to 4TB, per the official datasheet ('System Memory: 2TB, configurable to 4TB'); specific DIMM count/type is not published in the datasheet.
Network · 4x OSFP ports serving 8x single-port NVIDIA ConnectX-7 VPI (up to 400Gb/s NVIDIA InfiniBand/Ethernet each); 2x dual-port QSFP112 NVIDIA BlueField-3 DPU (up to 400Gb/s InfiniBand/Ethernet); management network: 10Gb/s onboard NIC with RJ45, 100Gb/s dual-port Ethernet NIC, and host BMC with RJ45.
Power figure covers the DGX B200 server chassis only -- it does not include top-of-rack network switches, external storage, or facility-level PUE overhead. NVIDIA's DGX SuperPOD B200 guidance elsewhere discusses deploying 2 systems per 42U rack, which would roughly double this per-system figure, but no NVIDIA-published dual-system rack total was located for this candidate.
NVIDIA DGX B200 Datasheet ↗Cooling · Not specified for the baseboard alone. The datasheet states the baseboard 'can be paired with advanced air or liquid cooling solutions and additional server components' -- cooling design is left to the OEM system integrator.
Memory / DIMMs · Not applicable / not stated -- the baseboard itself carries no system DIMMs; host CPU memory is provided by the OEM server board, which is out of scope for this NVIDIA baseboard datasheet.
Network · Not stated in this document. The baseboard integrates fifth-generation NVLink and fifth-generation NVSwitch for GPU-to-GPU interconnect, with 14.4 TB/s total NVLink bandwidth. Host-facing networking (e.g., ConnectX/BlueField NICs) is determined by the OEM system and is not part of the baseboard reference.
Not suitable for facility-level PUE math -- this is a GPU-only TBP sum for a bare 8-GPU OEM baseboard, not a measured or vendor-stated system power draw. Real HGX B200-based servers from OEMs publish their own higher system-level power specs (including CPUs, DIMMs, storage, fans, PSU overhead) that were out of scope for this NVIDIA-primary-source-only candidate.
PCF Summary for NVIDIA HGX B200 | Datasheet ↗Cooling · Not stated in the platform data sheet — no air/liquid designation is given for the 8x OAM UBB baseboard itself.
Memory / DIMMs · Not specified — this data sheet documents the 8x OAM GPU baseboard only; no host CPU/DIMM memory configuration is given.
Network · 8x PCIe Gen 5 x16 (128 GB/s) host I/O, one per GPU; intra-baseboard AMD Infinity Fabric mesh at 7x 128 GB/s per GPU, 896 GB/s peak aggregate bidirectional ring bandwidth.
This is a GPU-baseboard-only figure (8x OAM accelerators on a UBB 2.0), not a full-server or full-rack power draw. It excludes host CPU(s), system memory, storage, networking, fans, PSUs, and AC/DC conversion losses, so it is not suitable for facility-level PUE math on its own — only for screening accelerator-silicon load.
AMD Instinct MI300X Platform Data Sheet (GD-83, PID# 232405395-D) ↗Cooling · Not stated in the platform data sheet — no air/liquid designation is given for the 8x OAM UBB baseboard itself.
Memory / DIMMs · Not specified — this data sheet documents the 8x OAM GPU baseboard only; no host CPU/DIMM memory configuration is given.
Network · 8x PCIe Gen 5 x16 (128 GB/s) host I/O, one per GPU; intra-baseboard AMD Infinity Fabric mesh at 7x 128 GB/s per GPU, 896 GB/s peak aggregate bidirectional ring bandwidth.
This is a GPU-baseboard-only figure (8x OAM accelerators on a UBB 2.0), not a full-server or full-rack power draw. It excludes host CPU(s), system memory, storage, networking, fans, PSUs, and AC/DC conversion losses, so it is not suitable for facility-level PUE math on its own — only for screening accelerator-silicon load.
AI Data Sheet: AMD Instinct MI325X Platform (GD-83) ↗Cooling · Air or liquid capable up to 550W Max TDP per socket; liquid-cooling-only above that, up to 760W Max TDP per socket.
Memory / DIMMs · 128 GB HBM3 unified memory per APU package, coherently shared between the integrated CPU cores and GPU compute units (not conventional host DIMMs). No separate host-memory/DIMM configuration is given since the data sheet covers only the APU package.
Network · Per APU: 8x 128 GB/s AMD Infinity Fabric interfaces (4 dedicated Infinity Fabric, 4 assignable to Infinity Fabric or PCIe Gen 5 x16); scale-out listed as 400 Gbps Ethernet or InfiniBand; two x4 NVMe storage interfaces and two USB interfaces per APU (device counts/models not stated).
No AMD-published platform/system total power figure exists for any multi-APU MI300A configuration in this primary source — only per-socket TDP is stated, and the document's own '4-APU' reference architecture carries no accompanying wattage. Third-party OEM systems (e.g., Supermicro 4-way MI300A servers, HPE Cray EX blades used in El Capitan) do publish their own node-level power figures, but those are vendor-integrator sources, not AMD primary sources, and are out of scope for this candidate. This entry is included to document that the search was made and no genuine AMD platform-power figure was found, per the 'unknowns stay unknown' policy.
AMD Instinct MI300A APU Data Sheet (GD-83) ↗Cooling · Not stated on this page (no cooling method specified for this measured-power figure).
Memory / DIMMs · Not stated on this page. HBM2 is on-package (32 GiB, 1200 GBps); no separate DIMM/memory-channel detail given.
Network · Six inter-chip interconnect (ICI) links per chip enabling a 3D mesh/torus pod topology; x16 PCIe gen3 interface to host (direct connect), per the same document.
Chip-only figure, not a system/rack/pod total. Does not cover host servers, networking, or cooling. Not directly comparable to full-server or full-rack power figures used elsewhere in this dataset. Google has not published an equivalent figure for TPU v5e, v5p, v6e (Trillium), or TPU7x (Ironwood) in any official documentation or blog post found during this research.
TPU v4 | Google Cloud Documentation ↗Cooling · Not stated on this page (no cooling method specified for this measured-power figure).
Memory / DIMMs · Not stated on this page. HBM2 is on-package (32 GiB, 900 GBps); no separate DIMM/memory-channel detail given.
Network · 2D torus pod interconnect topology per the same document; per-chip host-connect interface not detailed on this page.
Chip-only figure, not a system/rack/pod total. Does not cover host servers, networking, or cooling. Not directly comparable to full-server or full-rack power figures used elsewhere in this dataset. Google has not published an equivalent figure for TPU v5e, v5p, v6e (Trillium), or TPU7x (Ironwood) in any official documentation or blog post found during this research.
TPU v3 | Google Cloud Documentation ↗Cooling · Not stated by AWS -- no cooling method or thermal design is published for the Trn2 UltraServer on the reviewed pages.
Memory / DIMMs · Not stated by AWS -- no host CPU/DIMM memory configuration is published for the Trn2 UltraServer on the reviewed pages.
Network · 64 Trainium2 chips connected via NeuronLink, delivering up to 83.2 FP8 petaflops of compute, 6 TB of total HBM3 with 185 TBps of total memory bandwidth, 12.8 Tbps of EFAv3 networking, and support for up to 32 TB of local NVMe storage.
