7 segments · 102 companies · as of Sep 18, 2026
Sovereign & Telco AI
Domestic AI capacity built by governments, telcos and national champions
Where it sits
Fed by: Servers & Systems · Data Center InfrastructureFeeds: Models & Platforms
Governments and large national businesses seek control over where models run, who operates the cloud and how compute is supplied. Telcos, model developers and infrastructure partners build these systems together. A sovereign label does not establish local ownership of every chip, model or financing vehicle.
Why now
Korea is commissioning public GPU services while its national center enters construction. European sovereign clouds are opening, but Stargate UK is paused. Gulf programs combine operating cloud services with much larger unbuilt projects, and September reports revise the UAE campus design.
Gulf AI programs and operators
Gulf programs bring state capital, power and domestic operators together with imported accelerators. Operating cloud services are distinguished from multi-gigawatt targets, preliminary joint ventures and sites still awaiting anchor customers or construction.
Core42 sovereign compute and Khazna campuses; July e& GPU service launched; Stargate UAE remains under construction
leaderAMD MI355X/Cisco infrastructure is live serving customers; next MI400 phase targets up to 250 MW in H2 2027 (est.)
majorCampus AI JV with Bpifrance/Mistral/Nvidia targets up to 3 GW across France (est.); expansion announced, not delivered
majorQIA-owned builder partners with Brookfield in a $20B AI infrastructure program (est.), including Qatar capacity
challengercenter3/HUMAIN proposed AI-campus JV remains under MOU; June extension leaves negotiations and approvals outstanding
emergingSyntys operates Gulf data centers; August Zankore AI venture extends sovereign compute plans into Indonesia
majorJuly Sovereign AI Compute with Core42 combines in-country GPUs, connectivity and enterprise distribution
majorAramco Digital partners with independent GroqCloud on Dammam inference; 2026 expansion delivery not quantified
nicheNEOM groundbreaking starts HUMAIN's 100 MW within a 360 MW first phase (est.); initial delivery targeted for 2028
emergingBytePlus partnership announces Riyadh cloud region for AI/media/gaming using Mobily data centers; opening not confirmed
nicheKKR stake agreement supports a $5B+ Gulf hyperscale/AI expansion program (est.); announcement is not delivered capacity
challenger
Sources: reuters.com, prnewswire.com, bam.brookfield.com, eand.com, spa.gov.sa, media.kkr.com
Stargate and other mega partnerships
Mega projects join compute buyers, campus developers, financing partners and power suppliers. Stargate is an umbrella rather than one funded asset. Capacity is shown at project scope, with paused programs and changed customer relationships separated from signed construction.
Agrees to lease ~8 GW IT at PORTS-Pike for 20 years (est.); first 800 MW expected in 2028, buildout through 2032
majorAbilene delivered 618 MW in 6 of 8 buildings by Sep 10; Dona Ana permit-hearing stay imposed Aug 23, lifted Sep 17
majorSB Energy develops OpenAI's Milam campus and PORTS-Pike; Ohio lease targets ~8 GW IT (est.), subject to phased build
majorMicrosoft takes 900 MW Abilene expansion after OpenAI pullback; original 1.2 GW Stargate campus continues
majorOracle/OpenAI Frontier and Lighthouse campuses target 1.4 GW and 902 MW IT respectively (est.)
majorOracle/OpenAI Saline campus >1 GW (est.); $16B financing secured (est.), subject to closing conditions
majorDirect Kentucky lease with TeraWulf: 401 MW/$19B (est.); expands beyond cloud-only procurement
majorAnthropic's $50B US program (est.) uses its GPU/TPU sites, including New York and Texas; delivery ramps from 2026
majorHyperion's $27B Blue Owl JV finances buildings and utilities within a larger campus; BlackRock's El Paso JV totals ~$14B
majorFairwater Wisconsin begins operation in June; linked training infrastructure expands alongside Azure regions
majorReported redesign of Stargate UAE's 5 GW masterplan (est.); initial 200 MW under construction, not verified live
majorEquity partner in Stargate and UAE buildout; planned program scale is not MGX cash already invested
majorMicrosoft contracts Kvandal's 230 MW (est.) and 30k Rubin GPUs (est.) from 2027, replacing OpenAI as anchor
majorQai partnership plans $20B in AI infrastructure (est.); managed program includes sovereign and compute projects
majorHyperVault/OpenAI India starts with 100 MW (est.), with expansion potential to 1 GW (est.)
