Stage 3 of 9
Data centre shell and cooling datacenter
Land, permits, water and cooling capacity turn connected megawatts into rack-ready megawatts. Density is rising with liquid cooling.
- Unit
- MW of IT load
- Sells
- a building with power, cooling and networking ready for racks, sold as capacity by the megawatt
- Buys from
- Grid, interconnection and transformers
- Lead time
- 1.5 years
- Status
- skeleton
Primer
What this stage is
The building that turns a grid connection into rack-ready megawatts: land, permits, a shell, power distribution, cooling and water, and the fibre to reach the outside world. Its unit is megawatts of IT load, the power available to the computing equipment itself.
How it works
A data centre is priced by the megawatt it can deliver to racks, and built in months once power is secured. Two ratios matter. Power usage effectiveness is total facility power divided by IT power; the difference is cooling and distribution loss. Rack density is the power one rack draws; GPU racks draw many times what conventional servers do, which is why liquid cooling has become standard for new AI capacity and why older buildings cannot simply be refilled.
Operators either build for themselves or lease from developers under long contracts. The developer takes the land and construction risk; the tenant takes the demand risk. Speed to power is the competitive variable, and it is set by the two stages below this one, not by construction.
What to watch
- Build cost per megawatt, and how much of it is power and cooling.
- Months from ground-breaking to first power.
- Vacancy and pre-leasing in the main US markets.
- Rack density and the share of new capacity that is liquid-cooled.
- The largest announced campuses: sponsor, megawatts, power source, status.
How it connects
Megawatts of IT load, divided by the power one GPU and its share of the server draw, give the number of GPUs a site can hold. That conversion is the bridge from the physical world to the unit everyone counts.
Figures
| Metric | Value | Unit | As of | Scope | Confidence | Source | Note |
|---|---|---|---|---|---|---|---|
| us_primary_market_vacancy_rate | 1.4 | % of built colocation capacity | 2026-06 | US | high | CBRE, North America Data Center Trends H1 2026 (published 27 Aug 2026) | Share of built colocation capacity in CBRE's eight primary US markets (Northern Virginia, Dallas, Chicago, Phoenix, Atlanta, Silicon Valley, Hillsboro, New York Tri-State) not leased at mid-2026; a record low, down from 1.6% a year earlier; Northern Virginia alone is at 0.2%. |
| us_primary_market_under_construction | 7,481.1 | MW of colocation capacity | 2026-06 | US | high | CBRE, North America Data Center Trends H1 2026 (published 27 Aug 2026) | Colocation capacity under construction across CBRE's primary markets at mid-2026, a record; 80.4% of it was already preleased, so almost none of it will relieve vacancy when delivered. |
| us_primary_market_net_absorption_2025 | 2,497.6 | MW per year | 2025-12 | US | high | CBRE, North America Data Center Trends H2 2025 (published 25 Feb 2026) | Net new colocation capacity leased in primary markets in calendar 2025 (space newly occupied minus space vacated); H1 2026 added a further 1,456.2 MW (CBRE H1 2026), i.e. absorption is running at roughly 3 GW a year. |
| us_wholesale_colocation_asking_rate | 195.94 | USD per kW per month | 2025-12 | US | high | CBRE, North America Data Center Trends H2 2025 (published 25 Feb 2026) | Average asking rent for a 250-500 kW colocation requirement across primary markets at year-end 2025; this is the price signal for scarce built capacity, and CBRE reports a further 4.3% rise in H1 2026 without publishing the dollar figure. |
| industry_average_pue | 1.54 | ratio, total facility energy / IT energy | 2025-07 | global | high | Uptime Institute, Global Data Center Survey 2025 (keynote report PDF, survey run Apr-May 2025) | Weighted average of self-reported annual PUE from 681 operator respondents; the same headline figure for six years; facilities of 20 MW and above average 1.44 and those commissioned since about 2020 average 1.48, so new AI builds should be assumed well below 1.54. |
| hyperscale_fleet_pue_google | 1.09 | ratio, total facility energy / IT energy | 2025-12 | Google (global data-center fleet) | high | Google Data Centers, Power usage effectiveness (efficiency page) | Trailing-twelve-month PUE across Google's large-scale data centers in all seasons including all overhead; the best public fleet figure, so a practical floor for what a new AI campus can target (9% overhead versus 54% at the industry average). |
| build_cost_per_mw_global_average | 10.7 | USD million per MW of capacity, shell plus power and cooling, excluding IT | 2025-12 | global | medium | JLL Research, 2026 Global Data Center Outlook (5 Jan 2026) | JLL's average construction cost per MW for 2025, up from $7.7M in 2020; JLL forecasts $11.3M for 2026 (a forecast, not recorded as a value); the figure excludes servers, GPUs and networking, which cost several times more per MW. |
| build_cost_per_watt_us_silicon_valley | 13.3 | USD per watt of IT capacity ($1/W = $1M/MW) | 2025-11 | US (Silicon Valley; most expensive US market) | high | Turner & Townsend, Data centre construction cost index 2025 - Data centre cost trends (5 Nov 2025) | Turner & Townsend's indicative construction cost per watt for a traditional (air-cooled) data centre; other US markets on the same page: New Jersey $12.9, Portland $10.9, Atlanta $9.9, Phoenix $9.8, Columbus $9.8, Charlotte $9.5; AI-ready liquid-cooled builds cost 7-10% more and the index rose 5.5% in 2025. |
| gpu_rack_power_gb200_nvl72 | 120 | kW per rack | 2024-10 | NVIDIA GB200 NVL72 (72 Blackwell GPUs, liquid-cooled) | high | NVIDIA Technical Blog, NVIDIA Contributes NVIDIA GB200 NVL72 Designs to Open Compute Project (15 Oct 2024) | NVIDIA's own design figure for a liquid-cooled GB200 NVL72 rack, versus a typical enterprise rack of under 10 kW; Microsoft says its Fairwater GB200/GB300 halls run about 140 kW per rack and Vantage's Frontier campus is designed for 250 kW+ racks. |
| typical_rack_density_modal_average | 7.5 | kW per rack | 2025-07 | global | high | Uptime Institute, Global Data Center Survey 2025 (keynote report PDF, survey run Apr-May 2025) | Average of operators' most-common (modal) rack density excluding high-density outliers of 30 kW or more; including them the average is almost 9 kW; more than 80% of operators have no rack above 30 kW, which shows how far most existing halls are from GPU densities. |