Captures only the compute/memory/network specs AWS publishes for the Trn2 UltraServer. No power, thermal, cooling, DIMM, or physical host-CPU details are published at chip, instance, or aggregate UltraServer level. Local NVMe capacity (up to 32 TB) is stated but device count is not. Not usable for PUE or facility power planning without an independently sourced power figure.
Amazon EC2 Trn2 Instances — AWS (EC2 Trn2 product page) ↗Cooling · Not stated by AWS -- no cooling method or thermal design is published for trn2.48xlarge on the reviewed pages.
Memory / DIMMs · AWS states 2 TB of total instance memory for trn2.48xlarge (see cpu_memory_gb) but does not publish DIMM count, type, or channel configuration.
Network · 16 Trainium2 chips deliver up to 20.8 FP8 petaflops of compute, 1.5 TB HBM3 total with 46 TBps memory bandwidth, and 3.2 Tbps of EFA networking; supports up to 8 TB local NVMe storage. The trn2.48xlarge instance has 192 vCPUs and 2 TB of instance memory.
No power, thermal, cooling, DIMM, or physical host-CPU (as opposed to vCPU) details are published by AWS for the Trn2 instance. Not usable for PUE or facility power planning without an independently sourced power figure.
Amazon EC2 Trn2 Instances — AWS (EC2 Trn2 product page) ↗Cooling · Not stated by AWS -- no cooling method or thermal design is published for inf2.48xlarge on the reviewed pages.
Memory / DIMMs · AWS states 768 GiB of total instance memory for inf2.48xlarge (see cpu_memory_gb) but does not publish DIMM count, type, or channel configuration.
Network · inf2.48xlarge: 12 Inferentia2 chips, 384 GB total accelerator memory (32 GB per chip), 9.8 TB/s of total memory bandwidth, 192 GB/s NeuronLink interconnect between chips, 192 vCPUs, 768 GiB instance memory, and up to 100 Gbps network bandwidth.
No power, thermal, cooling, DIMM, or physical host-CPU (as opposed to vCPU) details are published by AWS for Inf2 instances. Not usable for PUE or facility power planning without an independently sourced power figure.
Amazon EC2 Inf2 Instances — AWS (EC2 Inf2 product page) ↗Cooling · Not specified in this product brief (no air vs. liquid cooling designation is given alongside the 900W figure).
Memory / DIMMs · Not applicable / not stated — this is an accelerator mezzanine card, no DIMM slots are documented on it.
Network · 9.6 Tbps bi-directional networking capacity via 24x200 GbE RoCE v2 RDMA ports integrated on-die (native on-chip RDMA over converged Ethernet).
Single-accelerator component TBP only, not a system/server/rack figure. Cooling method not stated. Not itself an 8-accelerator platform figure — see the companion HLB-325 baseboard candidate for that.
Product Brief: Intel Gaudi 3 AI Accelerator HL-325L OAM Mezzanine Card ↗Cooling · Not specified (no air vs. liquid cooling designation given). Brief states the baseboard 'derives the majority of its power from 54V power input, only requiring a separate 12V input for standby.'
Memory / DIMMs · Not applicable / not stated — the baseboard carries 8 accelerator OAM cards, not CPU DIMMs.
Network · Baseboard provides 4.2 TB/s bi-directional all-to-all bandwidth directly between the 8 onboard accelerators (no separate switching IC required), plus an additional 1.2 TB/s bi-directional scale-out bandwidth via 6 OSFP connectors (24x200 Gbps RoCE v2 RDMA).
This is a baseboard/accelerator-tray-only power figure explicitly excluding the host CPU subsystem, system memory, storage, and chassis cooling — it is not a complete server or rack figure and should not be used directly for facility-level PUE math without adding host-server overhead. No cooling method (air/liquid) is stated.
Product Brief: Intel Gaudi 3 AI Accelerator HLB-325 Baseboard ↗Cooling · Not explicitly stated in the datasheet (no air vs. liquid cooling designation given); described only as a 19-inch rack-mount server chassis.
Memory / DIMMs · 1TB of host system memory stated for the dual-socket Xeon 8380 subsystem ('Dual socket Intel 3rd Generation Xeon 8380 CPUs with 1TB memory'); DIMM count/type not specified. Separately, 768GB of HBM2E is on-package accelerator memory across the 8 Gaudi 2 cards (96GB each, stated directly in the spec table).
Network · 24x100GbE RoCE v2 RDMA scale-out via 6x QSFP-DD, plus internal non-blocking all-to-all interconnect (21x100GbE RoCE ports from each accelerator to the other 7), plus two 2x100GbE PCIe host NICs for host connectivity (per block diagram description).
Power figure is Intel's stated 'Max Power Usage,' likely a design/nameplate ceiling rather than a continuously-measured typical draw; cooling method, PSU redundancy, and AC/DC conversion overhead are not stated. This document describes Gaudi 2 (not Gaudi 3) and is dated October 2023 per its filename/copyright; hosted on habana.ai, Intel's wholly-owned AI-accelerator subsidiary domain, and is Intel-branded throughout ('Intel® HLS-Gaudi®2'). An Intel-hosted landing page for this same datasheet exists at intel.com/content-details/784779 but returned HTTP 403 when fetched directly.
Intel HLS-Gaudi2 AI Accelerator Server datasheet ('AI Performance with Ethernet Scale') ↗Cooling · Liquid cooling, direct-to-chip. Datasheet specifies: Water Filtering 50 micron; Water Chemistry PG25; Water Flow Rate 100 L/min/system; Water Temperature 20 ± 2°C.
Memory / DIMMs · Not stated. The datasheet lists 44 GB / 21 PB/s on-chip SRAM as processor memory; no host DIMM/DRAM configuration is given for the CS-3 system itself.
Network · 1.2 Tb/s I/O bandwidth; management: 1x 1GbE management port (RJ45), 1x console port (RJ45), 1x PoE flow controller interface.
Single-system power draw only; excludes any external MemoryX/SwarmX or networking cabinets used in multi-system training/inference clusters (the datasheet's separate 'Inference Cluster Configurations' table gives multi-cabinet provisioned heat loads instead, from ~240 kW for an 8-node/5-cabinet cluster up to multi-MW for 64-88 node clusters). No PUE/facility overhead included. CPU/host memory/storage not specified by this datasheet.
Cerebras 'CS-3 System Specifications' datasheet (PDF), linked directly from cerebras.ai/system ↗Cooling · Combined liquid + air cooling. Datasheet states 'Water Cooling Load per Rack: 50.4 kW' and 'Air Cooling Load per Rack: 3.6 kW' (sums to the 54 kW total). Liquid cooling water coupling: 1.5" sanitary fittings, 2 pairs supply/return. Power feed: 2+1 redundant, 240V/415Y 3-phase, 60A, 5-wire, IEC 60309 plug.
Memory / DIMMs · Not stated.
Network · 8x QSFP-DD transceiver, 4x100G; management network: 1x 1GbE management port (RJ45), 1x console port (RJ45), 1x PoE flow controller interface.
