The foundation-model-providers segment is the capital-intensity and run-rate-scale top of the agentic AI market — and it is the segment most badly served by category-level commentary that reads all nine of the active labs as broadly comparable competitors. They are not. Our second IM Category Report scores OpenAI, Anthropic, Google DeepMind, Meta, Mistral AI, Cohere, DeepSeek, xAI (now a SpaceX subsidiary as of 2 February 2026), and Aleph Alpha across the full IM Framework, with the cohort’s structural shape — rather than any vendor’s headline benchmark — as the central editorial finding.
Download the full briefing report — 41 pages — here.
- The cohort is two-tiered, not nine-way comparable. OpenAI, Anthropic, Google DeepMind, and Meta operate at a structural scale — combined disclosed run-rate revenue approaching $100 billion (Anthropic alone crossed $47 billion in May 2026), combined committed forward compute spend in excess of $1 trillion through 2032 — that the cohort’s remaining five entities are not. The single most consequential fact in the segment is the size of the gap between its top four and its remaining five.
- Anthropic has leapfrogged OpenAI as the cohort’s most valuable lab. The 28 May 2026 $65 billion funding round (Altimeter Capital-led; FT) values Anthropic at $965 billion, surpassing OpenAI’s $852 billion. Anthropic’s stated run-rate revenue crossed $47 billion in May — a more-than-fivefold increase since the start of 2026.
- Robustness, not raw score, is the buyer’s optimisation target. Eight-scenario robustness leads at Google DeepMind (8.03) and trails to Aleph Alpha (3.69) — a wider spread than the coding-agent cohort we scored at the start of the year. The framework’s robustness statistic flattens what the macro-construct scores separate; for a buyer making a multi-year integration commitment, the Trajectory shape matters more than the headline number.
- Compute supply, not capability, is the binding constraint. Three top-tier vendors have signalled compute-constrained operations in 2026 — Anthropic on Claude Code rate limits, Alphabet through Pichai’s “we are compute constrained in the near term,” and OpenAI implicitly through the forward-commitment stack. NVIDIA’s Q1 FY27 print confirmed the supply-led reading: $75.2 billion Data Center revenue (+92% YoY) and a $91 billion ±2% revenue guide for Q2 FY27.
- The Cohere–Aleph Alpha merger has reshaped the European story but has not yet closed. On 24 April 2026 Cohere announced an all-stock acquisition of Aleph Alpha at a combined-entity valuation of approximately $20 billion, with a concurrent $600 million Series E commitment from Schwarz Group. The deal requires German foreign-direct-investment clearance and Canadian Investment Canada Act review. We score the two vendors separately and flag the transaction as the most consequential structural event in the European segment in 2026.
- Disclosure quality varies by an order of magnitude, and the framework weights it explicitly. Alphabet and Meta file audited 10-Qs and 10-Ks. OpenAI, Anthropic, and Cohere have company-stated revenue. xAI’s $3.2 billion 2025 revenue is now disclosed via the SpaceX S-1 filing (a material upgrade from the prior press-leaked-investor-materials regime). DeepSeek and Aleph Alpha revenue are effectively undisclosed. The cohort cannot be scored on Strategic Resilience as if these were comparable filings — and our scoring does not pretend they are.
The eight-scenario Trajectory Profile from our flagship Macro Briefing Eight Futures for Agentic AI lets us name, for every vendor, the futures the vendor is positioned for and the futures the vendor is exposed by. The cohort splits cleanly into three groups when read across the eight scenarios — a robust-everywhere top tier, a capability-frontier-and-rewire-leading middle tier, and a two-mode bottom tier where the European-sovereignty case and the asymmetric-bet case run on opposite shapes. Defensibility concentrates with the audited filers; Disruption Potential concentrates with the model-and-product alignment cases.
The next eighteen months read more as a supply story than a capability story. NVIDIA’s Q1 FY27 print, EU AI Act enforcement powers from 2 August 2026, and the Mistral Series D / pre-IPO that the $830 million Bruyères-le-Châtel debt raise implies all sit inside the briefing horizon. For buyers committing now, the question is which Trajectory shape — not which raw score — fits their own scenario weighting.
Download the full briefing report — 41 pages — here.
Martin De Saulles | Principal Analyst. #IM109. May 2026.