| us_datacenter_share_of_electricity_2023 | 4.4 | % of total US electricity consumption | 2023-12 | US | high | Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report (Dec 2024) | LBNL's bottom-up estimate that US data centers consumed 176 TWh in 2023 (up from 76 TWh, 1.9%, in 2018); the report's 2028 projection range is not recorded here because the brief asks for recorded figures only. |
| us_datacenter_direct_water_use_2023 | 66 | billion liters per year (on-site consumption) | 2023-12 | US | medium | Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report (Dec 2024) | Modelled direct (on-site, mostly evaporative cooling) water consumption of US data centers in 2023; indirect water consumed through electricity generation was nearly 800 billion liters (4.52 L/kWh), so power supply, not cooling, is the larger water footprint. |
Campuses
| Campus | Sponsor | State | Planned (MW) | Operating (MW) | Power source | Status | Likelihood (proposed, 1 to 5) | Sources |
|---|---|---|---|---|---|---|---|---|
| Meta Hyperion | Meta | LA | 5,000 | – | new Entergy gas plants | under construction | 3 | planned |
| Microsoft Fairwater Wisconsin | Microsoft | WI | 2,263 | 369 | operating; expanding | 4 | planned · operating | |
| Amazon Project Rainier, New Carlisle | Amazon; Anthropic | IN | 2,250 | 910 | Indiana Michigan Power, under its large-load tariff | operating; expanding | 5 | planned · operating |
| xAI Colossus | xAI | TN | 2,000 | 946 | gas turbines at Southaven, Mississippi, for much of the load | operating; expanding | 3 | planned · operating |
| Stargate Abilene | OpenAI; Oracle | TX | 1,200 | 421 | operating; expanding | 5 | planned · operating | |
| Microsoft Fairwater Atlanta | Microsoft | GA | – | 636 | operating | 4 | operating |
6 campuses · 12,713 MW planned · 3,282 MW operating
| Campus | Criterion | Evidence | As of | Confidence | Source |
|---|---|---|---|---|---|
| Meta Hyperion | Power secured | The Louisiana PSC approved a settlement (Docket U-37425) on 20 August 2025 by a 4-1 vote letting Entergy Louisiana build three combined-cycle gas plants and new transmission for the Meta campus. | 2025-08-20 | high | KNOE (Monroe, LA), 'Louisiana Public Service Commission approves settlement for Meta Data Center gas plant', 2025-08-20 |
| Meta Hyperion | Power secured | Entergy Louisiana broke ground on the two Richland Parish Franklin Farms combined-cycle units (~1,500 MW) on 1 December 2025, with both expected operational by late 2028. | 2025-12-01 | high | Entergy Louisiana newsroom, 'Entergy Louisiana breaks ground on new state-of-the-art generation facilities', 2025-12-01 |
| Meta Hyperion | Power secured | A second Entergy application (Docket U-37882) for seven more gas plants totalling 5,278 MW to serve the expanded 5 GW campus was fast-tracked 4-1 on 15 April 2026 and is not yet approved; the final vote is set for 18 November 2026. | 2026-04-21 | medium | WWNO, 'Advocates cite lack of transparency, rising costs as regulators fast-track second Meta data center', 2026-04-21 |
| Meta Hyperion | Interconnection | Entergy's approved plan includes a 100-mile 500 kV project, the 55-acre Rayville Smalling 500/230 kV substation, two utility substations, six customer-owned substations and eight new 230 kV lines to deliver power to the Meta site; none is reported energised. | 2025-08-27 | medium | Industrial Info Resources, 'Louisiana Greenlights Massive Power Projects for Meta Data Center', 2025-08-27 |
| Meta Hyperion | Permits | LDEQ issued the campus (as Laidley LLC) a minor-source Clean Air Act permit in July 2025 for 31 emergency diesel generators and approved an amendment on 6 June 2026 adding 44 gas boilers; the Sierra Club contends it should be a major-source permit. | 2026-08-04 | medium | Louisiana Illuminator, 'Meta data center pollution permit doesn't account for all its emissions, Sierra Club says', 2026-08-04 |
| Meta Hyperion | Permits | Emissions from Entergy's ten gas turbines are permitted separately by the utility, not under the data centre's permit. | 2026-08-04 | medium | Louisiana Illuminator, 2026-08-04 |
| Meta Hyperion | Water | Local officials voice concern about water but no constraint or permit action is documented. | 2026-05-12 | low | WAFB, 'Meta's $27 billion AI data center is transforming rural Louisiana', 2026-05-12 |
| Meta Hyperion | Water | Meta pledges to return 100% of the facility's water consumption to local watersheds; no water withdrawal permit or source has been published. | 2026-07-13 | high | Louisiana Economic Development news release, 2026-07-13 |
| Meta Hyperion | Construction | Site work moved from dirt and access roads to active construction in early 2026; the Entergy plants are due late 2028 and the data centre is not expected operational before 2028-2030. | 2026-05-12 | medium | WAFB, 2026-05-12 |
| Meta Hyperion | Construction | Meta expanded the campus to nearly 10 million square feet and 5 GW of IT capacity, with 7,500 peak construction jobs. | 2026-07-13 | high | Louisiana Economic Development news release, 2026-07-13 |
| Meta Hyperion | Financing | Meta and Blue Owl formed a joint venture (Blue Owl 80%, Meta 20%) to fund approximately $27 billion of development costs, with Meta leasing the campus on four-year initial terms and giving a 16-year residual value guarantee. | 2025-10-21 | high | Meta newsroom, 'Meta Announces Joint Venture with Funds Managed by Blue Owl Capital to Develop Hyperion Data Center', 2025-10-21 |
| Meta Hyperion | Financing | Meta says it is paying $11 billion toward the ten-plant Entergy generation deal. | 2026-03-30 | medium | The Lens (New Orleans), 'Meta triples order of gas-fired power plants for its AI campus in Louisiana', 2026-03-30 |
| Microsoft Fairwater Wisconsin | Power secured | Microsoft commits to matching fossil-sourced consumption one-for-one with carbon-free energy, including a 250 MW Portage County solar project under construction, and says it pays the full cost of its energy. | 2025-09-18 | high | Microsoft On the Issues blog, 'Made in Wisconsin: The world's most powerful AI datacenter', 2025-09-18 |
| Microsoft Fairwater Wisconsin | Power secured | The Public Service Commission of Wisconsin approved We Energies' Very Large Customer and Bespoke Resources tariffs on 24 April 2026 with a 15-year minimum term, a 100 MW threshold, and full (100%) cost responsibility for dedicated generation. | 2026-04-24 | high | BizTimes Milwaukee, 'PSC makes significant changes while approving We Energies data center electricity rates', 2026-04-24 |