Rack-level total for exactly 2x CS-3 systems only; excludes any additional WSE NET/interconnect cabinets used in larger multi-rack clusters (the datasheet's separate cluster-configuration table shows those as additional cabinets with their own heat loads, e.g. ~240 kW provisioned total for an 8-node/5-cabinet cluster). No PUE/facility overhead included.
Cerebras 'CS-3 Rack Specifications' datasheet (PDF), linked directly from cerebras.ai/system ↗Cooling · Not stated in this spec sheet. GroqCard is a PCIe Gen4 x16 add-in card with no dedicated cooling spec published here; it relies on host-server chassis airflow.
Memory / DIMMs · Not applicable / not stated. GroqCard has no DIMM slots — its 230 MB of memory is on-die SRAM, not DRAM.
Network · Up to 11 RealScale chip-to-chip connectors for multi-server/multi-rack scalability without external switches; PCIe Gen4 x16 host interface, up to 31.5 GB/s bi-directional bandwidth.
Component-level (single PCIe accelerator card) power figure only — not a system, node, or rack total. Groq's GroqNode server datasheet states it packages up to 8x this card plus dual host CPUs (2x AMD EPYC 7313) and 1TB DRAM in a 4U chassis with '4 x 2000W (220-240VAC)' power supplies, but that is a PSU nameplate-capacity figure (likely N+redundant), not a stated power-draw number, so no full-node or full-rack GroqNode/GroqRack power figure is included here. We also could not locate a working GroqRack spec PDF with any published rack-level power figure as of this check.
GroqCard™ Accelerator Product Spec Sheet, v1.5 (PDF), linked from groq.com/groqcard-accelerator ↗Cooling · Air-cooled; 8 hot-swappable fan modules in N+1 redundancy mode (per Hardware Description, Maintenance and Service Guide 05).
Memory / DIMMs · Up to 32 DDR4 DIMM slots (RDIMM), max speed 3200 Mbit/s, ECC/SEC-DED/SDDC + patrol scrubbing protected, 16 GB/32 GB/64 GB per module (all installed modules must share the same part number).
Network · CPU mainboard supports one FlexIO card providing 4x 25GE/10GE optical ports (PXE-capable); NPU carrier board supports up to 4 parameter-plane interface cards, each with 2x 200GE optical ports (fiber or copper).
Does not capture NPU (Ascend 910) accelerator count or exact model suffix (e.g. 910B) -- Huawei's own Overview page names the processor only generically as 'Ascend 910 AI Processors' without a quantity on the pages retrieved, so gpu_qty is left null and gpu_model reflects only the generic name actually published. Power figure is a manufacturer-stated maximum across configurations, not a measured or typical operating draw. Storage device count and accelerator HBM capacity are not stated on the retrieved pages.
Atlas 800T A2 Training Server User Guide 04 (Huawei Enterprise Support) — Overview, Technical Specifications, and Physical Specifications sections ↗Cooling · Air-cooled; deployable in standard air-cooled equipment rooms, using integrated fan modules plus Huawei-developed liquid-assisted air cooling (LAAC) modules. Each drawer integrates 5 fan modules with 4+1 hot-swap redundancy (single-fan failover time <= 2 minutes).
Memory / DIMMs · Not stated with specific capacity or DIMM type on the pages retrieved.
Network · Not stated with explicit port-count/speed figures on the pages retrieved. The server integrates separate 'LingQu UB' interconnect boards, documented by Huawei in a distinct 'LingQu UB Device Hardware' / 'LingQu UB Device Specifications' section, for supernode-to-supernode fabric connectivity.
Does not capture accelerator HBM memory capacity or interconnect bandwidth -- figures such as '128 GB HBM per NPU' or specific 400GE/UB link counts are widely reported by industry/trade press but were not found stated in these terms on the specific Huawei pages retrieved, so accelerator_memory_gb and detailed network specs are left null. The exact Ascend NPU variant (e.g. '910C') is not named as such in the retrieved Huawei text, which refers to it generically as an 'Ascend 910 AI module' (HiAM module). Power figure is a manufacturer-stated maximum across configurations, not a measured or typical operating draw, and excludes the separate LingQu UB Device module's own power draw.
Atlas 800I A3 User Guide 04 (Huawei Enterprise Support) — Product Features and Physical Specifications sections ↗A denser layer than MW alone: power-train stages, cooling topology bands, rack-density eras, official buyer/producer indexes, and equipment lead-time stress tests. Screening orientation only — manufacturer BOMs and site engineering still win procurement.
Transmission/distribution interconnection, customer substation, utility transformers, metering, protection, curtailment terms
ISO/RTO queues · utility large-load tariffs · EIA-930 system context
MV switchgear, feeders, bus duct, main-tie-main, on-site transformers to LV
DOE distribution-transformer rules · ENERGY STAR UPS adjacency · AHRI cooling power draw
UPS topology, battery/flywheel autonomy, STS, PDU/RPP, busway to rack
ENERGY STAR UPS directory
Busway, busbar, power shelves, PSU conversion, HVDC experiments (OCP)
NVIDIA/AMD primary BOMs · OCP rack specs
Diesel/HVO/gas gensets, fuel storage, paralleling, BESS, black start, emissions controls
EPA engine certification data · state air permits
Rack band · ≤ ~15–20 kW/rack typical; practical ceiling often cited ~30–40 kW with rear-door assist
PUE band · ~1.3–1.6 (design- and climate-dependent)
Water · Often cooling-tower or dry-cooler dependent; WUE is site-specific
Use when · Legacy enterprise, moderate-density colo, inference pods that stay air-coolable
Industry practice band for screening; validate with ASHRAE thermal guidelines and OEM envelopes
Rack band · ~20–40 kW/rack common screening band
PUE band · ~1.2–1.4 when free cooling is available
Water · Facility water loop; can reduce air-handler count
Use when · Mid-density GPU refresh inside existing shells
Industry practice band; pair with AHRI datacom ratings for equipment shortlists
Rack band · ~40–150 kW/rack; AI training racks often designed 100–140+ kW
PUE band · ~1.03–1.15 when liquid fraction is high
Water · Secondary loop flow is not water consumption; WUE still needs a water balance
Use when · GB200/GB300-class racks and dense OAM platforms
Aligned with public AI-rack power disclosures (e.g. NVL72 up to 142 kW) plus industry DLC practice
Rack band · ~100–200+ kW/rack screening band
PUE band · ~1.02–1.08 claimed in vendor case literature — treat as aspirational until measured
Water · Dielectric fluid inventory, leak detection, and heat-rejection water still apply
Use when · Extreme density experiments, specialized HPC, or shells designed around tanks
Secondary/vendor literature heavy — require FAT/SAT and third-party thermal tests in RFPs
| Era | Rack power | Note | Evidence class |
|---|---|---|---|
| Pre-AI enterprise (illustrative) | 4–12 kW | Historical average enterprise densities often cited near mid-single to low-double-digit kW/rack. | secondary_industry |
| Modern air-cooled GPU pods | 20–40 kW | Many retrofit halls target this band before full liquid conversion. | secondary_industry |
| NVIDIA GB200/GB300 NVL72-class AI rack | up to 142 kW (GB300 NVL72 manufacturer maximum) | NVIDIA's GB300 NVL72 Enterprise Reference Architecture states up to 142 kW for the full liquid-cooled rack — a manufacturer maximum, not a workload average. | primary_manufacturer |
| Next-wave liquid AI cabinets (roadmap / secondary) | 200–600+ kW discussed in industry and vendor roadmaps | Treat roadmap kW figures as scenario inputs, not purchaseable nameplate, until an OEM publishes a BOM. | secondary_roadmap |
Metric · Seasonally adjusted annual-rate $ put in place
May 2026 preliminary · ~$59.3B SAAR (secondary rebroadcast of Census private construction detail)
Cleanest official monthly dollar flow into U.S. data-center construction — a buyer CapEx index.