challengerProject Rainier runs over 1M Trainium2 chips for Anthropic (est.); April agreement adds up to 5 GW of AWS capacity (est.)
leaderGoogle/Broadcom add multiple GW for Anthropic from 2027, extending the prior 1 GW-plus TPU plan (est.) for 2026
majorCo-develops OpenAI custom accelerators for a planned 10 GW rollout (est.) from H2 2026 through 2029
majorOpenAI's 6 GW Instinct program (est.) starts with planned 1 GW MI450 in H2 2026; later tranches remain milestones
majorHyperion venture owns 80% for its funds (est.); Meta leases the campus and manages construction of long-lived infrastructure
majorGIP/HPS-backed El Paso venture targets 1 GW for Meta in 2028 (est.); announced ownership split is 80/20
major
Sources: openai.com, crusoe.ai, theglobeandmail.com, abqjournal.com, about.fb.com
Korean national AI infrastructure
Korea combines national GPU procurement, telco campuses and industrial AI factories. The announced national GPU allocation is multi-year, not a delivery count. Current services, consortium membership and exploratory Stargate agreements receive separate treatment.
KOACC national center breaks ground August; KRW2.5T/15k-chip project (est.) targets 2028; interim service in 2027
leaderUlsan targets 41 MW Nov 2027/103 MW Feb 2029; reported Nvidia up-to-2 GW and Anthropic GW-scale plans
majorOperates national GPU services via Naver Cloud; 60k-GPU Nvidia allocation (est.) spans years, not 2026 delivery
majorMicrosoft sovereign-cloud partnership and national-center consortium; September adds five-year KRW6T DC plan (est.)
majorRuns Gwangju national AI infrastructure and Factory X; September reports 7,656 B200 GPUs operating (est.)
majorSelected for 2026 national GPU programs; modular GPU sites serve academia and AICA public workloads
challengerPlans >50k Nvidia GPUs (est.) for semiconductor AI factory; national-center partner and Stargate memory LOI
majorStargate memory LOI supports Korean compute ambitions; supply volume and schedule still require definitive contracts
majorCoordinates SK AI infrastructure strategy; group-wide GPU/site ambitions are not standalone SK Inc. contracted capacity
nicheAI Box modular data centers support Korean GPU deployment; Busan expansion is a plan, not delivered capacity
challengerPaju AI campus targets 200 MW (est.); July adds KRW1.3T phase-two plan (est.), not commissioned supply
challengerPlans Nvidia AI factory with 50k GPUs (est.) for robotics/autonomy; complete delivery not verified
nicheNational AI-center consortium participant; OpenAI infrastructure cooperation remains distinct from binding offtake
nicheDevelops floating data centers and explores OpenAI cooperation under Samsung LOI; no verified contracted GPU campus
emergingReported KOACC consortium participant; equity/operating allocation remains undisclosed as the national center starts construction
niche
Sources: nvidianews.nvidia.com, samsungsds.com, en.sedaily.com, digitaltoday.co.kr, datacenterdynamics.com, reuters.com
Japanese, Indian and Taiwanese builders
National champions use established telecom, industrial and energy businesses to support domestic AI capacity. Named anchor customers and operating GPU services receive priority over national headline targets or long-term land and power reservations.
Jio/Jamnagar AI build; Meta signs 168 MW (est.) for phased delivery within two years; 3 GW campus remains a plan
leaderTPG-backed HyperVault owns India AI infrastructure; OpenAI initial 100 MW commitment (est.)