| Microsoft Fairwater Wisconsin | Power secured | The PSC's modifications lowered the tariff threshold to 100 MW, made it mandatory above that size, and removed We Energies' option to allocate 25% of dedicated generation cost to other customers. | 2026-04-24 | medium | Clean Wisconsin press release via Urban Milwaukee, 'State Regulators Make Substantial Modifications to We Energies AI Data Center Tariff Plan', 2026-04-24 |
| Microsoft Fairwater Wisconsin | Power secured | Microsoft signed a 15-year electric service contract with We Energies for the Mount Pleasant campus on 16 July 2026 after the PSC declined to reopen the tariff case; We Energies says the PSC-approved rules govern where the contract conflicts. | 2026-08-06 | medium | Wisconsin Watch, 'We Energies says Microsoft won't be held to contract terms that conflict with state regulators' order', 2026-08-06 |
| Microsoft Fairwater Wisconsin | Interconnection | We Energies is building a $335 million distribution project including an on-property substation for the Microsoft campus, exempt from PSC approval under the 2017 Foxconn law, with Microsoft paying its share. | 2024-03-05 | medium | Wisconsin Public Radio, 'We Energies to spend $335M on Microsoft development in Mount Pleasant', 2024-03-05 |
| Microsoft Fairwater Wisconsin | Interconnection | American Transmission Co. and We Energies filed a transmission charge agreement for Microsoft's Mount Pleasant campus with the PSC in late July 2026, running 15 years after the last transmission facility needed to serve the project comes online; so transmission build-out for later phases is still in progress. | 2026-08-03 | medium | Wisconsin Public Radio via Urban Milwaukee, ATC/We Energies transmission charge agreement for Microsoft, 2026-08-03 |
| Microsoft Fairwater Wisconsin | Interconnection | The load is being served: Epoch estimates 369 MW of IT power operating (Building 1 online 16 April 2026), with a completed substation at Building 2 and further substations under construction for Building 3 and Area 2. | 2026-09-30 | medium | Epoch AI, AI Data Centers directory, Microsoft Fairwater Wisconsin (satellite-based estimate; page updated 2026-09-30) |
| Microsoft Fairwater Wisconsin | Permits | Civic Media reports the full 15-building development is expected to be valued at more than $13 billion and to keep crews busy for up to a decade. | 2026-01-29 | medium | Civic Media / WRJN, 'Mount Pleasant board approves plans for 15 new Microsoft data centers', 2026-01-29 |
| Microsoft Fairwater Wisconsin | Permits | The Mount Pleasant village board unanimously approved site plans for up to 15 additional Microsoft data centers (two campuses, nearly 9 million sq ft, valued above $13 billion) on 26 January 2026; no litigation is reported. | 2026-06-23 | medium | Wisconsin Public Radio, 'Microsoft finishes construction of Mount Pleasant data center', 2026-06-23 |
| Microsoft Fairwater Wisconsin | Water | Racine's water utility supplies the campus from Lake Michigan within the 7 MGD diversion the DNR approved for Foxconn in 2018; records show a peak of 234,000 gpd for phase 1 rising to 702,000 gpd (8.4 million gallons a year) in later phases. | 2025-09-17 | high | Wisconsin Public Radio, 'Microsoft data centers will use up to 8.4M gallons of water each year, records show', 2025-09-17 |
| Microsoft Fairwater Wisconsin | Water | Microsoft says more than 90% of the facility uses closed-loop liquid cooling that consumes no water in operation. | 2025-09-18 | high | Microsoft On the Issues blog, 2025-09-18 |
| Microsoft Fairwater Wisconsin | Construction | The first Fairwater building on the 315-acre site began coming online in April 2026 and employs about 550; the second, adjacent data center is due for completion by 2028. | 2026-06-23 | high | Wisconsin Public Radio, 'Microsoft finishes construction of Mount Pleasant data center', 2026-06-23 |
| Microsoft Fairwater Wisconsin | Construction | Epoch projects Building 2 operational by 20 February 2027 and Buildings 3-11 by 13 May 2028, taking IT power from 369 MW today to 2,263 MW in Q2 2028. | 2026-09-30 | medium | Epoch AI, AI Data Centers directory, Microsoft Fairwater Wisconsin (estimate; page updated 2026-09-30) |
| Microsoft Fairwater Wisconsin | Financing | Microsoft's stated commitment is an additional $4 billion over three years for the second data center, taking its Wisconsin total above $7 billion, funded from Microsoft's own balance sheet. | 2025-09-18 | high | Microsoft On the Issues blog, 2025-09-18 |
| Microsoft Fairwater Wisconsin | Financing | Epoch estimates $14.0 billion of capital deployed so far, rising to $85.7 billion by Q2 2028 if all eleven buildings are filled; this is a model, not a Microsoft figure. | 2026-09-30 | low | Epoch AI, AI Data Centers directory (estimate; page updated 2026-09-30) |
| Amazon Project Rainier, New Carlisle | Power secured | I&M filed the large-load settlement jointly with Amazon Web Services, Microsoft, Google, the OUCC, the Data Center Coalition and the Citizens Action Coalition on 22 November 2024. | 2024-11-22 | high | AEP / Indiana Michigan Power news release, 'I&M and Stakeholders File Large Load Settlement to Advance Grid Reliability and Support Economic Growth', 2024-11-22 |
| Amazon Project Rainier, New Carlisle | Power secured | The Indiana Utility Regulatory Commission approved I&M's large-load Tariff I.P. settlement (Cause No. 46097) on 19 February 2025; the record identifies AWS's New Carlisle project as a 2,250 MW load and the largest customer in I&M's history. | 2025-02-19 | high | Indiana Utility Regulatory Commission, Order in Cause No. 46097 (I&M Tariff I.P. modifications), p. 7, 2025-02-19 |
| Amazon Project Rainier, New Carlisle | Power secured | The approved tariff binds large-load customers (including Amazon, a settling party) to a minimum 12-year contract after a load ramp of up to five years and an 80% minimum monthly billing demand, with I&M serving them as standard Tariff I.P. customers. | 2025-02-19 | high | Indiana Utility Regulatory Commission, Order in Cause No. 46097, pp. 32, 49, 53, 2025-02-19 |
| Amazon Project Rainier, New Carlisle | Interconnection | I&M told local media in May 2025 that two dedicated substations are being built on the AWS campus and that I&M plans to roughly double its generation from about 4-4.5 GW to about 8 GW by 2030-2031 to serve the new loads. | 2025-05-08 | medium | ABC57 (South Bend), 'The Silicon Prairie: Power Worries', quoting Indiana Michigan Power, 2025-05-08 |