Open series ↗Metric · Open notices for servers, UPS, generators, cooling, construction, ops
Continuous — filter by NAICS/PSC/CPV + place of performance
Buying intent before awards — the language of scopes, options, and delivery dates.
Open series ↗Metric · Index level, MoM and YoY %
Track via BLS/FRED; May 2026 prints near ~457 (index, not $)
Producer price pressure on the longest-lead electrical equipment.
Open series ↗Metric · Index level for multiuser host computers
Monthly BLS/FRED series
Hardware inflation/deflation input for server BOMs and lease pricing models.
Open series ↗Metric · Industry PPI for hosting / IT infrastructure provisioning
Monthly BLS industry product series
Service-side price index for colo/cloud-like outputs.
Open series ↗Metric · Listed-company monthly revenue
High-frequency producer throughput for AI servers and silicon
Producer supply heartbeat before annual reports — boards, power, cooling, and chips.
Open series ↗Metric · Value and quantity by HTS for transformers, generators, chillers, servers, optics
Monthly merchandise trade
Physical supply flows and tariff exposure for the equipment bill of materials.
Open series ↗Metric · TWh/year scenarios
Base case ~945 TWh by 2030 (~3% of world electricity)
Global demand ceiling used by policymakers and utilities; not a project queue.
Open series ↗Secondary survey and industry bands for schedule risk — not purchase-order commitments. Confirm with current OEM quotes.
| Equipment | Band | Evidence | Note | Planning use |
|---|---|---|---|---|
| Large power transformers | ~100–144+ weeks commonly reported; some GSU quotes longer | secondary_survey | Wood Mackenzie Q2 2025 survey figures widely reported: ~128 weeks power transformers, ~144 weeks GSUs. Use as a schedule stress test, not a purchase order. | Order at concept / before final design freeze; dual-source and accept-alternates language in RFP |
| MV switchgear | ~44–80 weeks reported depending on class and customization | secondary_survey | Trade-press summaries of 2025–2026 utility surveys; confirm with current OEM quotes. | Freeze one-line early; hold production slots |
| Diesel / large standby generators | ~50–78 weeks for many 1.5–3 MW-class blocks (indicative) | secondary_industry | Highly configuration-dependent; emissions aftertreatment and paralleling gear add time. | Air-permit family and fuel strategy lock with procurement |
| AI accelerator supply (GPU/OAM) | Allocation-driven; not a single public lead-time series | secondary_market | Track producer revenue (TWSE), foundry revenue, and OEM allocation letters rather than a BLS lead-time series. | BOM revision control + dual architecture options in RFP |
| Liquid-cooling CDUs / facility water skids | Project-specific; often on critical path once rack density exceeds air limits | secondary_industry | Pair AHRI datacom certifications with factory-witness tests; no single public national lead-time index. | Specify ΔT, approach, water quality, and controls interoperability explicitly |
Interconnecting TSPs must submit the Batch Zero LIF workbook, Dynamic Stability Study Form, and Section 9.2.1.2/9.2.1.4 attestations to BatchZero@ercot.com source
ILLEs must submit Form X (WLPUN designation), Form W (PCLR commitment), and Section 9.2.1.1 attestations to their interconnecting DSP/TSP source
PUCT approved rules for ERCOT to review an initial batch of large-load interconnection requests (75+ MW), including a $50,000/MW fee structure source
ERCOT tracking 438 GW of large-load requests; ~90% from data centers; 198 GW applied in Q1 2026 alone source
ERCOT reported receiving 137 new LLI submissions totaling approximately 140,000 MW of new large load by 2036, still being processed source
226 GW of large-load interconnection requests, up from 63 GW twelve months earlier; 73% from data centers source
| Month | Total queue | Approved | Observed peak | Source |
|---|---|---|---|---|
| 2026-03-31 | 438.0 GW | 9,042 MW | 3,883 MW | Utility Dive / PUCT; ERCOT Mar 2026 TAC report |
| 2025-11-30 | 226.0 GW | — | — | ERCOT public data via davefriedman.substack.com |
| 2025-03-31 | — | 9,042 MW | 3,883 MW | ERCOT LLI Status Update Mar 2026 (TAC) |
| 2024-11-30 | 63.0 GW | — | — | ERCOT public data via davefriedman.substack.com |
| 2024-01-31 | 39.3 GW | 4,479 MW | 2,587 MW | ERCOT LLI Status Update Jan 2024 |
| 2023-12-31 | 39.2 GW | 4,479 MW | 2,432 MW | ERCOT LLI Status Update Jan 2024 |
| 2023-10-31 | 39.3 GW | 3,926 MW | 2,379 MW | ERCOT LLI Status Update Jan 2024 |
| 2023-09-30 | 39.6 GW | 3,926 MW | 2,379 MW | ERCOT LLI Status Update Jan 2024 |
| 2023-08-31 | 42.5 GW | 3,744 MW | 2,347 MW | ERCOT LLI Status Update Jan 2024 |
| 2023-07-31 | — | 3,084 MW | 2,336 MW | ERCOT LLI Status Update Jan 2024 |
| 2023-05-31 | 41.8 GW | 2,620 MW | 2,072 MW | ERCOT LLI Status Update Jan 2024 |
| 2023-04-30 | 39.6 GW | 2,570 MW | 2,013 MW | ERCOT LLI Status Update Jan 2024 |
| 2023-03-31 | 39.4 GW | 2,570 MW | — | ERCOT LLI Status Update Jan 2024 |
| 2022-04-30 | 17.4 GW | — | — | ERCOT LLI Status Update Jan 2024 |
Dual axis · log efficiency (Gflops/W) + HPL power (MW) · measured as published, never estimated
| List | System | Country | HPL Rmax | Power | Gflops/W |
|---|---|---|---|---|---|
| 2024-11 | El Capitan | United States | 1,742.0 PF | 29.6 MW | 58.89 |
| 2024-06 | Aurora | United States | 1,012.0 PF | 38.7 MW | 26.15 |
| 2022-06 | Frontier | United States | 1,102.0 PF | 21.1 MW | 52.23 |
| 2020-11 | Fugaku | Japan | 442.0 PF | 29.9 MW | 14.78 |
| 2019-06 | Summit | United States | 148.6 PF | 10.1 MW | 14.72 |
| 2016-06 | Sunway TaihuLight | China | 93.0 PF | 15.4 MW | 6.05 |
| 2013-06 | Tianhe-2 | China | 33.9 PF | 17.8 MW | 1.90 |
| 2012-11 | Titan | United States | 17.6 PF | 8.2 MW | 2.14 |
| 2011-11 | K computer | Japan | 10.5 PF | 12.7 MW | 0.83 |
| 2009-11 | Jaguar | United States | 1.8 PF | 7.0 MW | 0.25 |
| 2008-06 | Roadrunner | United States | 1.0 PF | 2.3 MW | 0.44 |
Documented data centers, computing payloads, and proposals in non-standard environments — including Antarctic research IT (McMurdo IT&C, IceCube, South Pole Telescope). Communications platforms and edge systems are labelled when they are not full data centers; no unpublished MW is inferred. Commercial hyperscale does not operate on the Antarctic continent under the Treaty System.
