FedgeOps Field Guide · 2026.07

Data centers of
the world,
the companies behind them,
and what they cost the planet.

There are roughly 12,000+ data centers on Earth, half of them in the United States. Four companies — Amazon, Microsoft, Google, and Meta — are on track to spend $370 billion on them this year. Below is a full map, a country-by-country ranking, a state-by-state breakdown, an LLM-by-LLM tie-up, and the water, power, and square-footage reality of the AI build-out.

Researched 2026-07-27 Source: Synergy Research · Cargoson · Mappr · Epoch AI · IEA · LBNL · US Data Map v1.0
01 / By the numbers

The shape of the boom.

Snapshots, not forecasts. Every number on this page links to its source at the bottom.

12,000+
Operational data centers · global
Approximately. The count includes hyperscale, colocation, enterprise, and edge sites.
5,427
United States
45% of the world total. More than the next nine countries combined.
+ ~3,000 new planned or under construction
122.2 GW
Installed IT power · global
Q1 2025. Roughly equivalent to 120 large nuclear reactors running flat-out for compute.
415 TWh
Annual electricity · 2024
≈ 1.5% of global electricity demand. IEA forecasts 945 TWh by 2030 (≈ 3%).
+128% by 2030
$527B
Global market size · 2025
Statista. Gartner puts data-center systems spend at $489.5B, up 46.8% YoY.
$370B
Hyperscaler capex · 2025
Amazon + Microsoft + Google + Meta. Up from $244B in 2024. ~$446B+ projected for 2026.
55 B L
Direct water use · US · 2023
Hyperscale + colocation. Indirect (power generation) adds ~800 B L/year.
1,189
Hyperscale data centers · global
Q1 2025. Hyperscalers control 44% of global capacity — projected 61% by 2030.
USA 53.7 GW (44%) Europe 20.8 GW (17%) China 19.6 GW (16%) APAC ex-CN 16.9 GW (14%) LatAm 6.1 GW (5%) Other 5.1 GW (4%)
02 / Where they live

A world map of compute.

Hover any country for its data-center count. The choropleth uses log-scaled color so a 5-facility country doesn't vanish next to the U.S.

Data centers by country

SOURCE: Cargoson / Cloudscene · n ≈ 12,000 · Nov 2025
Few / none 10–99 100–299 300–599 600+
03 / Top 25 countries

The US is its own continent.

If you stack the next nine countries against the US, you still don't reach the US count.

Top 25 countries · data centers

SOURCE: Cargoson · Statista · Straits · Nov 2025

Hyperscale data centers

SOURCE: Synergy Research

Hyperscale facilities (≥10,000 sq ft, >1 MW) account for 44% of global capacity, up from 20% in 2017. Projected 61% by 2030.

Global IT power · GW

SOURCE: Synergy · Q1 2025
  • United States · 53.7 GW (44%)
  • Europe · 20.8 GW (17%)
  • China · 19.6 GW (16%)
  • APAC ex-CN · 16.9 GW (14%)
  • Latin America · 6.1 GW (5%)
  • Other · 5.1 GW (4%)

The US, Europe, and China together control 77% of all installed IT capacity on Earth.

04 / United States

5,427 reasons the US is the AI capital.

Hyperscaler self-builds + colocation + sovereign-cloud build-outs. The 8-state top-8 holds 51% of US facilities.

Top US states · data centers

SOURCE: USDataMap · Jul 2026

Northern Virginia — known as "Data Center Alley" — alone holds ~1,020 facilities within a 20-mile radius, more than every country on Earth except the US itself.

Top US operators

SOURCE: USDataMap · facilities

Equinix and Digital Realty are pure colocation. Google, Meta, Microsoft, and AWS each self-build hyperscale campuses and additionally lease capacity from those two. The hyperscalers are scaling both.