majorAdaniConneX helps build Google Vizag AI hub; 1 GW/$15B overall program (est.), not Adani-only investment
majorNxtra builds Google-linked Vizag infrastructure; March $1B funding round (est.) expands domestic capacity
majorVia NTT DATA, supplies domestic Blackwell cloud and global AI campuses; former NTT DATA listing rolls to parent
majorOsaka Sakai AI center operating since January; GB200 NVL72 cloud serves domestic enterprise workloads
majorSakai AI center now targets FY2027; domestic GPU cloud commercial launch scheduled for October 2026
majorJapanese sovereign GPU cloud supplies domestic enterprises and researchers with government-backed expansion
majorShakti Cloud anchors IndiaAI GPU supply; September Rubin/GB300 expansion is planned; IPO not completed
majorReported Visonbay.ai GB300 factory plan of 27 MW (est.) in Taiwan; full commissioning after H1 target unverified
challengerLunping AI data-center project begins in 2026; 36 MW full-build ceiling (est.) is not current energized capacity
challengerVayu GPU Cloud provides locally hosted Nvidia compute for Indian enterprise and sovereign workloads
nicheTakane and PRIMERGY private-AI platforms support Japanese-controlled model deployment; 2026 rollout extends to Europe
nicheBuilds NCHC's operating Nano4 H200/GB200 supercomputer; national system delivery, not merchant cloud ownership
niche
Sources: group.ntt, newsroom.kddi.com, tata.com, googlecloudpresscorner.com, global.fujitsu, asus.com
European sovereign AI infrastructure
European programs combine national control requirements with privately operated clouds and bank-financed clusters. The region has operating GPU factories, but large announced campus targets and proposed public gigafactories should not be mistaken for delivered supply.
T-Systems Industrial AI Cloud opened February; Munich factory has 10k Blackwell GPUs (est.) with Nvidia/SAP
majorCo-owns France's Campus AI with MGX/Bpifrance; June target expands to a 3 GW national network (est.), not live capacity
majorWith Post Telecom and CleverCloud, wins one of four EU sovereign-cloud awards under a shared EUR180M six-year ceiling
majorVia STACKIT, wins an EU Commission sovereign-cloud award; the four-provider framework totals up to EUR180M over six years
majorVia Scaleway, secures one of four EU sovereign-cloud awards; six-year procurement pool is up to EUR180M, not its own award
majorEU AI Cloud integrates sovereign infrastructure; works with T-Systems on the German Industrial AI Cloud
majorWins EU sovereign-cloud framework with S3NS, Clarence and Mistral; EUR180M ceiling is shared across four providers
challengerSwiss AI Platform runs on DGX SuperPOD; Fastweb adds domestic Italian AI infrastructure and models
majorBleu with Capgemini offers French Azure-based cloud/GPU services; SecNumCloud certification not yet complete
challengerDeploys AI-ready Spanish edge data centers with Nokia; broader gigafactory proposals remain separate
nicheNorwegian AI Factory combines Nvidia compute with Red Hat AI to serve locally controlled production workloads
nicheGerman-operated sovereign GPU infrastructure serves regulated enterprises and European public-sector customers
nicheFinland and other European sites supply local compute; most global contracted power is due online over 2-3.5 years
majorNarvik adds European AI capacity; the Aker joint venture rolls into Nscale, which now solely manages the Norwegian project
majorReported EUR30B French AI buildout plan (est.); disclosed SEK95B Sweden program (est.) remains future infrastructure
majorBacks Nscale/Kvandal; Microsoft $6.2B site contract (est.) replaces OpenAI as disclosed anchor customer
nicheControlled S3NS operates PREMI3NS, qualified SecNumCloud 3.2 in December 2025; Google technology remains locally operated
majorOwns 50% of Bleu with Orange (est.); French Azure cloud offers GPU services while SecNumCloud qualification remains pending
nicheSovereign Cloud targets EU-controlled operations for government, defense and health; rollout does not prove a new GPU fleet
nicheSeptember Aleph Alpha combination is signed, not closed; planned German operations extend sovereign model deployment
challengerSovereign AI Cloud uses Verda at Helsinki Data Center; Telia resells and manages the customer service, not Verda's whole fleet
nicheSwedish Armed Forces selects sovereign cloud services; Stockholm/France campus plans are separate from commercial availability
niche
Sources: t-systems.com, reuters.com, assets.nebius.com, news.sap.com, thalesgroup.com, ec.europa.eu
Regional telco and sovereign factories
Telcos and regional groups distribute domestic AI compute using existing networks, enterprise customers and power access. Large national campus ambitions are kept separate from current GPU cloud products, signed capacity and local regulatory permissions.
