01
United States
45%Share of global data-centre electricity use (IEA, 2024)
AI INFRASTRUCTURE EVIDENCE ATLAS / REVIEWED SEPTEMBER 2026
Who is investing
in AI infrastructure—
and what is real?
An independent guide to reported spending, company forecasts, public programmes and future commitments. It explains what each number includes, where it comes from and whether it describes an announcement, a project in progress or operating infrastructure.
Four company forecasts for 2026ILLUSTRATIVE PACE
$0.00B
Estimated amount by today if the annual forecasts were spread evenly across the year
Combined forecast$695B
Average per second$22,038
Amazon + Microsoft + Alphabet + Meta / company-wide forecasts, not verified AI spending / 00:00:00 local
01$500B all data centres investment in 2024
02$695B four company forecasts for 2026
03EUR 200B EU public-private investment target
04RMB 380B Alibaba company plan over three years
THE PURPOSE OF THIS ATLASINTERPRETATION, NOT ANOTHER DIRECTORY
The headline number is only the beginning.
AI infrastructure announcements mix different currencies, time periods and stages of development. This atlas does not claim to list every data centre. Its job is to make unlike claims easier to compare and to show where the evidence is incomplete.
01Separate actual investment from forecasts and targets
02Keep public funding distinct from private company spending
03Expose scope, overlap, uncertainty and original sources
01 / GLOBAL INVESTMENT
Comparing AI investment
requires careful definitions.
The International Energy Agency estimated approximately $500B of worldwide data-centre investment in 2024. This includes every type of data centre—not only facilities used for AI.
02
China
25%Share of global data-centre electricity use (IEA, 2024)
03
Europe
15%Share of global data-centre electricity use (IEA, 2024)
04
Rest of world
15%Share of global data-centre electricity use (IEA, 2024)
$500Ball data centres / 2024
US45%
CN25%
EU15%
ROW15%
COMPANY FORECASTS FOR 2026
Four companies expect to invest about $695B
Amazon, Microsoft, Alphabet and Meta operate worldwide. Their forecasts cover their complete businesses, so these amounts are neither US-only figures nor exact measures of AI spending.
Amazon$200B
AI, chips, robotics and satellites
VIEW COMPANY SOURCEMicrosoft$190BCalendar-year investment forecast
VIEW COMPANY SOURCEAlphabet$175-185BMostly technical infrastructure
VIEW COMPANY SOURCEMeta$115-135BAI efforts and core business
VIEW COMPANY SOURCEAmount announcedEUR 200M
Status and timeframeFully funded
Groningen AI Factory; government, region and EuroHPCVIEW ORIGINAL SOURCE02 / SELECTED INFRASTRUCTURE EXAMPLES
Follow projects from plan
to operation.
This is an evidence map, not a comprehensive directory. It contains the complete official set of 19 EuroHPC AI Factories plus 17 selected public and commercial examples. Choose a marker to inspect its reported status and source.
COMPLETE EUROHPC INVENTORY
Europe has 19 AI Factories—not one.
EuroHPC selected 19 AI Factories across 16 EU Member States in three rounds ending in October 2025. All 19 are included below and on this map. Their exact operating phases vary, so the shared map status is Selected / deploying unless a newer site-specific source confirms more.
01
LUMI AI FactoryKajaani, Finland
02
HammerHAIStuttgart, Germany
03
PharosAthens, Greece
04
IT4LIABologna, Italy
05
L-AI FactoryBissen, Luxembourg
06
BSC AI FactoryBarcelona, Spain
07
MIMERLinköping, Sweden
08
AI:ATVienna, Austria
09
BRAIN++Sofia, Bulgaria
10
AI2FFrance — decentralised network
11
JAIFJülich, Germany
12
PIAST AI FactoryPoznań, Poland
13
SLAIFMaribor, Slovenia
14
CZAIOstrava, Czechia
15
LitAI FactoryVilnius, Lithuania
16
NLAIF / Dutch AI FactoryGroningen, Netherlands
17
Gaia AI FactoryKraków, Poland
18
RO AI FactoryBucharest, Romania
19
1HealthAISantiago de Compostela, Spain
COVERAGE BOUNDARY
Only the 19 EuroHPC AI Factories form a complete programme-level set. Every other marker is a deliberately selected example and must not be used to compare facility counts between countries. For broader global coverage, use the specialist databases linked in the evidence section. Markers show approximate cities or country references, not exact entrances.