| Amazon Project Rainier, New Carlisle | Interconnection | The load is demonstrably served: Epoch estimates 910 MW of IT power operating at the campus as of September 2026, with buildings 1-16 operational and a further substation under construction at the Buildings 19-32 expansion area. | 2026-09-30 | medium | Epoch AI, AI Data Centers directory, Anthropic/Amazon New Carlisle (satellite-based estimate; page updated 2026-09-30) |
| Amazon Project Rainier, New Carlisle | Permits | IDEM notified Amazon on 8 June 2026 of two air-permit violations (oversized 1,600 kW backup generators installed against a 1,500 kW approval, missing Affidavit of Construction, failed exhaust-temperature sensors); no enforcement action was taken. | 2026-06-16 | medium | WVPE (NPR, Elkhart/South Bend), 'Amazon data center violates air permits in multiple ways', 2026-06-16 |
| Amazon Project Rainier, New Carlisle | Permits | Amazon's Section 401 water quality certification for the campus (utility pumping station, substations, roads; 4.88 acres of emergent wetland and 3,986 linear feet of stream permanently impacted) was a resubmittal still in public comment in July 2026, i.e. not yet issued. | 2026-07-10 | high | Indiana Department of Environmental Management, Section 401 Water Quality Certification public notice (2025-78-71-ENH-WQC-2), Amazon Data Services Inc., hearing 2026-07-08, comments extended to 2026-07-10 |
| Amazon Project Rainier, New Carlisle | Permits | IDEM issued Amazon Data Services a Part 70 (Title V) air operating permit on 1 May 2025 for the New Carlisle source; a significant modification adding two generators went to 30-day public comment on 16 September 2026 and is not yet issued. | 2026-09-16 | high | Indiana Department of Environmental Management, Notice of 30-Day Period for Public Comment, Significant Permit Modification No. 141-50312-00650, 2026-09-16 |
| Amazon Project Rainier, New Carlisle | Water | Residents' wells near the Amazon and GM sites went dry in May 2025 and the Indiana DNR opened an investigation; the area draws on the Kankakee aquifer, and Epoch reports the buildings use direct air cooling rather than external cooling infrastructure. | 2025-05-29 | low | WNDU (South Bend), 'Residents raise concerns after wells lose water near Amazon, GM projects in St. Joseph County', 2025-05-29; cooling quote from Epoch AI directory page (updated 2026-09-30) |
| Amazon Project Rainier, New Carlisle | Water | The St. Joseph County Drainage Board tabled Amazon's request to extend construction dewatering of up to 35 million gallons per day into the Niespodziany Ditch for another 365 days on 3 March 2026, pending more information. | 2026-03-03 | medium | WSBT (South Bend), 'Amazon's request to dewater millions of gallons from data center near New Carlisle tabled', 2026-03-03 |
| Amazon Project Rainier, New Carlisle | Construction | AWS activated Project Rainier on 29 October 2025 with nearly 500,000 Trainium2 chips across seven buildings on 1,200 acres, dedicated to Anthropic, with a stated plan for about 30 buildings. | 2025-10-29 | medium | Converge Digest, 'AWS activates Project Rainier data center campus', reporting AWS announcement, 2025-10-29 |
| Amazon Project Rainier, New Carlisle | Construction | Epoch's satellite tracking shows buildings 1-7 operational by June 2025, 8-16 by March 2026, 17-18 under construction in June 2026, and the Buildings 19-32 expansion projected operational 17 March 2028 with 1,925 MW IT by Q1 2028. | 2026-09-30 | medium | Epoch AI, AI Data Centers directory, Anthropic/Amazon New Carlisle (satellite-based estimate; page updated 2026-09-30) |
| Amazon Project Rainier, New Carlisle | Financing | Amazon's stated commitment is $11 billion, and St. Joseph County approved a 10-year 50% real-property abatement on 870 acres plus an 85% personal-property exemption worth about $1.9 billion net to the company (August 2024). | 2024-08-13 | medium | Inside Indiana Business, 'Incentives approved for $11 billion Amazon data center project', 2024-08-13 |
| Amazon Project Rainier, New Carlisle | Financing | Spend is on Amazon's balance sheet (no project finance reported); Epoch estimates $34.5 billion of capital cost to date rising to $72.9 billion by Q1 2028, far above the $11 billion headline. | 2026-09-30 | low | Epoch AI, AI Data Centers directory (estimate; page updated 2026-09-30) |
| xAI Colossus | Power secured | Colossus 2 (5420 Tulane Road) draws essentially all its power from behind-the-meter turbines at the former Duke Energy Southaven Power Plant site that xAI bought, estimated at ~1.3 GW across ~59 turbines, with Solaris Energy Infrastructure renting ~1,140 MW of them. | 2026-05-13 | medium | Cleanview, 'Behind-the-meter data centers' report, Colossus 1 & 2 profile (satellite imagery dated 2026-05-13) |
| xAI Colossus | Power secured | Colossus 1 (Paul Lowry Road) receives 150 MW of MLGW/TVA grid power with a second 150 MW requested, giving about 500 MW total with on-site generation. | 2026-05-13 | medium | Cleanview report, 2026 |
| xAI Colossus | Interconnection | MLGW states it has no new contracts at the Colossus 2 site and is not supplying power to Colossus 2; the site's general services agreement delivers about 0.5 MW. | 2025-07-22 | high | Memphis Light, Gas and Water, statement on xAI at 5420 Tulane Road, 2025-07-22 |
| xAI Colossus | Permits | MDEQ issued an air permit on 10 March 2026 for 41 permanent gas turbines (about 1.2 GW) at the Southaven site; roughly 20 turbines were already running before the permit. | 2026-03-10 | high | E&E News, 'xAI gets air permit for unauthorized gas turbines', 2026-03-10 |
| xAI Colossus | Permits | SELC appealed the MDEQ permit in April 2026, and under an August 2026 agreement with MDEQ the company will remove its 69 temporary mobile turbines, which SELC says can keep running until at least July 2027. | 2026-08-03 | medium | Memphis Flyer, 'Environmental groups call out SpacexAI and MDEQ over agreed order', 2026-08-03 |
| xAI Colossus | Water | The $80 million Colossus Water Recycling Plant (up to 13 million gallons/day of reclaimed water), broken ground October 2025, was put on indefinite pause in April 2026, leaving the campus drawing on the Memphis Sand Aquifer. | 2026-04-09 | medium | Action News 5 (Memphis), 'xAI is prioritizing other projects; water plant is still planned', 2026-04-09 |
| xAI Colossus | Water | Rep. Steve Cohen publicly urged xAI to resume the water recycling plant after its work stoppage. | 2026-04-10 | high | Office of Rep. Steve Cohen, press release, 2026-04-10 |
| xAI Colossus | Construction | Musk said Colossus 2 had to be finished and made stable before the water plant resumes, signalling the campus is still in build-out. | 2026-04-09 | medium | Action News 5 quoting Elon Musk on X, 2026-04-09 |
