Global Sovereign AI Infrastructure Programmes Outlook to 2030

Executive Summary

Sovereign AI infrastructure programmes have moved from announcement to construction. Cumulative committed capex across named programmes exceeds US$1.5 trillion through 2030, anchored by Stargate (US$500 billion / 10 GW by 2029), HUMAIN (Saudi Arabia, 600,000 NVIDIA GPUs over 3 years), IndiaAI Mission (₹10,372 crore outlay, more than 38,000 GPUs deployed), Mistral (US$830 million March 2026 financing, 1.4 GW 2030 campus), and Isambard-AI (UK, £225 million, live July 2025).

Three forces drive the trajectory. The May 13, 2025 rescission of the BIS AI Diffusion Rule by the Trump administration replaced country-cap architecture with bilateral partnership models, opening Stargate-class export deployments to UAE, Saudi Arabia, and other GCC partners. MGX cross-border capital flow now links UAE sovereign funds to Stargate US, Mistral France's 1.4 GW campus, and Anthropic's February 2026 US$30 billion Series G, a structurally important pattern. And the binding execution constraint has shifted from capital to physical: grid power availability, GB300 allocation, advanced packaging (CoWoS) capacity, and cooling infrastructure gate deployment timelines.

Market Overview

Sovereign AI infrastructure scope captures national-scale compute programmes that pair government policy with hyperscaler-vendor capex: Stargate (US, UAE, and Argentina), HUMAIN (Saudi Arabia and US extension), IndiaAI Mission, Mistral (France and EU sovereign positioning), Isambard-AI (UK), and the broader cohort of national AI strategies in Korea, Japan, Singapore, Canada, and emerging programmes. The category is distinct from commercial hyperscaler capex (covered in our Generative AI Foundation Models outlook) by its sovereign-policy-anchored governance and procurement model.

The category sits at the intersection of three forces. AI-data-centre power demand is forecast to grow approximately four times through 2030 per IEA estimates; sovereign programmes provide the policy and capital substrate to build dedicated capacity at scale. US export-control posture (the May 13, 2025 BIS Diffusion Rule rescission, the December 2025 H200 revenue-share arrangement, the January 13, 2026 BIS final rule on case-by-case licenses) shapes which sovereign partners can access frontier compute. And the binding execution constraint is physical: grid power availability is the binding bottleneck for Stargate sites, HUMAIN deployment, and Mistral's 1.4 GW campus.

Geopolitically, MGX cross-border capital flow is the canonical pattern. UAE sovereign capital participates in Stargate US (via G42 and partnership architecture), Mistral France 1.4 GW 2030 campus (NVIDIA and MGX), and Anthropic's February 2026 US$30 billion Series G. This compounds with PIF participation in HUMAIN US extension and broader cross-border flows that did not exist prior to 2024.

Market Size and Growth Outlook

Cumulative committed capex compounds rapidly through 2026 as Stargate (US$500 billion / 10 GW) and HUMAIN (partnership of more than US$50 billion), IndiaAI Mission, Mistral, and Isambard-AI all advance. From 2027 onward, growth moderates as the capex cycle shifts from announcement to operational deployment; the binding constraint shifts from capital to physical. Operational sovereign-AI compute capacity grows from approximately 1.2 GW (2024) to 22 GW (2030) if announced programmes deliver on schedule.

Market Segmentation

By Programme

Stargate at 38% of committed capex anchors the category. HUMAIN at 18% includes the US-Saudi Investment Forum announcement of up to 600,000 NVIDIA GPUs over 3 years and the Aramco-aligned ecosystem. IndiaAI Mission's smaller absolute scale is offset by more than 38,000 GPUs deployed (versus 10,000 target) and the Yotta deployment (20,000 NVIDIA Blackwell Ultra by August 2026, 10,000 dedicated to IndiaAI). Mistral's commitment includes 13,800 NVIDIA GB300 at the 44 MW Bruyères-le-Châtel facility and the 1.4 GW 2030 campus.

By Technology Stack

NVIDIA dominates sovereign-AI compute at 62%. Stargate UAE deploys NVIDIA stack with G42 integration. HUMAIN's first phase uses 18,000 NVIDIA GB300 and InfiniBand. Mistral's first DC uses 13,800 NVIDIA GB300. IndiaAI's Yotta deployment uses NVIDIA Blackwell Ultra. The structural concentration to NVIDIA shapes capacity allocation across sovereign programmes, a constraint covered in detail in our Advanced Packaging and HBM outlook.

By Region

The Middle East at 32% reflects HUMAIN, Stargate UAE, and the broader GCC sovereign-AI cluster. The US at 28% captures Stargate domestic siting (Shackelford TX, Doña Ana NM, Lordstown OH, Milam TX). Europe is anchored by Mistral France, UK Sovereign AI Unit, emerging Germany, and Nordic programmes. India scales fast on cost-leadership in renewable electricity and IndiaAI Mission framework.