REVIEWED 28 SEPT 202603 / AI MODELS AND AGENTS
Compare leading AI models
and agent tools.
Select a model or tool to see who makes it, what it is designed for and how it can be accessed. “Open weights” means the model files can be downloaded, but the licence may still impose restrictions.
AI MODELAGENT OR TOOL
CLOSED ACCESSDOWNLOADABLE OR OPEN TOOLING
Closed/API Open weights Agent/framework
ClosedUS
GPT-6 Astra
Computer use, reasoning, professional work and tools
Rolling out through ChatGPT, the OpenAI API, Microsoft Azure and AWS BedrockOpenAI / VISIT WEBSITEClosedUS
Claude
Writing, analysis and coding
Available as a product and through an APIAnthropic / VISIT WEBSITEClosedUS
Gemini
Text, images, audio and video
Available as a product and through an APIGoogle DeepMind / VISIT WEBSITEClosedUS
Grok
General assistance and current information
Available as a product and through an APIxAI / VISIT WEBSITEOpen weightsUS
Llama
Downloadable general-purpose models
Downloadable files with a custom licenceMeta / VISIT WEBSITEOpen weightsUS
Gemma
Smaller models for local use
Downloadable files under Gemma's termsGoogle DeepMind / VISIT WEBSITEOpen weightsChina
Qwen
Languages, coding and problem solving
Downloadable files and a hosted APIAlibaba / VISIT WEBSITEOpen weightsChina
DeepSeek
Problem solving and coding
Downloadable files and a hosted APIDeepSeek / VISIT WEBSITEOpen weightsEurope
Mistral
Multilingual business and developer use
Downloadable files and a hosted APIMistral AI / VISIT WEBSITEOpen weightsChina
Kimi
Documents, research and coding
Downloadable files and a hosted productMoonshot AI / VISIT WEBSITEAgentUS
Codex
Software development
Hosted coding agentOpenAI / VISIT WEBSITEAgentUS
Claude Code
Software development from a terminal
Hosted agent and command-line toolAnthropic / VISIT WEBSITEAgentUS
Gemini CLI
Open command-line coding tool
Open-source command-line toolGoogle / VISIT WEBSITEAgentUS
Devin
Completing software tasks
Hosted coding agentCognition / VISIT WEBSITEAgentAsia
Manus
Research and online tasks
Hosted agentManus / VISIT WEBSITEAgentUS
LangGraph
Building agent workflows
Open-source development frameworkLangChain / VISIT WEBSITEAgentUS
Muse
Personal tasks, web actions and background work
Available in the US through web, iOS, Android and WhatsAppMeta / VISIT WEBSITE04 / CONTINUOUS MONITORING
A weekly radar for
new capital signals.
The harvesting layer watches official channels and selected international news outlets. A weekly AI research pass then checks original sources and updates the curated Atlas only when the evidence supports a material change.
Starting first collection
COLLECTION WINDOWEvery Monday at 06:15 UTC
LAST COMPLETEDAwaiting first run
SOURCE HEALTHChecking sources
The first weekly collection is being assembled.
Official feeds and selected publisher feeds are being checked. Newly found records will appear below with their original links and discovery status.
—priority signals
—official records
—new this run
MONITORING INPUT
Source signals considered by the weekly AI review
Appearing here means a report was collected, not that an Atlas statistic changed. The AI review checks material candidates against original evidence before updating the curated data.
COLLECTINGNo discovery records have been stored yet.
The first pass runs in the background. Return shortly to see source-linked findings.