| xAI Colossus | Construction | Epoch estimates 946 MW of IT power operating at Colossus 2 and projects 1,531 MW by Q1 2027; Musk announced 'nearly 2 gigawatts' for Colossus 1 and 2 combined. | 2026-09-24 | medium | Cleanview report, 2026 (construction timeline; Epoch estimate for MW per stack/campuses.csv, updated 2026-09-24) |
| xAI Colossus | Financing | xAI raised a $20 billion round in October 2025 split into about $7.5 billion of equity and up to $12.5 billion of debt through an SPV that buys Nvidia GPUs for Colossus 2 and rents them to xAI for five years. | 2025-10-08 | medium | Free Malaysia Today (Bloomberg report), 'Nvidia to invest in Musk's xAI as part of US$20bil funding', 2025-10-08 |
| xAI Colossus | Financing | SpaceX acquired xAI on 2 February 2026 in an all-stock deal, so Colossus spend now sits on SpaceX's balance sheet; no campus-level capex figure is published. | 2026-02-02 | low | WSMV (Nashville), headline, 2026-02-02 |
| Stargate Abilene | Power secured | Lancium holds an approved 1.2 GW grid interconnection for the Abilene Clean Campus, supplemented by behind-the-meter batteries and solar. | 2025-03-18 | high | Crusoe newsroom, 'Crusoe Expands AI Data Center Campus in Abilene to 1.2 Gigawatts', 2025-03-18 |
| Stargate Abilene | Power secured | An initial 360 MW of on-site gas generation (GE Vernova LM2500XPRESS and Solar Titan 350 units) was completed in January 2026, with up to ~1,180 MW planned behind the meter. | 2026-04-13 | medium | Cleanview, 'Behind-the-meter data centers' report, Stargate Abilene profile (satellite imagery dated 2026-04-13) |
| Stargate Abilene | Interconnection | Crusoe states the Abilene site connects directly to the ERCOT grid. | 2026-04-13 | medium | Cleanview report quoting Crusoe statement (2026) |
| Stargate Abilene | Interconnection | The campus is served by a 200 MW substation (built 2022, expanded 2024) and a larger 1 GW substation that was approaching completion in January 2026. | 2026-09-30 | medium | Epoch AI, AI Data Centers directory, OpenAI Stargate Abilene (satellite-based; page updated 2026-09-30) |
| Stargate Abilene | Permits | TCEQ minor (permit-by-rule) authorisations cover 10 gas turbines and 62 diesel generators; a first major air permit application for 41 more turbines and 18 more generators was filed and is pending. | 2026-07-09 | medium | Floodlight News, 'This permit, common for dry cleaners, is now being used to build AI power plants', 2026-07-09 |
| Stargate Abilene | Permits | Crusoe says the permitted turbines will be used only for backup power despite being permitted for continuous operation. | 2026-07-09 | medium | Floodlight News, 2026-07-09 |
| Stargate Abilene | Water | Cooling is direct-to-chip liquid with a zero-evaporation closed loop that is filled once and recirculated; no water-rights permit or constraint has been reported. | 2025-03-18 | high | Crusoe newsroom, 2025-03-18 |
| Stargate Abilene | Water | Lancium and Crusoe executives told Abilene city leaders the loop is filled once with relatively no loss over the project life. | 2026-03-05 | medium | KTXS (Abilene), 'Construction on Stargate One data center campus set to run through early 2027', 2026-03-05 |
| Stargate Abilene | Construction | The eighth and final building topped out in November 2025; the first two buildings (200+ MW IT) have been operational for Oracle/OpenAI since September 2025. | 2025-11-18 | medium | W.Media, 'Crusoe tops off final AI data center in Abilene, Texas', 2025-11-18 |
| Stargate Abilene | Construction | Construction on the eight buildings is expected to run through Q1 2027, later than the mid-2026 energisation originally targeted for phase 2. | 2026-03-05 | medium | KTXS (Abilene), 2026-03-05 |
| Stargate Abilene | Construction | Epoch estimates 421 MW of IT power operating (buildings 1-4) with buildings 5-8 projected operational by 1 November 2026 and 843 MW IT by Q4 2026. | 2026-09-30 | medium | Epoch AI, AI Data Centers directory (estimate; page updated 2026-09-30) |
| Stargate Abilene | Financing | Total financing raised for the campus reached $11.6 billion under a $15 billion joint venture. | 2025-05-22 | medium | RCR Wireless, 'Crusoe secures $11.6bn for OpenAI's Stargate DC in Texas', 2025-05-22 |
| Stargate Abilene | Financing | A J.P. Morgan-led consortium provided a $7.1 billion construction loan to the Blue Owl/Crusoe/Primary Digital Infrastructure joint venture for phase 2 (six buildings, 1.2 GW total). | 2025-05-28 | medium | REBusiness Online, 'Newmark Arranges $7.1B Construction Loan for West Texas Data Center Project', 2025-05-28 |
| Microsoft Fairwater Atlanta | Power secured | Microsoft designed Fairwater Atlanta around grid (utility) power with no on-site generation, UPS or dual-corded distribution for the GPU fleet, targeting 99.99% availability at the cost of 99.9%. | 2025-11-12 | high | Microsoft Official Blog, 'Infinite scale: The architecture behind the Azure AI superfactory', 2025-11-12 |
| Microsoft Fairwater Atlanta | Power secured | The Georgia PSC on 19 December 2025 certified about 9,885 MW of new Georgia Power generation (Dockets 56298 and 56310) to serve large-load growth, while the record showed executed large-load contracts were still a small share of projected demand. | 2025-12-19 | medium | Perkins Coie, 'Georgia Public Service Commission Approves Georgia Power's Major Generation Expansion', 2026-01-05 (PSC vote 2025-12-19) |
| Microsoft Fairwater Atlanta | Power secured | Georgia Power reported to the Georgia PSC that as of 31 March 2026 it had 12,400 MW of large-load commitments (contracts plus requests for service) across 31 customers, of which 8,500 MW had broken ground and ten projects were online; it does not name Microsoft or QTS, so the Fayetteville campus's specific agreement is not public. | 2026-03-31 | medium | Georgia Power Company, Q1 2026 Large Load Economic Development Report filed with the Georgia Public Service Commission (public disclosure version), as of 2026-03-31 |
| Microsoft Fairwater Atlanta | Interconnection | The load is served: Epoch estimates 636 MW of IT power operating across four buildings (Buildings 1-2 from 14 October 2025, Buildings 3-4 from 14 June 2026) with about 795 MW of air-cooled chiller capacity; no utility statement on the substation is public. | 2026-09-30 | medium | Epoch AI, AI Data Centers directory, Microsoft Fairwater Atlanta (satellite-based estimate; page updated 2026-09-30) |