Axiom Space · International Space Station
Commercial cloud device operationalized aboard the ISS in 2022; commercial AI inference conducted in orbit. Edge device, not a full data center.
Axiom Space · International Space Station
Data Center Unit-1 powered by Red Hat Device Edge; cloud computing, AI/ML, data fusion, and space-cybersecurity test applications.
Axiom Space · Free-flying low Earth orbit nodes
First two dedicated orbital data-center nodes launched January 11, 2026; 2.5 Gbps-capable optical intersatellite links.
NASA + Hewlett Packard Enterprise · International Space Station
HPE ProLiant DL360 server plus Edgeline 4000 server; research and commercial edge-compute validation.
Starcloud · Low Earth orbit
First NVIDIA H100 GPU in orbit; company reports Gemini inference and nanoGPT training in space.
Axiom Space · International Space Station
Planned additional ISS ODC node; Axiom says it plans at least three interconnected ODC nodes by 2027.
Starcloud · Sun-synchronous orbit
GPU cluster; persistent storage; 24/7 access; proprietary thermal and power systems in a smallsat form factor.
European Space Agency · ESA Discovery and Preparation study
Generic space-data-centre architecture and simulation; the work examines technical operational and economic constraints before mission approval.
Lonestar Data Holdings · Odysseus landing site near the lunar south pole
Commercial lunar data-storage and edge-processing test; data was stored on the Moon and returned to Earth.
Highlander / HiCloud · Lingshui Li, Hainan, China
Builder-backed project reported as running smoothly after its first module launched March 31, 2023; commercial undersea data-center service and deployment program.
Microsoft · European Marine Energy Centre, North Sea
240 kW; 12 racks; 864 servers; 27.6 PB; deployed June 2018 and retrieved July 2020.
Korea Ministry of Oceans and Fisheries + 12-org consortium (KIOST, UNIST, POSCO, GS E&C, SK Telecom) · Offshore Ulsan, South Korea
Site selected April 2026; test module at 20 m depth (as reported); target PUE 1.2; 51.1 billion won total investment through 2030; demonstration facility targeted by 2030, commercial construction from 2031. Nothing deployed yet.
Highlander / CCCC · China coastal regions
Strategic partnership signed June 2026 for scaled undersea compute deployment and computing-power / offshore-energy synergy.
Subsea Cloud · Subsea deployment portfolio
Modular subsea data centers on a build/deploy/maintain model; company marketing claims up to 700 MW per-site capability (a capability claim, not deployed capacity); Project OTTO rolling customer trial began October 2024 off south-west Norway with a 16-rack unit at 150 kW+ per rack.
Microsoft · Pacific Ocean seabed
First underwater data-center pilot operated approximately one kilometer off the Pacific coast from August to November 2015.
NetworkOcean · San Francisco Bay, California, USA
Company claimed a 500 kW capsule and offered 2,048 NVIDIA H100 GPUs for reservation; sought no permits and SF Bay regulators warned of heavy fines (Wired, Sept 2024). No public evidence the test ever took place; current status unknown.
Denv-R / Geps Techno · Quai Wilson, Loire River, Nantes, France
200 kW in a 100 m2 solar-covered container housing four racks; passive Loire river-water cooling without pumps; operational since October 2024. A swell-powered offshore version is planned, not built.
Nautilus Data Technologies · Stockton, California, USA
6.5 MW critical IT load across four data halls; Tier III; ~55 kW rack densities via river-water cooling; commissioned 2021. Listed for sale at $45m in November 2024 at 86% leased as the operator pivots to licensing its EcoCore cooling technology.
Aker BP / Armada · Norwegian Continental Shelf, Norway
Agreement announced March 2026; begins with a single containerized Galleon unit for on-rig processing of drilling and operational data. Edge compute on a working platform, not a colocation data center; no capacity figures published.
Keppel Data Centres · 25 Loyang Crescent, Singapore
25 MW project with a four-story 19.2 MW waterborne module; seawater-cooled; capacity committed to an unnamed global hyperscaler; funded by Keppel Data Centre Fund II. Construction began Q1 2026; go-live targeted 2028.
Samsung Heavy Industries · No site announced
Concept/pre-order stage targeting commercialization as soon as Q2 2028; dedicated barge with on-board electrical infrastructure, initially shore-grid powered with water cooling. No vessel built, no committed capacity.
Mitsui O.S.K. Lines / Hitachi / Hitachi Systems · Japan (Malaysia and US under consideration)
MOU signed March 30, 2026 for demand verification and feasibility studies; operations targeted 2027 or later. Illustrative example: a car carrier offers ~54,000 m2 floor area; ~1-year conversion vs several years for a land build. No vessel converted yet.
Arctic World Archive (Piql / SNSK) · Mine no. 3, Longyearbyen, Svalbard, Norway
Offline, air-gapped piqlFilm vault 300 m inside the mine at about -4C; launched March 2017; hosts the GitHub Arctic Code Vault. Archival storage with manual retrieval, not an online data center - zero compute, off-grid by design.
Bluebird Network · Springfield, Missouri, USA
6 MW built-out capacity (per Data Center Map listing); 85 feet underground encased in solid limestone; three diverse utility feeds, three backup generators, SOC 2 Type II; colocation plus fiber transport and Internet services.
Tencent · Gui'an New Area, Guizhou, China
Capacity for 300,000 servers with about one-sixth housed in five tunnel caves; more than 30,000 m2 of tunneled space on a ~470,000 m2 site; maximum PUE ~1.1 logged in a Ministry of Industry and Information Technology inspection.
GPI · Val di Non, Italy
European data center built inside an active mine approximately 100 meters below the surface; officially online in June 2026.
Lefdal Mine Data Centers · Måløy, Norway
Six underground levels and 75 chambers; up to 120,000 m2 potential white space; Level 3 capacity is 80 MW.
Bahnhof · Stockholm, Sweden
Operating data center built inside a Cold War civil-defense / nuclear fallout shelter beneath Vita Bergen.
Green Mountain · Stavanger, Norway
Former high-security NATO ammunition storage converted to a Tier III mountain-hall data center; 25 MW site capacity and 22,600 m2 space.
MOUNT10 · Swiss Alps, Switzerland
Underground data-storage and data-center site in former Swiss military bunkers.
Iron Mountain · Butler County, Pennsylvania
220 feet underground; 15.5 MW power capacity; 330,000 square feet; geothermal cooling from an underground reservoir.