MONITORED CHANNELS6 source groups
Government sources and publisher feeds are queried directly. Every discovery links to the originating publication.
POLITICO Europe · Artificial intelligenceNews outlet / European Union
Finextra · Artificial intelligenceNews outlet / Global
TechCrunch · Artificial intelligenceNews outlet / United States / Global
US Federal RegisterGovernment / United States
European Commission Press CornerGovernment / European Union
UK Government newsGovernment / United Kingdom
Automatic updates, evidence-gated
The AI research pass promotes a finding only when original sources support a material Atlas change. Ambiguous or conflicting reports leave the statistics unchanged.
Harvest Monday at 06:15 UTC; AI evidence review follows05 / EVIDENCE AND DEFINITIONS
Follow every claim
back to its source.
Actual investment, company forecasts, government programmes and future commitments are different kinds of information. They are kept separate and shown in their original currencies and time periods.
1
Use original sources
Prefer company filings, investor reports, government publications and intergovernmental research.
2
Separate spending from plans
A multi-year investment target is not presented as money that has already been spent.
3
Avoid double counting
Large programs may include investments announced elsewhere, so related totals are not automatically added together.
4
State what is uncertain
Ranges, missing information and spending unrelated to AI remain visible.
PROJECT EVIDENCE
One announcement can pass through five different stages.
A project is never described as operational merely because a large investment target was announced. The source must support the stage shown.
01Announced
A company or government has stated an intention. Money may not yet be committed.
02Funded
A budget, financing package or formal programme has been approved.
03Permitted
A location has entered the planning or regulatory process.
04Building
Construction or equipment installation is reported to be under way.
05Operational
The facility or computing capacity is reported as available for use.
Important limitation
No audited worldwide total for AI infrastructure investment currently exists. The IEA's $500B estimate covers all data centres in 2024. The $695B figure is a separate combination of four company-wide forecasts for 2026 and must not be read as actual global AI spending.
SPECIALIST DATABASES
Use the right source for the question.
These independent resources provide broader or deeper facility-level coverage than this atlas. They are credited here so readers can inspect the underlying landscape rather than mistake our selected map for a global inventory.
01Epoch AI Data CentersResearch-grade estimates for major AI facilities
Power, compute, cost and project timelines supported by permits, imagery and public records.
OPEN DATABASE ↗02AI Data Center IndexStructured international facility directoryFacility, operator, capacity, status and market-level exploration across many countries.
OPEN DATABASE ↗03Jatevo DC MapBroad global data-centre footprintA large map combining colocation, cloud-region and frontier-AI infrastructure sources.
OPEN DATABASE ↗CLAIM-BY-CLAIM SOURCE LEDGER
Amounts are labelled by evidence type, period and scope.
Region / sourceAmountType / periodScope and limitations
GlobalGlobal data-centre investment
$500BReported2024
IEA baseline covers all data centres, not AI alone.High source confidence / VIEW SOURCE
US-based companyAmazon investment forecast
$200BCompany forecast2026
Company-wide forecast covering AI, chips, robotics and satellites—not AI alone.High source confidence / VIEW SOURCE
US-based companyMicrosoft investment forecast
$190BCompany forecast2026
Company-wide forecast. A substantial share supports cloud and AI infrastructure, but no AI-only amount is reported.High source confidence / VIEW SOURCE
US-based companyAlphabet investment forecast
$175-185BCompany forecast2026
Mostly for technical infrastructure, but not exclusively for AI.High source confidence / VIEW SOURCE
US-based companyMeta investment forecast
$115-135BCompany forecast2026
Supports both AI development and Meta's established products.High source confidence / VIEW SOURCE
United StatesStargate
$500BCommitment2025-2029
A target shared by several partners, not money already spent. Some supplier investments may be included elsewhere.Medium source confidence / VIEW SOURCE