RE:AI: 12 MW secured, 6 MW contracted and SGD600M multi-year TCV (est.); ramps through FY2027
majorBell AI Fabric buys BUZZ compute for Cohere; September Saskatchewan expansion targets up to 1.2 GW (est.)
majorRimouski sovereign AI factory operating; Kamloops expansion supplies Canadian-controlled Nvidia cloud capacity
majorLintasarta sovereign GPU cloud; new Zankore venture targets 200 MW in 2027 (est.), not current delivered MW
majorYTL AI Cloud operates Johor GB200 infrastructure; wider 600 MW park (est.) includes future and non-GPU capacity
majorDomestic AI factories in Vietnam/Japan operate H100/H200 clusters; B300 service expansion announced in March
majorVietnam sovereign AI cluster deployed in March with 22 DGX B200 systems (est.), not 22 individual GPUs
challengerNeutraDC develops Indonesian AI-ready campuses; proposed outside investment is not yet a closed sale
challengerGSA joint venture with Singtel/AIS operates Thai liquid-cooled capacity; later-campus MW remain development plans
challengerGSA partner provides Thai enterprise access and local connectivity to AI-ready data-center infrastructure
nicheProject Southgate participates in Nvidia Australia buildout; national 2 GW target (est.) is shared and unbuilt
majorAustralian sovereign GPU supplier in Nvidia September expansion; 212 MW secured power (est.) is not live capacity
challengerSeptember Vodafone Business partnership introduces Egypt sovereign GPUaaS; local delivery stages and MW remain undisclosed
emergingAnchors Bell/BUZZ Canadian sovereign compute; September Aleph Alpha merger signed but regulatory close pending
majorePLDT's Pilipinas AI combines local GPU services at VITRO Sta. Rosa with Dell/Katonic infrastructure and Vantiq applications
nicheEgypt business announces Cassava sovereign AI factory and local Nvidia GPUaaS; partnership launch is not a full-capacity delivery
niche
Sources: cdn1.singteldigital.com, canada.ca, nvidianews.nvidia.com, fpt.com, vantiq.com, menafn.com
Technology partners for sovereign AI
Technology suppliers enable domestic clusters through accelerators, networking, cloud software and operating models. This segment tracks named sovereign programs rather than generic exposure to government software spending or general IT consulting.
DSX/NCP ecosystem supports Australian 2 GW buildout (est.) and Korea multi-year GPU allocations, not delivered totals
leaderMI355X/EPYC power HUMAIN's live Saudi service; MI400 expansion targets up to 250 MW from H2 2027 (est.)
challengerSilicon One, Nexus and 800G networking equip HUMAIN's operating MI355X cloud; additional Saudi capacity is planned
majorHUMAIN targets 200 MW of AI200/AI250 capacity from 2026; initial AI100 deployment supports Adobe's inference workloads
challengerAzure infrastructure underpins KT sovereign partnership and local cloud control options for regulated customers
majorAWS supports SK Ulsan and Saudi AI-zone programs; national project targets are distinct from energized AWS regions
majorGoogle supports India Vizag hub and KDDI Gemini on-prem services; TPU supply expands sovereign compute options
majorOCI supplies sovereign regions and Stargate UAE technology; UAE initial 200 MW (est.) remains under construction
majorSovereign Core gives local operators control over AI-cloud identity, deployment and governance infrastructure
nicheSovereign EU cloud and T-Systems partnership connect enterprise data workloads to domestically operated AI capacity
nicheFoundry/AIP software joins Rackspace Blackwell sovereign pod announced in September; infrastructure operator is Rackspace
nicheOperates September Blackwell/Palantir sovereign pod for regulated enterprises; early commercial rollout
challengerGC2 on AWS GovCloud keeps orchestration in agency VPCs; Nutanix Enterprise AI deploys on-premises LLMs
majorUS Air Force PANDA predictive-maintenance AI expands under a $13M task order (est.); $450M ceiling is not funded revenue
nicheAcquired Ask Sage serves Army enterprise generative AI on cArmy IL5 infrastructure; acquisition closed December 2025
niche
Sources: nutanix.com, c3.ai, bigbear.ai, qualcomm.com, newsroom.cisco.com
Glossary
- Sovereign cloud
- Cloud infrastructure organized around local legal, operational and data-control requirements; definitions vary by project.
- AI factory
- A cluster and its supporting power, cooling and software for model training or inference; not necessarily a physical factory.
- Offtaker
- The buyer signing to use future capacity; a memorandum or expression of interest is not a binding offtake contract.
- SPC
- A special-purpose company holding a project, its contracts and financing, often shared by public and private sponsors.
Heat, spend, exposure and shares are editorial estimates. How to read the map ↗