US hyperscale build-out · selected mega-projects

SOURCE: FT · TechCrunch · Epoch AI · company filings
Project Owner / Anchor tenant Location Scale Capex
Stargate InitiativeOpenAI + Oracle + SoftBankMultiple US sites5+ GW planned$500B
Meta HyperionMetaRichland Parish, LA5 GW · 2,250 acres$10B+
Anthropic-Amazon New CarlisleAmazon (anchor: Anthropic)New Carlisle, IN~2.2 GW · 1.5M sq ft$11B (Project Rainier)
xAI Colossus 2xAIMemphis, TN (+ MS)1.2+ GW · 1,112k H100-eq~$10B est.
Microsoft Fairwater AtlantaMicrosoft (OpenAI as user)Fayetteville, GA769k H100-equndisclosed
Meta PrometheusMetaNew Albany, OH763k H100-equndisclosed
OpenAI Stargate AbileneCrusoe + Oracle + OpenAIAbilene, TX1.2 GW · 509k H100-eq~$40B
Microsoft Fairwater WisconsinMicrosoftMount Pleasant, WI446k H100-eq~$3.3B
Google ColumbusGoogleColumbus, OH409k H100-equndisclosed
AWS Georgia expansionAmazonGeorgia, USA11+ data centers$35B
AWS Mississippi complexAmazon (anchor: Anthropic)Mississippi, USA2 sites · Project Rainier$10B
Aligned Data CentersBlackRock consortium50+ US campuses$40B acquisition

Compute capacity is given in H100-equivalent GPUs as estimated by Epoch AI from satellite imagery, permits, and public documents. Real performance is typically 20–50% of theoretical peak.

05 / China

Where does China actually train these models?

The open-weight Chinese models land on Hugging Face, but where they were trained is mostly a black box. Here is the most honest picture I can draw from public sources.

Why so little is known. In 2023 China tightened the rules on disclosing AI training compute (Cyberspace Administration's "Generative AI Interim Measures"). Labs now self-report headline specs, not where the GPUs live. Most of what follows is a synthesis of academic papers, IPO filings, government renewable-energy auctions, and land-use permits — not lab announcements.

Three patterns

how Chinese labs actually get compute
A
Rent from a Chinese hyperscaler. The default for almost every mid-tier lab. Alibaba Cloud, Tencent Cloud, Baidu Cloud, and Huawei Cloud sell GPU-hours as a service. The lab never owns the racks and usually doesn't pick the building.
B
Spin up inside the parent company's cluster. When the lab is owned by a larger Chinese tech or finance firm, training happens in the parent's existing facility. DeepSeek → High-Flyer's Fire-Flyer 2 cluster in Hangzhou. Moonshot → Alibaba's rented capacity (Alibaba owns 36%).
C
Sovereign "Western data" regions. The big hyperscalers all build their mega-campuses in Inner Mongolia (Hohhot), Guizhou (Guiyang), or Ningxia (Zhongwei). Cold climate, cheap coal-and-wind power, government-owned land. The address isn't published; you know it exists because of the electricity-utility contracts.

Where the chips are

post-export-control silicon mix

After the U.S. banned the H100 in 2022 and the H20 in 2024, China's frontier training runs on a constrained mix. DeepSeek trained V3 on 2,048 Nvidia H800s (a sanctioned China-export chip); newer work is shifting to Huawei Ascend 910B/910C. Domestic supply is the bottleneck, not power or land.

9 open-weight Chinese models — and where they likely trained

model · lab · pattern · best evidence · disclosure
Model Lab How they get compute Where Best public evidence
DeepSeek V3 / R1 DeepSeek (Hangzhou) Pattern B · parent cluster Hangzhou, Zhejiang 2,048 H800s · 671B MoE · V3 paper cites Fire-Flyer lineage
Qwen3 / Qwen3-Max Alibaba (Qwen team) Pattern A · self-hosted Alibaba Hangzhou-Yuhang + Beijing-Tongzhou 235B-A22B MoE · 100+ open-weight releases · 40M downloads
Kimi K2 / K2.5 Moonshot AI (Beijing) Pattern A · colocation Beijing (racks in Alibaba + smaller landlords) 1T params · 32B active · trained on 15.5T tokens · MoE
GLM-4.5 / 4.6 Zhipu AI (Beijing) Pattern A · colocation Beijing colocation · undisclosed Open-weight MoE · competitive with Claude 4 Sonnet on some benchmarks
Doubao 1.5 Pro ByteDance Volcano Engine Pattern C · self-build Inner Mongolia & Guizhou (Doubao + TikTok) Closed-weight · estimated 7–10 GW self-build by 2027
ERNIE 4.5 / 4.5 Turbo Baidu Pattern C · self-build Yangquan (Shanxi) + Baoding (Hebei) Closed-weight · sovereign enterprise cloud focus · Kunlun 2 silicon
Hunyuan Turbo / T1 Tencent Cloud Pattern A · self-hosted Shenzhen + Tianjin + Shanghai Closed-weight · Mamba-style hybrid architecture
Pangu / MindSpore Huawei Cloud Pattern C · self-build Guizhou + Dongguan + Inner Mongolia Ascend 910B/910C silicon · only credible Nvidia competitor at scale
MiniMax-M2 / Yi-Lightning 01.AI (Beijing) Pattern A · colocation Beijing (Alibaba + smaller) Yi-Lightning closed-weight · M2 open-weight · founder Kai-Fu Lee