| Microsoft Fairwater Atlanta | Interconnection | Cleanview lists the four Fairwater Atlanta buildings in Fayette County as developed by QTS and operating, with Buildings 1 and 2 online in 2025 and Buildings 3 and 4 in 2026. | 2026-09-30 | medium | Cleanview data center tracker, 'Microsoft Fairwater Atlanta - Building 1', accessed 2026-09-30 |
| Microsoft Fairwater Atlanta | Permits | Fayette County issued a public fact statement on 14 May 2026 clarifying that a smart-meter changeover, not unauthorised connections, caused the QTS water-billing gap, and that QTS paid the back bill at the construction rate. | 2026-05-14 | high | Fayette County Government, 'County Releases Facts About QTS Data Center Campus Project' (Fayette Newsnet), 2026-05-14 |
| Microsoft Fairwater Atlanta | Permits | The campus is the QTS 'Project Excalibur' development on about 600 acres off Georgia Highway 54 in Fayetteville, Fayette County, currently 13 buildings totalling about 6.2 million sq ft with up to 16 at full build-out; Fayette County publicly defends the project and no zoning litigation is reported. | 2026-07-13 | medium | Atlanta News First, 'Records raise questions about Georgia's largest data center and its water use', 2026-07-13 |
| Microsoft Fairwater Atlanta | Water | Water comes from Fayette County Water System (surface water, no wells), permitted for 22.8 MGD and producing about 17.3 MGD; QTS's monthly use is under 1% of that, and the county says the project does not affect residential pressure. | 2026-05-14 | high | Fayette County Government, 'County Releases Facts About QTS Data Center Campus Project', 2026-05-14 |
| Microsoft Fairwater Atlanta | Water | County records show about 109 million gallons billed to QTS across 13 lines during construction, with operational use projected at 9,000-12,000 gallons a month via closed-loop cooling; Microsoft says the closed loop's initial fill equals 20 homes' annual use with no evaporation. | 2026-07-13 | medium | Atlanta News First, 2026-07-13; second clause from Microsoft Official Blog, 2025-11-12 (https://blogs.microsoft.com/blog/2025/11/12/infinite-scale-the-architecture-behind-the-azure-ai-superfactory/) |
| Microsoft Fairwater Atlanta | Construction | Microsoft announced Fairwater Atlanta operational on 12 November 2025 with GB200/GB300 racks at about 140 kW each linked to the Wisconsin site; Epoch shows four buildings and 636 MW operating by June 2026 with no published expansion date. | 2025-11-12 | high | Microsoft Official Blog, 2025-11-12 |
| Microsoft Fairwater Atlanta | Construction | The wider QTS campus is not expected to finish construction until 2030, so further Microsoft phases beyond the four operating buildings are possible but undisclosed. | 2026-07-13 | medium | Atlanta News First, 2026-07-13 |
| Microsoft Fairwater Atlanta | Financing | The buildings are developed and owned by QTS (Blackstone) and leased to Microsoft, so Microsoft's spend is lease-based and not disclosed; Epoch estimates $24.1 billion of capital in the operating IT load. | 2026-09-30 | low | Cleanview (developer attribution, accessed 2026-09-30) and Epoch AI directory (capital estimate; page updated 2026-09-30) |
Conversions
| From → To | Factor | Unit | As of | Confidence | Source | Note |
|---|---|---|---|---|---|---|
| Grid, interconnection and transformers → Data centre shell and cooling | 0.65 | MW of IT load per MW of facility power drawn (= 1/PUE) at the industry-average PUE | 2025-07 | high | Uptime Institute, Global Data Center Survey 2025 (keynote report PDF, survey run Apr-May 2025) | 1 / 1.54 = 0.649, so a 100 MW grid connection at the industry-average PUE supports about 65 MW of IT load; use this for older or colocation halls, not for new hyperscale AI builds. |
| Data centre shell and cooling → GPUs, servers, racks and networking | 120 | kW of IT load per liquid-cooled GB200 NVL72 rack | 2024-10 | high | NVIDIA Technical Blog, NVIDIA Contributes NVIDIA GB200 NVL72 Designs to Open Compute Project (15 Oct 2024) | Design figure; 1 MW of IT load supports about 8 such racks (1,000 / 120 = 8.3); Microsoft reports ~140 kW per rack in production for GB200/GB300, which gives about 7 racks per MW. |
| Data centre shell and cooling → Data centre shell and cooling | 0.36 | liters of on-site water consumed per kWh of data-center electricity (WUE, US average) | 2023-12 | medium | Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report (Dec 2024) | LBNL modelled US average site WUE through 2023; a 100 MW facility at PUE 1.3 and 0.36 L/kWh would consume about 100,000 kW x 1.3 x 8,760 h x 0.36 L = 410 million liters a year; AWS reports a fleet WUE of 0.15 L/kWh and closed-loop designs (Microsoft Fairwater, Crusoe Abilene) approach zero evaporation; indirect water via power generation is far larger at 4.52 L/kWh. |
| Grid, interconnection and transformers → Data centre shell and cooling | 0.92 | MW of IT load per MW of facility power drawn (= 1/PUE) at Google's fleet PUE | 2025-12 | high | Google Data Centers, Power usage effectiveness (efficiency page) | 1 / 1.09 = 0.917, so a 100 MW connection supports about 92 MW of IT at best-in-class efficiency; new liquid-cooled AI campuses (Microsoft Fairwater, Crusoe Abilene) are designed toward this end of the range. |
| Grid, interconnection and transformers → Data centre shell and cooling | 10.7 | USD million of construction cost per MW of capacity (shell, power and cooling; excludes IT) | 2025-12 | medium | JLL Research, 2026 Global Data Center Outlook (5 Jan 2026) | Global average for 2025; US markets range from $9.5/W (Charlotte) to $13.3/W (Silicon Valley) in Turner & Townsend's index, i.e. $9.5M-$13.3M per MW, with a 7-10% premium for AI-ready liquid-cooled halls; a 1 GW campus is therefore roughly $10-13B before any GPUs. |
| Data centre shell and cooling → GPUs, servers, racks and networking | 72 | GPUs per GB200 NVL72 rack | 2024-10 | high | NVIDIA, GB200 NVL72 product page | 72 Blackwell GPUs plus 36 Grace CPUs per rack; at 120 kW per rack that is 120 / 72 = 1.67 kW per GPU at the rack (including CPUs, NVLink switches and fans but excluding facility cooling and power losses, which the PUE factor adds). |
| Data centre shell and cooling → GPUs, servers, racks and networking | 479 | Blackwell GPUs per MW of campus IT load, all-in (compute, storage, networking) | 2026-09-24 | medium | Epoch AI, AI Data Centers directory (estimates; page updated 24 Sep 2026) | Arithmetic from Epoch's Abilene estimate: 100.8k B200 + 100.8k B300 = 201,600 GPUs / 421 MW IT = 479 GPUs per MW, i.e. 2.09 kW of IT load per GPU; Epoch's Colossus 2 figures give a similar 440,000 / 946 = 465 per MW; SemiAnalysis counted ~110k GB200 for ~200 MW at Colossus 2 (550 per MW); these are tracker estimates, not operator disclosures. |