Trends and Developments

BIS AI Diffusion Rule rescission reshapes sovereign architecture

The May 13, 2025 rescission by the Trump administration replaced country-cap architecture with bilateral partnership models. The Tier 2 GPU caps (~50,000 H100-equivalent per company per Tier 2 country) never came into force. Stargate UAE launched May 22, 2025, nine days after rescission, with G42, OpenAI, Oracle, NVIDIA, Cisco, and SoftBank. The BIS January 13, 2026 final rule shifted to case-by-case license review for H200 and MI325X exports to mainland China, Hong Kong, and Macau. The December 2025 H200 revenue-share arrangement (25% to US Treasury for qualified Chinese exports) created a precedent for commercial export-control mechanisms.

MGX cross-border capital flow as canonical pattern

UAE sovereign fund MGX participates across Stargate US (via G42 partnership architecture), Mistral France 1.4 GW 2030 campus (NVIDIA and MGX co-investors), and Anthropic February 2026 US$30 billion Series G (co-led by GIC and Coatue with MGX participation). PIF mirrors the pattern in HUMAIN US extension. This cross-border sovereign capital flow is structurally new and shapes commercial-relationship architecture across the broader frontier-AI ecosystem.

Power as the binding execution constraint

Stargate sites face grid-interconnection challenges in Texas, New Mexico, and Ohio. Mistral's 1.4 GW campus depends on French grid and possible nuclear-anchored siting. NEOM Green Hydrogen capacity could theoretically anchor sovereign-AI power demand. The strategic implication is that sovereign-AI execution timelines through 2030 are gated by grid, interconnection, data-centre liquid cooling, and advanced packaging supply, not by capital availability.

Other relevant developments include the IndiaAI Compute Portal subsidy (₹65 per GPU hour), the Korean ETRI and Naver sovereign-AI cluster, the Japan Mitsui and SoftBank Stargate-equivalent discussions, the UK £900 million supercomputer fund and £500 million Sovereign AI Unit, and Argentina Stargate (US$25 billion / up to 500 MW) October 10, 2025.

Competitive Landscape

Stargate (US$500 billion / 10 GW by 2029, more than US$400 billion committed through Q3 2025) anchors the category. HUMAIN (up to 600,000 NVIDIA GPUs over 3 years, partnerships with xAI, Global AI, AWS and US data centre extension) is the second-largest committed programme. The Mistral cluster includes US$830 million debt financing (March 2026, 7 European banks) and the 1.4 GW 2030 campus. Cross-cutting through all is NVIDIA as the dominant compute substrate (62% of sovereign-AI deployments) and MGX as the dominant cross-border sovereign capital partner.

Key Policies and Regulatory Environment

BIS AI Diffusion Rule rescission and successor framework

The BIS AI Diffusion Rule was issued January 15, 2025 by the Biden administration and rescinded May 13, 2025 by the Trump administration days before the May 15 effective date. The rescission replaced country-cap architecture with bilateral partnership models. Stargate UAE launched May 22, 2025 within this revised framework. The BIS January 13, 2026 final rule shifted to case-by-case license review for H200 and MI325X exports to mainland China, Hong Kong, and Macau.

EO 14179 (January 2025) and successor US AI policy

Executive Order 14179 (January 2025) revised EO 14110 (October 2023), shifting federal AI policy from safety-led to capability-led. Removed safety-evaluation requirements that had been imposed on frontier-model providers. Combined with the BIS Diffusion Rule rescission, the framework signals a US federal posture optimised for US frontier-lab competitiveness and sovereign-AI partnership architecture.

Other relevant frameworks include the IndiaAI Mission framework (₹10,372 crore outlay, MeitY cabinet approval March 2024), UK AI Sector Plan and Sovereign AI Unit (£500 million, up to 1 million GPU hours per startup on Isambard-AI), Korea Hydrogen Economy and AI Strategy, Japan METI AI compute strategy, and Singapore IMDA Sovereign AI investments. The EU AI Act GPAI provisions (covered in our Generative AI Foundation Models outlook) interact with sovereign-AI deployments where Mistral and other EU-headquartered models are involved.

Future Outlook

The sovereign AI infrastructure category in 2030 reaches approximately US$1.55 trillion of cumulative committed capex with operational compute capacity of approximately 22 GW. Stargate (10 GW by 2029), HUMAIN, IndiaAI, Mistral, UK, and emerging programmes deliver the operational base. The structural shift through 2030 is from announcement to physical execution; capacity availability becomes the binding strategic variable, not capital.

The competitive landscape consolidates at the substrate layer: NVIDIA's structural dominance compounds (62% in 2024 expected to hold near 60% by 2030 as competitors compete for the residual 38%). MGX cross-border capital position entrenches across multiple programmes. The geographic mix shifts as Middle East scale execution maturity demonstrates and India's cost-leadership compounds.

The biggest risk is power-availability gating in 2027-28. Stargate, HUMAIN, Mistral, IndiaAI, and emerging programmes collectively require approximately 22-30 GW of new grid-firm power by 2030. Grid interconnection, permitting, and advanced packaging (CoWoS) supply gate physical execution. The leading indicator is operational MW additions per quarter at top-3 programmes through 2026-27.