Disclosure rule of thumb: Alibaba, ByteDance, Tencent, Baidu, Huawei control a known facility (Pattern A or C). The frontier research labs — DeepSeek, Moonshot, Zhipu, 01.AI — are overwhelmingly Pattern A or B: they rent racks in hyperscaler buildings and don't own the addresses.

The unspoken geography: Inner Mongolia

where the compute actually lives

If you're tracking where AI training happens in China, watch three provincial names: Inner Mongolia (Hohhot), Guizhou (Guiyang), and Ningxia (Zhongwei). These are China's "national data center clusters" — designated by MIIT since 2020 specifically to absorb hyperscale compute.

  • Cold air: mean winter temps of −10°C to −25°C cut cooling OPEX 30–40% vs. coastal sites.
  • Wind + coal: Inner Mongolia alone generated 13% of China's wind power in 2025; Ningxia is on the coal grid. Power is ~$0.04/kWh vs. ~$0.08 in Beijing.
  • Land: state-owned, large contiguous parcels. Alibaba's Hohhot campus is on a 1,500-acre parcel — comparable to Meta's Hyperion.
  • Water: scarce. The trade-off is the same one US western sites face, just hidden behind national-grid pricing.
  • Disclosure: the GPU counts and exact tenants of these sites are not public. Investors and journalists track them via local utility renewable-purchase agreements, not press releases.
Hohhot · Inner Mongolia Alibaba · Tencent · ByteDance Guiyang · Guizhou Huawei · Apple (iCloud) · Tencent Zhongwei · Ningxia Amazon (CDN) · Alibaba · MIIT pilot Beijing Moonshot · Zhipu · 01.AI Hangzhou DeepSeek · High-Flyer · Qwen Western data corridor (Pattern C) Coastal lab clusters (Pattern A/B)
06 / Who owns what

The LLM-to-data-center tie-up.

Most frontier AI labs don't own their data centers. They sign exclusive compute deals with the four big hyperscalers — and increasingly, with Oracle, Crusoe, and CoreWeave.

$125B · 2025

Amazon Web Services

Anchor: Anthropic
Self-built regions36 globally
Active AI sites28 in US
Anchor customerAnthropic (Project Rainier)
2026 capex~$126–130B

$8B in Anthropic + Trainium silicon deal. Operates Project Rainier (New Carlisle, IN) and Mississippi campus for Anthropic workloads. 1.5M+ sq ft.

$88.7B · FY25

Microsoft Azure

Anchor: OpenAI (weakening)
Self-built regions60+ globally
Active AI sites32 in US
Anchor customerOpenAI (Fairwater ATL/WI)
2026 capex$120B+ (FY26)

Originally exclusive OpenAI host. Fairwater Atlanta (769k H100-eq) and Fairwater Wisconsin (446k H100-eq) anchor the AI footprint. FY26 capex ~$120B+ including Stargate co-funding.

$91–93B · 2025

Google Cloud

Anchor: Gemini / internal
Self-built regions40 globally
Active AI sites35 in US
Anchor customerGemini · Anthropic (TPUs)
2026 capex"significant increase"

Builds its own TPUs. Hosts Anthropic on TPU clusters via a multi-billion-dollar deal. Also hosts smaller AI startups (Lovable, Windsurf) as "primary computing partners."