European UnionInvestAI
EUR 200BCommitmentMulti-year
A target for investment from public and private sources. The EUR 20B AI-factory facility is included within it, not an additional amount.Medium source confidence / VIEW SOURCE
FranceAI summit announcements
EUR 109BCommitmentMulti-year
Private announcements in France; execution schedules vary.Medium source confidence / VIEW SOURCE
China-based companyAlibaba AI and cloud plan
RMB 380BCommitmentThree years
Alibaba's company plan, not a total for China as a country.High source confidence / VIEW SOURCE
European UnionEuroHPC AI Factories
19Public programSelected through October 2025
Complete official set across 16 EU Member States. This count excludes the separate Antenna and Gigafactory programs.High source confidence / VIEW SOURCE
European UnionAI Factory Antennas
13Public programOfficial inventory
Service extensions that broaden access to existing AI Factories; they are not counted as additional AI Factory sites on the map.High source confidence / VIEW SOURCE
European UnionFuture AI Gigafactories
Up to 5CommitmentSeparate initiative
A different, larger-scale program. The formal selection process is separate from the 19 AI Factories.High source confidence / VIEW SOURCE
NetherlandsGroningen AI Factory
EUR 200MPublic programTo 2027
EUR 70M national, EUR 60M regional and EUR 70M EuroHPC funding.High source confidence / VIEW SOURCE
GlobalNVIDIA compute financing platforms
>$500BCommitmentOver time
Aggregate third-party capital that six independent financing platforms aim to mobilize. It is not NVIDIA revenue, a single fund or money already deployed; definitive agreements are still required.High source confidence / VIEW SOURCE
United StatesNscale / Anthropic compute lease
$45BCommitment6 years; capacity expected from late 2027
460 MW of planned Nscale capacity in West Virginia. This is a signed lease commitment reported independently by Reuters and the Financial Times, not capital already deployed or operational capacity.Medium source confidence / VIEW SOURCE
United StatesCornelis Networks infrastructure financing
$205MFundedRaised September 2026
Company-announced financing for scale-up and scale-out ai and hpc networking products, including manufacturing and customer deployments. CN5000 shipping; CN6000 sampling with broader availability expected in Q4 2026. It is not data-centre capacity or spending already completed.High source confidence / VIEW SOURCE
US-based companyLambda GPU infrastructure term loan
$926MReportedTo 2030
Closed senior secured financing for GPU servers and related infrastructure supporting a committed customer deployment; separate from Lambda's May 2026 credit facility.High source confidence / VIEW SOURCE
US-based funda16z Machine Age Fund
$1.1BReported2026 fund
Raised for physical AI build-out spanning chips, memory, data centres, energy and robotics; it is not a data-centre-only investment total.High source confidence / VIEW SOURCE
CanadaSovereign AI Compute Strategy
CAD 2BPublic programMulti-year
Public compute strategy, including a CAD 700M compute challenge.High source confidence / VIEW SOURCE
CanadaMeta Alberta AI campus
C$13BReportedConstruction 2026-2029
1 GW campus, scalable to 1.8 GW. Alberta records the project as under construction; the amount excludes AI chips and its separate power plant.High source confidence / VIEW SOURCE
CanadaBell Saskatchewan AI infrastructure expansion
>C$50BCommitmentPhased; no firm completion date
Non-binding memorandum for up to 900 MW of additional capacity, creating a path to 1.2 GW. Full buildout depends on customer commitments, commercial agreements, permits, approvals and environmental review.High source confidence / VIEW SOURCE
United KingdomAI Research Resource expansion
GBP 1BPublic programTo 2030
Public compute expansion; separate from private data-centre investment.High source confidence / VIEW SOURCE
LAST EVIDENCE REVIEW28 September 2026
This review added Meta's C$13B Alberta campus as under construction and Bell's proposed Saskatchewan expansion as an announced, non-binding pathway to more than C$50B and 1.2 GW. The selected infrastructure map now contains 36 locations. The four-company forecast total is unchanged; proposed IPOs, financing commentary and non-infrastructure AI news were excluded.