Global Generative AI Foundation Models Outlook to 2030

Executive Summary

Generative AI foundation models are in their commercial revenue inflection. The integrated category, spanning frontier-lab provider revenue, hyperscaler-hosted inference, and the enterprise application layer, grows from US$30 billion in 2024 to US$420 billion by 2030 at 46% CAGR.

Three forces drive the trajectory. Anthropic's run-rate revenue passed US$30 billion (gross) in April 2026, a 30-fold growth from US$1 billion in January 2025. OpenAI tracked from US$6 billion (end-2024) to US$24-25 billion (April 2026). Capital concentration to top-3 frontier labs reached strategic-asset-class scale: Anthropic Series G US$30 billion at US$380 billion post-money (February 12, 2026), Google's up-to-US$40 billion commitment April 2026, Amazon's cumulative US$33 billion. Open-weights and Chinese-developed models (DeepSeek, Alibaba Qwen, Meta Llama) grew from 1.2% to 30% of global usage in 2025. And EU AI Act GPAI obligations entered force August 2, 2025; full Commission enforcement powers activate August 2, 2026; pre-existing model compliance August 2, 2027.

Market Overview

The generative AI foundation-model scope captures three layers: frontier-lab provider revenue (Anthropic, OpenAI, Google, xAI, Mistral, Cohere, Meta Llama, DeepSeek, Alibaba Qwen), hyperscaler-hosted inference (AWS Bedrock, Google Vertex AI, Microsoft Azure AI, Oracle OCI), and the enterprise application layer (Microsoft 365 Copilot, ChatGPT Enterprise, Claude for Work, Google Workspace AI). Revenue is partially overlapping across the layers (Anthropic Claude is sold through AWS Bedrock and Google Vertex), so we size the integrated category at gross-revenue layer.

The category sits at the intersection of three forces. Enterprise software discipline post-2022 favours providers (Anthropic on closed-weight side, DeepSeek/Qwen on open-weights) demonstrating cost-per-quality. The AI-power-and-grid binding constraint (covered in our Sovereign AI Infrastructure outlook) gates training and inference scale through 2028. And geopolitical export-control posture: BIS AI Diffusion Rule rescinded May 13, 2025; January 13, 2026 final rule shifted to case-by-case license review for H200 and MI325X exports to China; December 2025 H200 revenue-share arrangement (25% to US Treasury for qualified Chinese exports).

The regulatory layer is moving from voluntary to binding. EU AI Act GPAI provisions are the first binding regime. UK AI Security Institute and Bletchley/Seoul/Paris Summit framework provides non-binding international coordination. China's Interim Measures on Generative AI and Deep Synthesis Provisions regulate the domestic Chinese market. The US framework remains posture-led at federal level (EO 14179) and state-level overlay (California SB 53, Colorado AI Act).

Market Size and Growth Outlook

The 2025-26 acceleration to 85-117% growth is the revenue inflection itself: Anthropic's 30-fold ARR ramp, OpenAI's continued four-fold growth, and Microsoft's September 2025 Anthropic integration into Microsoft 365 Copilot. From 2027 onward growth moderates to 25-45% as base effects compound and frontier-lab provider revenue concentration approaches saturation among Top-1000 enterprises.

Market Segmentation

By Model Class

Closed-weight frontier dominates at 58%. The fastest-growing segment is open-weights Chinese, expected to reach 22% by 2030 at 62% CAGR, driven by DeepSeek's cost-of-capability disruption and Qwen's cross-jurisdictional adoption. Closed-weight challenger compresses to 6% by 2030 as the strategic-squeeze thesis plays out.

By Buyer Cohort

Enterprise at 36% is the most strategically important segment. Anthropic's 40% share of enterprise LLM spend (per Menlo Ventures 2025) reflects the most decisive enterprise-share gain in software-category-challenger history. Government and sovereign expand from 6% to 12% by 2030 driven by IndiaAI Mission deployments, Mistral's European positioning, and Anthropic's AWS GovCloud expansion.

By Region

North America's 58% share reflects the concentration of frontier labs (Anthropic, OpenAI, Google, xAI, Meta) and US enterprise adoption. China's 14% share is anomalous: Chinese open-source models grew from 1.2% to 30% of global usage in 2025 but commercial revenue capture remains concentrated in domestic and non-US-aligned international markets.

Trends and Developments

Anthropic enterprise ascendancy reshapes vendor selection

Anthropic's growth from 12% (2023) to 24% (2024) to 40% (2025) of enterprise LLM spend per Menlo Ventures is the most decisive vendor-selection shift in foundation-model history. Microsoft's September 2025 integration of Anthropic into Microsoft 365 Copilot broke prior OpenAI-exclusivity in the world's largest enterprise AI distribution channel. The implication is that hyperscaler distribution is now multi-vendor across the top-3 frontier labs.