$66–72B · 2025

Meta

Anchor: Llama · internal
Self-built sites33 in US
Mega-campusesPrometheus (OH) · Hyperion (LA)
Anchor customerLlama 4 · "superintelligence" labs
2026 capex~$100B

Zuckerberg committed $600B US infra through 2028. Hyperion (LA, 5 GW, nuclear-adjacent) and Prometheus (OH, 763k H100-eq) are the two new flagships. Also $10B/yr with Google Cloud.

$100B+ · 2025

Oracle Cloud

Anchor: OpenAI · xAI
OpenAI deal$300B · 5 yr · starts 2027
Stargate roleBuild partner
Fuel cells2.8 GW Bloom deal (1.2 GW live)
GPU supply~400,000 Nvidia GPUs @ Abilene

Rose to the top tier after the Stargate announcement. Operates Abilene site for OpenAI with Crusoe. Sept 2025: 5-year $300B deal with OpenAI for compute starting 2027.

~$10B · 2025

xAI

In-house · Memphis · on-site gas
Colossus 2 capacity1,112k H100-eq
SiteMemphis, TN (+ MS)
PowerOn-site Doosan + mobile Titan turbines
SpeedColossus 1 built in 122 days

First true gigawatt-scale data center in execution (SemiAnalysis). Uses behind-the-meter gas turbines to bypass 5–7 yr grid interconnection queues. NAACP lawsuit over unpermitted turbines in MS; DOJ intervened citing national security.

$14B · 2025

Anthropic

Anchor: AWS · Project Rainier
AWS deal$8B investment + Trainium
Capacity~686k H100-eq @ New Carlisle
StrategyTrainium 2 silicon, kernel-level co-design
Google dealMulti-$B TPU commitment

Anchor customer on AWS Project Rainier (Indiana + Mississippi). Also buys TPU capacity from Google. Trains Claude on a Trainium + Nvidia hybrid — co-designed kernels for cost efficiency.

$100B · 2025

OpenAI

Anchor: Microsoft · Oracle · Stargate
Microsoft deal~$14B since 2019 (now non-exclusive)
Oracle deal$300B · 2027–2032
Nvidia deal$100B GPU-for-stock (Sep 2025)
AMD dealGPU-for-stock arrangement

No longer Microsoft-exclusive. Distributing compute across Microsoft (Fairwater), Oracle (Abilene + 5 more Stargate sites), Nvidia, and Crusoe-built sites. The most distributed of any AI lab.

~$15B · 2025

CoreWeave

Neo-cloud · GPU-as-a-service
StrategyPure GPU colocation
Capacity~360k GPUs across 32 sites
PowerBloom fuel cells + Volo, IL pilot
Anchor customerMicrosoft (~$15B through 2030)

Built on crypto-mining foundations, became the first major "neo-cloud" pure-play GPU provider. Signed a $15B compute contract with Microsoft. Now adding Bloom Energy fuel cells at the Volo, IL site for behind-the-meter power.

07 / AI vs traditional

Two species of data center.

The split that's reshaping the entire industry. AI facilities are not just "data centers but bigger" — they're a fundamentally different build.

Dimension Traditional cloud / colocation AI / "AI factory"
Primary workload Multi-tenant SaaS, web, email, databases, video streaming Single-customer training runs or inference; one model, one company (sometimes one nation)
Hardware CPU racks · 8–32 kW per rack GPU racks · 60–130+ kW per rack (Nvidia Blackwell direct-to-chip liquid)
Site footprint ~70% IT room, 30% support · dense racks ~30% IT room, 70% support · power & cooling dominate
Power density ~1.0 kW / sq ft ~2.5–6.0 kW / sq ft (newer sites target 8+)
Cooling Air / chilled water loop Direct-to-chip liquid + rear-door heat exchangers + evaporative cooling towers (~19,000 L/min each)
Typical PUE 1.4 – 1.6 1.1 – 1.3 (Google claims 1.10 industry-leading)
Water use ~0.3 L / kWh (mild) ~1.0–2.0 L / kWh for evaporative (cooling-tower) sites
Tenant model Hundreds-to-thousands of customers sharing racks "AI factory" — often one tenant; Jensen Huang's term
Latency priority Distributed across metro regions Co-locate as many GPUs as possible on one campus to minimize inter-chip latency
Demand signal Proven — "proven model with proven returns" Speculative — "build it and they will come"
Capex per facility $200M – $1B $5B – $40B+ per campus

Cooling shift

Air cooling used to be the default. AI racks now run liquid-to-chip — cold plates bolted directly to the GPU die. Microsoft shifted the bulk of new builds to water cooling in 2024.