| Data centre shell and cooling → Data centre shell and cooling | 0.86 | acres of campus land per MW of capacity (Vantage Frontier design) | 2025-08-19 | medium | Vantage Data Centers press release, Vantage Data Centers Unveils Plans for Frontier, a $25B Mega-campus in Texas (19 Aug 2025) | Arithmetic: 1,200 acres / 1,400 MW = 0.86 acres per MW, with 3.7 million sq ft of buildings for 1.4 GW (2,640 sq ft per MW); Meta Hyperion is nearly 10 million sq ft for 5 GW (2,000 sq ft per MW); land is rarely the binding constraint, power and permits are. |
| Grid, interconnection and transformers → Data centre shell and cooling | 0.65 | MW of IT load per MW of facility power (1 / PUE, global weighted-average PUE 1.54) | 2025 | high | Uptime Institute, Global Data Center Survey 2025 (Keynote Report 180, July 2025) | 1 / 1.54 = 0.649 (my arithmetic): a data centre drawing 100 MW at the grid connection supports about 65 MW of servers and networking at the industry-average PUE; the figure has barely moved for six years. |
| Grid, interconnection and transformers → Data centre shell and cooling | 0.69 | MW of IT load per MW of facility power (1 / PUE, data centres of 20 MW and above, PUE 1.44) | 2025 | high | Uptime Institute, Global Data Center Survey 2025 (Keynote Report 180, July 2025) | 1 / 1.44 = 0.694 (my arithmetic); large, newer facilities of the kind hyperscalers build are more efficient, so a 1 GW grid connection yields roughly 690 MW of IT load rather than 650 MW; facilities commissioned since about 2020 average PUE 1.48. |
| Grid, interconnection and transformers → Data centre shell and cooling | 9 | kW per rack (average of modal rack densities, 2025 survey) | 2025 | high | Uptime Institute, Global Data Center Survey 2025 (Keynote Report 180, July 2025) | Typical rack density across all surveyed data centres, so 1 MW of IT load is about 111 conventional racks (1,000 / 9, my arithmetic); AI racks are far denser (an NVIDIA GB200 NVL72 rack is rated around 120 kW), which is why AI halls need liquid cooling and far more power per square foot. |
| Wafers and advanced packaging → Data centre shell and cooling | 1.79 | kW per GPU, all-in at the server level (DGX B200, 8 GPUs) | 2026-09 | high | NVIDIA, DGX B200 product page, specifications table | Arithmetic: the 8-GPU DGX B200 system draws about 14.3 kW maximum, so 14.3 / 8 = 1.79 kW per GPU including the two host CPUs, NICs, fans and power-supply losses; this is the number to use for IT load per GPU in an air-cooled Blackwell server, versus about 1.0 kW for the GPU chip alone. |
| Wafers and advanced packaging → Data centre shell and cooling | 120 | kW of IT load per GB200 NVL72 rack (72 GPUs), liquid-cooled | 2024-07-17 | medium | SemiAnalysis, GB200 Hardware Architecture - Component Supply Chain & BOM (July 17, 2024), free section | Rack-level power for the 72-GPU NVL72 (18 compute trays at a 6.3 kW max TDP each plus NVLink switch trays); arithmetic: 120 / 72 = 1.67 kW per GPU all-in, and a hosting provider (Pantheon, Aug. 2026) says NVIDIA's rack guide lists approximately 120 kW for GB200 NVL72 and up to 142 kW for GB300 NVL72. Requires direct liquid cooling; typical air-cooled H100 racks were about 40 kW. |
| GPUs, servers, racks and networking → Data centre shell and cooling | 72 | GPUs per NVL72 rack | 2026-09 | high | NVIDIA, GB200 NVL72 product page (specifications table) | One GB200 or GB300 NVL72 rack holds 72 Blackwell GPUs and 36 Grace CPUs across 18 compute trays (4 GPUs + 2 CPUs each) plus 9 NVLink switch trays; a Vera Rubin NVL72 rack keeps the same 72-GPU count. |
| GPUs, servers, racks and networking → Data centre shell and cooling | 8 | GPUs per HGX/DGX 8-GPU server | 2026-09 | high | NVIDIA, DGX B200 product page (specifications table) | The HGX baseboard design used by every OEM 8-GPU server (Supermicro, Dell XE9680, HPE, etc.) carries eight GPUs; a DGX B200 occupies 10 rack units, so air-cooled racks typically hold 2 to 4 such servers (16 to 32 GPUs) against 72 in a liquid-cooled NVL72. |
| GPUs, servers, racks and networking → Data centre shell and cooling | 1.83 | kW per GPU at the rack (NVL72, rated rack power incl. CPUs, NVLink switches, in-rack CDU) | 2026-09 | high | Supermicro, 'Supermicro NVIDIA GB200 NVL72 Datasheet' (PDF), rack specifications table | Arithmetic: 132 kW rated rack power / 72 GPUs = 1.833 kW per GPU; using the datasheet's operating range of 125-135 kW gives 1.74-1.88 kW per GPU. Excludes scale-out network switches, storage and facility cooling overhead (PUE), which belong to the datacenter stage. |
| GPUs, servers, racks and networking → Data centre shell and cooling | 1.79 | kW per GPU at the server (HGX/DGX B200 8-GPU, max system power incl. CPUs and NICs) | 2026-09 | high | NVIDIA, DGX B200 product page (specifications table) | Arithmetic: 14.3 kW max system power / 8 GPUs = 1.79 kW per GPU, including the two CPUs, memory, storage and ConnectX-7/BlueField-3 NICs inside the server but not the external switches; per GPU this is close to the NVL72 figure despite the NVL72's far higher density per rack. |
| GPU-hours → Data centre shell and cooling | 1.67 | kW of rack IT load per GPU (GB200 NVL72: 120 kW rack / 72 GPUs) | 2024-10-15 | high | NVIDIA Technical Blog, 'NVIDIA Contributes NVIDIA GB200 NVL72 Designs to Open Compute Project' (15 Oct 2024) | Arithmetic: 120 kW / 72 GPUs = 1.667 kW per GPU at the rack (GPU, Grace CPU, NVLink switches, in-rack cooling). Excludes scale-out network, storage and facility overhead (PUE), which the datacenter stage adds. |
| GPU-hours → Data centre shell and cooling | 1.27 | kW of system power per GPU (DGX H100: 10.2 kW max / 8 GPUs) | 2026-09-27 | high | NVIDIA DGX H100/H200 User Guide, 'Introduction to the NVIDIA DGX H100/H200 Systems', hardware specifications | Arithmetic: 10.2 kW / 8 GPUs = 1.275 kW per GPU for an air-cooled H100 server at maximum system power ('10.2 kW max.' on the same spec table; H100 SXM TDP is up to 700 W, so the server roughly doubles the GPU's own draw). Excludes network switches and facility overhead. |
| Data centre shell and cooling → GPU-hours | 5,256,000 | GPU-hours per MW-year of rack IT load (GB200 density, 100% uptime) | 2024-10-15 | medium | NVIDIA Technical Blog, 'NVIDIA Contributes NVIDIA GB200 NVL72 Designs to Open Compute Project' (15 Oct 2024); arithmetic by this note | Arithmetic: 1,000 kW / 1.667 kW per GPU = 600 GPUs per MW of rack IT load; 600 x 8,760 hours = 5,256,000 GPU-hours per MW-year if every GPU is rentable every hour. Sellable hours are lower by utilisation (Oracle reported 97.9%) and by facility overhead: at PUE 1.2 the figure per MW of facility power is 4,380,000. At H100 density (1.275 kW per GPU) it is 784 GPUs and 6,870,588 GPU-hours per MW-year. |