DeepSeek and Qwen demonstrate cost-of-capability disruption

DeepSeek V3's approximately US$5.6 million training cost (per Nature paper September 2025; the headline US$294,000 figure refers only to the R1 reinforcement-learning phase). Alibaba Qwen 2.5-Max release January 29, 2025 outperformed DeepSeek V3 on multiple benchmarks. Mixture-of-experts architecture reduces computational cost by approximately 30% versus dense models. Chinese open-source models grew from 1.2% to 30% of global usage in 2025.

Capital concentration reaches strategic-asset-class scale

Anthropic Series G February 12, 2026: US$30 billion at US$380 billion post-money (led by GIC and Coatue, co-led by D. E. Shaw, Dragoneer, Founders Fund, ICONIQ, MGX). Google's up-to-US$40 billion commitment April 24, 2026 (US$10 billion wired immediately at US$350 billion). Amazon cumulative US$33 billion. xAI Series E US$20 billion early 2026 at over US$200 billion implied valuation. Top-3 frontier-lab cumulative committed capital exceeds US$200 billion.

Other relevant developments include Microsoft's three-frontier-lab procurement architecture (OpenAI, Anthropic, and Microsoft first-party in Microsoft 365 Copilot), reasoning and agentic models displacing conversational chat as the value frontier (GPT-5, Claude 4 with extended thinking, Gemini 2.5 Deep Think), and EU GPAI Code of Practice signatory status (Anthropic, OpenAI, Google, Mistral, Cohere signed; Meta did not).

Competitive Landscape

Top-3 closed-weight frontier labs collectively control 76% of enterprise LLM spend. Anthropic is the structurally most defensible position: dual-hyperscaler backing (Amazon and Google), Microsoft 365 Copilot integration, and 30-fold ARR ramp creates the strongest vendor moat. OpenAI retains the strongest consumer franchise (60% of revenue) and Microsoft enterprise distribution. Google has the deepest substrate stack (TPU, Google Cloud, Gemini, AlphaFold 3, and Isomorphic Labs vertical) but Gemini's 11% enterprise share lags its capability tier. xAI is the most heavily capitalised single-operator capex push outside top-3 (valuation of more than US$200 billion) but Grok ARR remains under US$2 billion. Mistral has the cleanest sovereign-European position (US$830 million debt March 2026, 1.4 GW 2030 campus with NVIDIA and MGX).

Key Policies and Regulatory Environment

EU AI Act: General Purpose AI provisions

GPAI obligations entered force August 2, 2025. Full Commission enforcement powers activate August 2, 2026. Pre-existing GPAI model compliance deadline August 2, 2027. Providers must produce technical documentation, copyright controls, training-data summary, and (for systemic-risk GPAI over 10^25 FLOP) model evaluation, adversarial testing, incident reporting, cybersecurity. Fines up to 3% of global annual turnover or €15 million. This is the first binding compliance regime for foundation-model providers globally.

EU GPAI Code of Practice

Signed July 10, 2025 by Anthropic, OpenAI, Google, Mistral, Cohere. Not signed by Meta. Signatories receive presumption of compliance with the EU AI Act's GPAI transparency obligations. Code of Practice signatory status is a procurement-criterion advantage in EU enterprise sales through 2027.

Other relevant frameworks include US Executive Order 14179 (January 2025, capability-led posture replacing EO 14110 safety-led posture), California SB 53 / Frontier AI Transparency Act (September 2025), Colorado AI Act (effective February 2026), China CAC framework (Interim Measures on Generative AI and Deep Synthesis Provisions), UK AI Security Institute, and BIS export controls on advanced compute (May 13, 2025 Diffusion Rule rescission, January 13, 2026 final rule).

Future Outlook

The generative AI foundation-model market in 2030 reaches approximately US$420 billion. Frontier-lab provider revenue accounts for 38% of value, hyperscaler-hosted inference 28%, and enterprise application layer 34%. The structural shift through 2030 is migration from API revenue toward integrated application revenue, the cloud-computing layer-cake stratification pattern compressed to a six-year timeline. Domain-vertical applications, including AI-led drug discovery, capture an outsized share of the application-layer pie because they wrap proprietary data and regulated workflows around frontier-model inference.

The competitive landscape consolidates further at the top. Top-3 frontier labs collectively control 78% of enterprise LLM spend by 2030. Mid-tier closed-weight challenger compresses by approximately 50% in absolute share. Open-weights and Chinese segment expands from 5% to 22% of category value.

The biggest risk is capability convergence between closed-weight frontier and open-weights/Chinese models by 2027-28. If the absolute capability gap closes entirely, frontier-lab pricing power compresses by an estimated 30-40% on premium-tier API rates and the back-half forecast (2028-30) compresses by 20-30%. Secondary risk is a frontier-lab capex-revenue mismatch event during a market correction.