Behind-the-meter

Grid queues in the US are 5–7+ years. Hyperscalers increasingly bring their own power: gas turbines (xAI, Crusoe), fuel cells (Oracle/CoreWeave), even co-located nuclear (Meta Hyperion).

PUE matters

Power Usage Effectiveness = total facility power / IT power. AI sites have lower PUE because cooling is denser — but absolute power and water per site are dramatically higher.

08 / Footprint

Power, water, square footage.

The three numbers that decide whether a data center actually gets built.

122.2 GW
Installed IT power · global
Q1 2025. Doubling every 4–5 years.
415 TWh
Annual electricity · 2024
≈ 1.5% of global electricity. By 2030: ≈ 945 TWh / 3%.
55 B L
Direct water · US · 2023
LBNL. 42% of Microsoft's water came from water-stressed basins.
800 B L
Indirect water · US · 2023
Water embedded in the electricity grid that powers DCs. ≈ water use of 2M US homes.
19,000 L/min
Per cooling tower
Evaporative loss. A single 50 MW liquid-to-chip AI site may run 4–8 towers.
5 GW
Meta Hyperion footprint
Power capacity of one US campus — equal to ~5 large nuclear reactors, all on one 2,250-acre site.

Power consumption · 2024 vs 2030

SOURCE: IEA
Region 2024 (TWh) 2030 projected Growth
United States183426+133%
China65175+170%
Europe64109+70%
Japan1934+80%
Rest of world84201+140%
Global415945+128%
09 / Watch it

A 90-second walkthrough.

Built with Pexels stock footage and HyperFrames. Pexels free-to-use license; video is CC-BY-style shareable with credit.

Pexels + HyperFrames

Clips: Pexels / Mikael Blomkvist, Tom Fisk, Dima Krivoy, MrColo, Pressmaster, Pavel Danilyuk, Kuiyibo Campos. Data: Synergy Research · IEA · LBNL · Epoch AI · US Data Map · FT · TechCrunch · company filings.

10 / Sources

Everything I used.

If a number on this page matters to you, click through to the source.

Cargoson · Cloudscene
Country-by-country data center count, Nov 2025.cargoson.com/en/blog/...
Mappr
Interactive map of global data centers 2025.mappr.co/countries-with-most-data-centers
Straits Research
Top 10 countries by data center count, Mar 2025.straitsresearch.com/...
DCPulse
Top data center cities + growth hotspots 2025.dcpulse.com/article/...
USDataMap
800+ US facilities, owners, capacity, planned.usdatamap.com
Synergy Research Group
Hyperscale counts and global IT power by region (Q1 2025).srgresearch.com
IEA · Data Centres & Data Transmission Networks
415 TWh (2024) → 945 TWh (2030).iea.org/reports/data-centres...
LBNL · United States Data Center Energy Usage Report 2024
55 B L direct + 800 B L indirect water (US, 2023).eta-publications.lbl.gov/...
Epoch AI · AI Data Centers
Satellite + permit-tracked AI facility database (75 sites).epoch.ai/data/ai-data-centers
Financial Times · Inside the AI capacity race
Hyperscaler capex, Colossus 2, water story.ig.ft.com/ai-data-centres
TechCrunch · $1T+ AI infrastructure deals
Stargate, Microsoft, OpenAI, Oracle, Nvidia.techcrunch.com/2026/02/28/...
NextBigFuture · Major AI DC build projects 2026–2028
Cross-referenced hyperscaler tracker.nextbigfuture.com/2026/06/...
Statista · Data center market size 2025
$527.46B global market, 6.98% CAGR.statista.com/statistics/...
Visual Capitalist · World's Data Centers 2025
Innovation clusters + global DC distribution.visualcapitalist.com/...