Rhodium — China AI models ~10% of OpenAI+Anthropic ARR

Opening

Rhodium puts a hard ARR gap under China vs U.S. frontier labs — usage ≠ revenue. CNBC (Evelyn Cheng) reports Rhodium Group estimates that all Chinese AI models combined generate only about 10% of the revenue reported for OpenAI and Anthropic, using annual recurring revenue (ARR) constructed from recent monthly figures ×12.

The numbers

Chinese AI models combined ≈ 10% of OpenAI + Anthropic revenue — Rhodium via CNBC. DeepSeek ARR $500 million; MiniMax $800 million; Moonshot $1 billion — Rhodium via CNBC. Z.ai latest ARR $1.8 billion (company to investors Wed); year-end ARR target raised to $3 billion from $2.4 billion — CNBC. ByteDance $4 billion; Alibaba $2.4 billion — Rhodium via CNBC. OpenAI $40 billion ARR; Anthropic $65 billion — Rhodium via CNBC. Valuation/revenue ratios cited: Moonshot ~50x, DeepSeek ~163x vs OpenAI ~34x, Anthropic ~21x — Rhodium via CNBC. >60% of equity investment in Chinese AI chips/servers from state-affiliated sources — Rhodium via CNBC.

Rhodium partner Logan Wright (to CNBC): financing gap makes sustainable scale harder for Chinese frontier labs; equity-market dependence; government funding more hardware- than lab-focused.

Why it matters

Separates open-weight adoption / token volume narratives from dollar ARR and financing capacity — relevant when comparing China labs’ “catch-up” stories to U.S. closed-model revenue bases ahead of listing talk.

Positioning

Frontier lab monetization vs open weights — Low ARR relative to usage is a feature of open distribution and lower task pricing (CNBC cites Artificial Analysis on cost-per-task), not proof of capability parity in dollars.

Connects to

OpenRouter spend vs tokens — Same spend-vs-volume distinction on a Western routing venue (discussion). OpenAI vs Anthropic OpenRouter share flip — Share fights among high-ARR U.S. labs (discussion).