Westland Bunker · Montgomery County, Texas, USA
Built 1982 as Westland Oil headquarters with a 40,000 sq ft nuclear bomb shelter; expanded with a 107,000 sq ft data center; 23 MW power capacity (per Baxtel); zero-downtime performance through Hurricane Ike.
Deltalis · Near Attinghausen, Swiss Alps, Switzerland
Up to 10,000 m2 planned colocation space in a Cold War command-and-control bunker; Deltalis exited retail colocation in 2018 and the site was reportedly repositioned toward cryptocurrency mining/storage.
SoftBank / HAPSMobile · Stratosphere at approximately 19 km
62,500-foot / 19-km test flight; first LTE connectivity from a fixed-wing HAPS. Communications platform, not a data center.
AALTO / Airbus · Stratosphere above 60,000 feet
Solar-electric stratospheric aircraft with observation and connectivity payloads. Airborne platform, not a data center.
SoftBank / Sceye · Stratosphere at approximately 20 km
Long-duration helium HAPS; pre-commercial service scheduled in Japan for 2026. Communications platform, not a data center.
Academic proposal · High-altitude platform
Research proposal for a flying data-center-enabled HAP; no commercial deployment identified.
Cape Peninsula University of Technology (A. A. Periola) · Conceptual; modeled at 16-30 km altitude
Peer-reviewed MATLAB simulation of stratosphere-based data centres: modeled 43.9% average PUE improvement from an intelligent server-awakening architecture. No hardware built, flown, or deployed.
Project queues, product certification, server/UPS/storage/network BOMs, water and emissions, tenders, trade, construction, supplier revenue, climate, and physical risk. Each card states its signal, cadence, access path, and authority so a forecast is never mistaken for a transaction or a certified model.
Statewide registry of every air permit issued to a data center in Virginia (backup diesel-generator permits): site name, registration number, issuance date, program type and county, with 198 rows as of July 6, 2026. Bot-blocked; scraping requires a browser-like session.
Research proposal for a flying data-center-enabled high-altitude platform, including energy, solar, thermal, communication, and offloading analysis. It is a study, not a deployment record.
Primary platform source for the solar-electric Zephyr HAPS operating above 60,000 feet with earth-observation and connectivity payloads. It is tracked as an airborne compute-adjacent platform, not a data center.
Primary HAPS source covering Sunglider testing at 19 km, Sceye long-duration platform, solar power, payloads, and planned pre-commercial communications service. HAPS is tracked as atmospheric edge infrastructure, not a full data center.
Lets backup-generator RFPs connect engine families to certified emissions characteristics before air-permit screening.
Closest official price index for the service side of the data-center product — a buyer/seller contract benchmark, not a facility MW measure.
Practical API layer for stitching construction-spend (buyer), transformer PPI (producer), and server PPI into one monitoring graph.
National capacity map for large users: available transport capacity, congestion, queue counts, and requested MW by supply area. It is a constraint and connection-readiness source, not a data-center project list.
Official PJM capacity-market auction library with results, supply curves and planning parameters for every delivery year back to 2007/08. The 2027/2028 BRA cleared at $333.44/MW-day RTO-wide (up from $329.17), with total cleared-capacity cost of $16.4B.
Free JSON APIs over every SEC filing: the XBRL company-concept endpoint returns full quarterly time series of tagged line items such as purchases of property and equipment for Microsoft, Alphabet, Amazon, Meta and the data-center REITs. Filings appear within minutes of release; a descriptive User-Agent header is required.
Paid-tracker firm publishing headline numbers free in its article stream: hyperscale operator capex of $142B in Q3 2025, 1,297 hyperscale data centers worldwide with 770 in pipeline, and US share of hyperscale capacity at 55%.
Supports free-cooling hours, design-day temperature, evaporative-water stress, flood/precipitation, and long-run climate-normal comparisons.
Enables transparent GPU-only load and memory arithmetic while exposing what a full platform RFP still must specify.
Turns CPU, memory, PSU, and measured efficiency claims into a comparable machine-readable buying universe.
Provides a defensible bridge from AI capacity targets to rack count, memory, network ports, liquid cooling, and electrical scope.
Bridges component BOMs to facility busway, busbar, and HVDC design language used in hyperscale RFPs.
Maps signed connection projects and high-/low-voltage sites with contracted capacity. It can inform location and queue diligence for large loads, but it does not publish a data-center-specific queue and access requires registration.
Official monthly survey of US construction spending with a dedicated private 'Data center' line, published as machine-readable XLSX (seasonally adjusted annual rate plus unadjusted history). May 2026 preliminary shows $59.3B SAAR, exceeding all general office construction.
Provides the clearest official monthly buyer-side index of dollars flowing into U.S. data-centre construction.
Moves cooling RFPs from marketing claims toward comparable, third-party-certified thermal performance.
Annual survey (1,033 respondents in 2025) of data-center cooling technology mix, including direct liquid cooling adoption at 22% of organizations, DLC drivers and barriers, and the rack-density threshold at which operators expect air cooling to fail.
Daily model-based estimate of Bitcoin network electricity consumption with a free CSV download API (134.19 TWh annualised best guess as of 2026-07-11) and a companion mining map with country and US-state hashrate shares. CC BY-NC-SA licensed.
Official data-center connection policy, application form, technical assessment, constrained-area criteria, demand fees, maximum-import-capacity process, and energisation prerequisites. Individual applications are not published as a public queue.
IEA's dedicated study of data-centre and AI electricity demand: ~415 TWh consumed globally in 2024 (~1.5% of world electricity), projected to ~945 TWh by 2030, with US at 45% of 2024 consumption and China at 25%. Underlying dataset downloadable from the report page.
ERCOT's paid emergency curtailment program in which qualified loads commit MW deliverable in 10 or 30 minutes, procured four times per year; awarded MW and cost by service type are posted per contract term via the NP3-144-M data product.
Mandatory EU-wide reporting scheme requiring every data centre with >=500 kW IT demand to report PUE, WUE, energy-reuse and renewable factors to a European database each May, with an aggregated public dashboard and periodic assessment report. Commission reporting puts EU data-centre electricity use at 70 TWh in 2024.
DOE-commissioned bottom-up model of US data-center electricity use: 176 TWh in 2023 (4.4% of US electricity), projected to 325-580 TWh (6.7-12%) by 2028, with server/storage/network/cooling component breakdowns. Companion free benchmarking tools (DC Pro, PUE estimators) hosted on the same site.
Adds a carbon and industrial-neighbor layer for siting diligence beyond eGRID subregion averages.
Creates a repeatable environmental diligence layer around candidate campuses, generators, cooling discharge, and nearby industrial burden.
Leading development signal that appears before many commercial inventory databases — one permit is not one campus.
Free press releases carrying hard numbers from paid quarterly trackers: data center physical infrastructure (power/cooling) revenue ~$12B in 1Q 2026 (+28% YoY, thermal management +~50%), and data center semiconductor/component revenue +116% YoY.
A procurement-grade cross-check for whether specified power and mechanical equipment has filed required federal certifications.