| GPU-hours → Data centre shell and cooling | 34.27 | USD m of revenue per MW-year of rack IT load at $6.52 per GPU-hour (B200 on-demand median, GB200 density, 100% utilisation) | 2026-09 | medium | AIMultiple, 'Cloud GPU Rental Price Index' (Sep 2026) for the price; NVIDIA GB200 NVL72 blog for the density; arithmetic by this note | Arithmetic: 5,256,000 GPU-hours per MW-year x $6.52 = $34.27 million per MW-year, a ceiling using on-demand list prices at full utilisation. At H100 density and the $3.25 H100 median: 6,870,588 x $3.25 = $22.33 million. Contract-level cross-checks are lower: Nebius cites $20-25 million per MW of annual contract value and Together AI/HUMAIN expect over $5 billion on 250 MW (= $20 million per MW-year). |
| GPU-hours → Data centre shell and cooling | 20 | USD m of annual contract value per MW (low end of Nebius' $20-25 million range on recent large deals) | 2026-08-12 | high | Nebius Group N.V., Form 6-K Exhibit 99.2, Letter to shareholders Q2 2026 (12 Aug 2026) | A company-stated contract yield per megawatt for multi-year GPU capacity deals, the best public anchor for 'revenue per MW-year on contract'. Implied contract price: $20-25 million / 5,256,000 GPU-hours (GB200 density) = $3.81-4.76 per GPU-hour, below the $6.52 B200 on-demand median. High end 25. |
| GPU-hours → Data centre shell and cooling | 1.71 | USD m of lease revenue per MW-year paid by a GPU cloud to its data-centre landlord (Applied Digital portfolio) | 2026-05-31 | medium | Applied Digital Corporation, Form 10-K for the fiscal year ended May 31, 2026; arithmetic by this note | Arithmetic: $36.2 billion / 1,410 MW / 15 years = $1.71 million per MW-year. The landlord's take is under a tenth of the $20-25 million per MW-year a GPU cloud books on the same megawatt, which shows where the margin (and the GPU capex) sits. |
Players
| Company | Ticker | Role | Listed | Note |
|---|---|---|---|---|
| Equinix | NASDAQ:EQIX | Retail and interconnection colocation REIT | yes | Largest colocation operator by footprint with 280 data centers in 36 countries and 77 markets (April 2026 boilerplate), but its retail/interconnection model hosts little of the multi-hundred-MW AI training capacity. |
| Digital Realty | NYSE:DLR | Wholesale and hyperscale colocation REIT | yes | Largest carrier-neutral wholesale landlord with 300+ data centers in 55+ metros and 30+ countries; develops hyperscale campuses partly through a $7B joint venture with Blackstone. |
| Vantage Data Centers | private | Hyperscale campus developer (DigitalBridge/Silver Lake-backed) | no | Building the $25B+, 1.4 GW 'Frontier' campus on 1,200 acres in Shackelford County, TX, with 10 buildings totalling 3.7 million sq ft designed for 250 kW+ racks; first building due H2 2026. |
| QTS (Blackstone) | private (owned by Blackstone, NYSE:BX) | Hyperscale campus developer | no | Blackstone calls QTS the largest independent data center operator in the world; it is the vehicle for Blackstone's $25B+ Pennsylvania digital and energy infrastructure commitment (July 2025) and has a fully pre-leased multi-GW pipeline. |
| Crusoe | private | AI data-center developer and neocloud | no | Built the first two Stargate Abilene buildings (200+ MW IT) from a June 2024 groundbreaking to live service by Sep 2025; the campus is planned at 1.2 GW, plus a separate 900 MW Microsoft campus announced Mar 2026 (2.1 GW site total). |
| Applied Digital | NASDAQ:APLD | Bitcoin miner turned AI data-center landlord | yes | 400 MW of critical IT leased to CoreWeave at Polaris Forge 1 (Ellendale, ND) under ~15-year leases worth about $11B in contracted revenue; the campus is engineered to scale to 1 GW. |
| Core Scientific | NASDAQ:CORZ | Bitcoin miner converting sites to GPU hosting | yes | Contracted to deliver approximately 500 MW of critical IT load to CoreWeave by H2 2026 under 12-year hosting contracts with $8.7B of projected cumulative revenue. |
| IREN | NASDAQ:IREN | Bitcoin miner turned AI cloud and data-center operator | yes | Signed a ~$9.7B five-year contract (20% prepaid) to give Microsoft access to GB300 GPUs at its 750 MW Childress, TX campus, housed in new liquid-cooled halls totalling 200 MW of critical IT (Horizon 1-4). |
| Cipher Mining | NASDAQ:CIFR | Bitcoin miner turned AI hosting landlord | yes | Signed a ~$5.5B, 15-year lease with AWS for 300 MW delivered in two phases in 2026, on top of Fluidstack leases backstopped by Google; AI hosting contracts total about $8.5B. |
| Vertiv | NYSE:VRT | Power and thermal equipment (UPS, switchgear, CDUs, chillers) | yes | FY2025 net sales $10,229.9M with 26% organic growth, backlog $15.0B (+109%) and Q4 organic orders +252%; guides 2026 sales of $13.25-13.75B, the clearest public demand signal for data-center electrical and cooling gear. |
| Schneider Electric | EPA:SU | Electrical distribution and cooling (Motivair liquid cooling) | yes | Took a controlling interest in liquid-cooling specialist Motivair in Feb 2025 and now sells coolant distribution units from 105 kW to 2.5 MW alongside switchgear and UPS, positioning as 'grid to chip'. |
| Eaton | NYSE:ETN | Electrical equipment; liquid cooling via Boyd Thermal | yes | Agreed in Nov 2025 to buy Boyd Thermal for $9.5B (Boyd expected 2026 sales $1.7B, $1.5B of it liquid cooling), closing expected Q2 2026; Eaton's own release timed out for the fetcher, so a secondary copy is cited. |
Signals
| Date | Kind | Actor | Counterparty | Claim | Value | Confidence | Source |
|---|---|---|---|---|---|---|---|
| 2026-08-27 | statement | CBRE | US primary colocation markets at mid-2026: a record 7,481 MW under construction, 80.4% of it preleased, and a record-low vacancy of 1.4% | 7,481 MW | reported | CBRE, North America Data Center Trends H1 2026 (2026-08-27) | |
| 2026-08-11 | buildout | CoreWeave | Active power reached 1.5 GW and total contracted power about 3.7 GW at June 30, 2026; nearly 500 MW went live in the quarter | 1.5 GW | confirmed | CoreWeave Q2 2026 earnings release (8-K exhibit 99.1, 2026-08-11) | |
| 2026-05-07 | statement | CoreWeave | Chief executive: surpassed 1 GW of active power and on the way to more than 8 GW by 2030 | 8 GW | confirmed | CoreWeave Q1 2026 earnings release (8-K exhibit 99.1, 2026-05-07) |