Deep|LLM: 26H1 Update (Part 2): Frontier Labs, Chinese Model Vendors & the Compute Bottleneck

After covering industry dynamics in Part 1, we discuss individual frontier labs and their implications on infrastructure spending in Part 2 of the report.

1. Leading-Company Updates: Frontier Labs and Chinese Model Vendors

Model-layer value concentrating in the Anthropic–OpenAI duopoly

Figure: Anthropic vs OpenAI quarterly ARR; Source: The Information

Model-layer revenue in H1 2026 is highly concentrated in Anthropic and OpenAI, with the industry entering a phase of duopoly co-expansion. Anthropic pulled ahead around April on the strength of Claude Code and enterprise contracts, while OpenAI re-accelerated from late May on GPT-5.5, Codex, its API, and ChatGPT distribution, and the gap is now narrowing. ARR trajectory: end-2024 $1B → end-2025 $9B → February 2026 $14B → March $19B → April overtaking OpenAI ($30B vs OpenAI $24B) → May run-rate above $47B vs OpenAI ~$33B (public / The Information basis). The customer base is 300,000 enterprises, with 1,000+ paying $1M+ annually, and a 54% market share in the coding category; average monthly spend per user on Claude is $211, 8.4× OpenAI's $25 on a comparable basis. In terms of revenue mix, Anthropic derives ~80% of its revenue from enterprise contracts, and OpenAI ~85% from consumer subscriptions.

To add context: the data in The Information's chart is closer to a revenue snapshot from early to mid-May and does not fully reflect late-May changes. Per our subsequent channel checks, OpenAI re-accelerated on GPT-5.5, Codex, and the API side in late May, and the gap between the two may narrow further, with token budgets and agent workflows becoming the new entry point for enterprise productivity budgets.