Continuously maintained ArcGIS web map of existing, under-construction and proposed data centers across Virginia, compiled from county land-use filings, with a companion diesel-generator air-quality map. A regional nonprofit tracker, but widely cited as the de-facto state dataset.
Creates a mandatory, comparable facility-performance layer spanning electricity, water, efficiency, and waste-heat indicators.
Rare national primary meter: 23% of Ireland metered electricity in 2025 — a buyer-side system-impact index, not a developer claim.
Primary U.S. energy baseline: 176 TWh (2023) and 325–580 TWh (2028 range) — the denominator for national share claims.
Monthly Manufacturers' Shipments, Inventories and Orders survey with downloadable XLSX tables breaking out NAICS 335 electrical equipment: May 2026 advance shows new orders of $18.4B and unfilled orders of $54.6B (+5.9% YoY NSA).
System-planner data-center capacity scenarios and connection diligence. The companion connection-reporting process records proponent, site, coordinates, capacity, projected demand, commissioning date, and project stage; detailed portal access is controlled.
Primary demand-and-supply planning and firm-generation procurement signal. EMA identifies data centers and semiconductors as high-demand drivers; it does not identify individual data-center projects.
Current long-term forecast report, tables, data workbook, data-center accuracy report, and load-adjustment evidence. PJM vets large-load requests for commitments and double-counting; this is a forecast and validation source rather than a public queue.
Official system-planning entry point for data-center demand integration, connection processes, forecast development, and the 2026 ISP modelling/data workbooks. It is a planning source, not a public project queue.
Supports location-based carbon, marginal procurement context, and grid-resource-mix comparisons without relying on vendor sustainability claims.
Authoritative global demand envelope (base case ~945 TWh by 2030, just under 3% of world electricity) used to size national/grid conversations without confusing MW requests for metered load.
Adds hourly and forecast load context for testing whether data-center requests are large relative to the system that must serve them.
Provides the denominator and utility-level context needed to compare proposed hyperscale load with existing customers, sales, and programs.
Places data-center servers inside the official long-run electricity outlook rather than as an orphan side estimate.
Offers a rare operator-published leading indicator: projects in application, approval, and implementation before they appear in metered demand.
Operator planning forecast (e.g. up to ~30 GW by 2030) for the densest U.S. interconnection market outside ERCOT deep-dive coverage.
Time-coincident load, wind, and solar generation profiles. ERCOT actuals cover 2017-2018; NYISO/MISO/SPP actuals cover 2018-2019. Includes deterministic and probabilistic forecasts for existing sites and projected capacity-growth scenarios derived from interconnection-queue proposals.
The canonical US-wide interconnection-queue dataset covering all 7 ISOs/RTOs plus ~50 non-ISO utilities (~98% of US generating capacity), with project-level Excel data on capacity by fuel, status, region, queue duration and completion rates. The 2026 edition covers ~8,200 active projects with data through end-2025.
Operator requirements and planning context for non-conforming large loads, including data centers and advanced manufacturing. Useful for operational-readiness signals; it is not a public, project-level queue.
Monthly dashboard of high-voltage connection requests to the Italian grid that explicitly breaks out data centers as a consumer-load category, with request counts, GW and authorisation status navigable to regional/provincial level. Bulk files available via the portal's Download Center.
Public interconnection queue spreadsheet whose scope explicitly includes new and materially modified load facilities alongside generation and transmission, with queue position, MW, zone, status and dates per project.
Transmission-system connection data published from Statnett’s connection database: reserved capacity and mature queue projects by company, industry category, MW volume, location, and planned completion. Category filtering can surface data-center / industrial demand where disclosed.
Project-scale large-load request workflow, study and fee milestones, energisation path, status materials, and committee records. This is the current deepest public data-center / large-load queue in the registry.
Primary system-operator paper with operating data-center peak demand, contracted capacity, demand share, connection applications, technical constraints, and 2032 adequacy scenarios. Some figures combine data centers with other new-technology loads and are labelled that way in Grid Watch.
Free brokerage research with market-by-market vacancy, net absorption MW, under-construction MW and asking-rate trends; the 2026 edition reports North America top-4 net absorption of 2,236 MW in Q1 2026 and Northern Virginia vacancy of 0.3%.
Quarterly earnings 8-Ks on EDGAR with detailed HBM product and supply disclosures; FQ3-2026 revenue was $41.46B vs $9.30B a year earlier, with HBM4 in high-volume shipment and FQ4 guidance of ~$50B. EDGAR is the bot-friendly path (Micron's IR site times out).
Daily DRAM and NAND flash spot prices by part (DDR5/DDR4 by density, NAND) plus the DXI market index, updated intraday; DDR5 16Gb averaged $47.80 on July 10, 2026.
Adds the often-missed network portion of IT load and makes RFP shortlists auditable at model level.
The industry-standard submarine cable map exposes free unauthenticated GeoJSON endpoints for cable routes and landing points (695 cable systems and 1,917 landing points verified on 2026-07-12), with per-cable owner, length and ready-for-service metadata.
Primary press releases for the flagship operating-fleet nuclear PPAs with data-center buyers: Microsoft-Crane Clean Energy Center (~835 MW Three Mile Island restart, 20-year) and Meta-Clinton (1,121 MW, 20-year from June 2027), with MW and duration on-page.
Vendor and deployment source for underwater intelligent computing cabins, private cloud, undersea data centers, deployment projects, and the 2026 marine computing-power / offshore-energy partnership.
Primary mission record for the two underwater prototypes: deployment, retrieval, power, rack/server payload, storage, reliability, water use, and the current research-stage status.
Commercial subsea data-center platform describing modular build, deployment, maintenance, cooling, connectivity, and public performance claims. No public MW or operating customer count is assumed.
Quarterly disclosure of orders (record €17.7B in Q2 FY2026), order backlog (€154B) and book-to-bill (1.72) by segment; Grid Technologies covers HVDC, switchgear and grid-connection equipment with US-driven growth.
Quarterly disclosure of total orders ($18.3B in Q1 2026), total backlog ($163B) and, uniquely, gas power equipment backlog plus slot-reservation agreements in GW (83 to 100 GW during Q1 2026, guiding to 110+ GW by end-2026).
Generator-level inventory: nameplate capacity, fuel type, owner/utility, in-service and planned/actual retirement dates, prime mover, technology, and environmental equipment. Texas/ERCOT plants are a filterable state subset, not a separate file.
Quarterly tracker of US data-center projects blocked or delayed by local opposition, moratorium proposals and state legislation; Q1 2026 counted at least 75 projects worth ~$130B blocked or delayed, with 300+ state bills filed in six weeks. Note the name collision with datacentr.net's own branding — attribute carefully.
Primary orbital data-center program record covering AxDCU-1, the first two free-flying ODC nodes launched January 11, 2026, optical links, the ISS node roadmap, and long-term kilowatt-to-megawatt expansion intent.
Adds a consistent multi-hazard screen for flood, hurricane, wildfire, heat, winter weather, earthquake, and other correlated siting risks.
Supports primary/secondary/tertiary power-chain scoping with comparable UPS ratings and part-load efficiency.
Anchors transformer efficiency, compliance, and replacement assumptions while supplier lead-time signals are assessed elsewhere.
Quarterly P25 solar/wind PPA offer-price index built from marketplace offers across AESO, CAISO, ERCOT, MISO, PJM and SPP; public posts carry directional QoQ/YoY changes each quarter (Q1 2026: P25 solar +4.6% QoQ, wind +8% QoQ), with absolute levels subscriber-only.
Provides an open, cross-border buyer-intent feed for data-centre construction and equipment with both pre-award and award-stage evidence.
Surfaces buying intent and RFP language for servers, accelerators, cooling, UPS, generators, substations, construction, and operations before awards.
Official monthly producer price indexes via free JSON API for transformers (PCU335311335311), switchgear/switchboards (PCU335313335313) and electrical contractors on nonresidential work (PCU23821X23821X). Transformer PPI is up ~5.5% YoY as of May 2026 and has more than doubled since pre-2020.
Primary producer-price pressure for the long-lead substation equipment that gates campus energization.
Producer-side price index for server hardware used in RFP escalation clauses and CapEx sensitivity screens.
Upstream silicon/component price pressure that eventually flows into accelerator boards and networking silicon.
A direct project and customer-identity signal that appears earlier than many commercial inventory datasets and can be reconciled to permits and grid requests.
FERC's generic proceeding on large loads co-located at generating facilities (AD24-11) plus the related company-specific dockets (Talen/AWS ISA amendment, BGE/PECO/ComEd/Pepco filings) and the EL25-49 show-cause proceeding on PJM co-location rules. Full filing history available via FERC eLibrary docket sheets.
North American reliability assessment with planner-submitted demand forecasts, large-load / data-center growth, resource-adequacy risk, and regional exposure. The source is a validated planning outlook, not an interconnection queue.
Monthly average retail electricity price by state and sector with YoY comparison (Table 5.6.A); April 2026 shows US industrial average 8.66 c/kWh, Texas 6.33, Virginia 9.86. Also queryable via the EIA v2 API for automation.
Free downloadable Excel of monthly worldwide semiconductor billings by region spanning four decades, updated monthly from the WSTS Blue Book; latest data May 2026. This is the machine-readable series behind SIA's monthly press numbers ($120.6B in May 2026, +104.1% YoY).
Statutory monthly revenue disclosure (NT$ millions, MoM/YoY, YTD) published around the 10th of each month; May 2026 revenue was up 30.1% YoY. TSMC fabricates nearly all NVIDIA/AMD accelerator dies.
Connects storage architecture and capacity procurement to measured energy characteristics rather than nameplate guesses.
Creates a high-frequency producer-side view of AI server, PCB, power, cooling, and semiconductor demand before annual financial statements.
Japan's cross-regional grid coordinator publishes English editions of the annual Aggregation of Electricity Supply Plans and seasonal supply-demand outlooks, with 10-year demand and capacity projections for the 10 service areas; data-center and chip-fab demand upgrades surface here first.
Searchable database of GASB 77 tax-abatement disclosures from state and local government financial statements, FY2015-FY2025, covering all 50 states plus major localities; a companion April 2026 report quantifies data-center-specific losses (Georgia $2.5B, Virginia $1.94B, Texas $1B per year).
Measures physical supply flow and price/value pressure for transformers, switchgear, generators, chillers, servers, semiconductors, and optical equipment.
Free press-release summaries of Wood Mackenzie's paid transformer supply and lead-time survey: 30% power-transformer and 10% distribution-transformer supply deficits in 2025, demand +116% since 2019, imports covering 80% of power-transformer supply, and average lead times around 120 weeks.
Company announcement for an operating data center approximately 100 meters below the surface inside an active mine in Val di Non.
Operating high-security data center in converted mountain halls / former NATO ammunition storage, with published site capacity, space, PUE, grid supplies, and build-out description.
Operating underground data-center campus in a former mine with published chambers, white-space potential, levels, power backbone, and 80 MW Level 3 capacity.
Operating Stockholm data center built inside a former Cold War civil-defense / nuclear fallout shelter, with published security, connectivity, cooling, and operations detail.
Operating secure underground data-storage / data-center infrastructure in former Swiss military bunkers in the Alps.
Operating colocation site 220 feet underground in a former limestone mine, with published 15.5 MW power capacity, 330,000 square feet, geothermal cooling, utility, and connectivity.
Makes watershed availability, drought response, intake/discharge context, and cooling-water risk measurable at site scale.
Cooling topology is often decided by water rights and discharge permits before a CDU model is selected.
Annual environmental report disclosing total operational water (2024: 11,011 million gallons withdrawn, 8,135 consumed, +28% YoY), 64% freshwater replenishment, and per-site water data across Google's data center cities.
Weekly national drought classification (D0-D4) with downloadable statistics by state, county and urban area, DSCI index, GIS shapefiles and web services for programmatic pulls.
Free republication of ICE-traded daily day-ahead firm power prices (weighted-average index, high/low, volume) at seven hubs including ERCOT North and PJM West, with current-year and multi-year XLSX archives.
Machine-readable appendix to Meta's annual sustainability report with 2020-2024 time series for data-center WUE (0.30 down to 0.19 L/kWh), PUE (1.10 to 1.08), water withdrawal by source in megaliters, and withdrawals from high-water-stress areas.
Official fleet efficiency page publishing PUE and WUE for Microsoft's owned datacenter fleet, globally and by Americas/EMEA/APAC, for the last two fiscal years: FY25 global WUE 0.27 L/kWh and PUE 1.17.
Critical-facilities, MEP, commissioning, and operations roles from public Greenhouse, Lever, and Ashby boards — plus optional sponsored listings when Adzuna keys are configured. No Indeed or LinkedIn scraping.
Daily digest signup not configured yet — see deploy/daily-digest-setup.md.
Project-level LLI tracking, county permits, CSV/API export, docket alerts.
Open pro database Free while in beta
A public, no-PII feed of what the automation behind Data Center Watch is doing right now — pipeline refresh runs, curator/audit agent outcomes, new job postings, and anonymized subscriber signups. This never shows an email address.
A primary-source intelligence dashboard for data-center power: interconnection queues, live grid demand, generation capacity, HPC systems, and the power, cooling, and supply-chain dependencies behind new sites.
Every figure links to its primary source. Live grid demand and generation capacity come from EIA Forms EIA-930 and EIA-860/860M. ERCOT queue and watch-feed figures come from its own LLI status updates, TAC/LLWG decks, GIS reports, and PUCT filings. The remaining sources are catalogued in the worldwide connection & buildout sources section — nothing is interpolated to fill a gap.
ERCOT has the most complete public interconnection-queue reporting, so Data Center Watch tracks its project-level queue and watch-feed line items in the greatest depth. Other regions get live demand and generation capacity from EIA but not a project-level queue.
No. Data Center Watch is not affiliated with ERCOT, EIA, or any grid operator. Data may contain errors — verify against primary sources before making decisions.
Yes. The facility registry is downloadable as JSON and CSV, the news monitor is available as RSS, and the pro database has its own JSON API — see the worldwide connection & buildout sources section and the pro database section for links.
An automated pipeline refreshes the underlying data multiple times a day; the live activity console on this page shows each refresh run as it happens.