China AI Substitution Monitor

The thesis in one picture: up and to the left is better. Each dot is a flagship line from the price table joined to its LMArena Elo; lines without a current arena entry are omitted.

The second substitution front, and the one where Chinese labs are already on the podium: on the video boards ByteDance (Seedance), MiniMax, Alibaba (Wan, HappyHorse) and Kuaishou (Kling) crowd the top ten, while image generation is still US-led. Each row compares the best Chinese entry with the best non-Chinese one, because a Chinese model can hold #1 here. Elo is relative to each board's own pool: a 30-point gap on video and on image are not the same distance, and none of these figures compare with the text arena above or with other sites' video arenas. Context only – no index weight. Price per minute / per image is not joined yet; it rides the Artificial Analysis key. Source: LMArena leaderboard dataset (CC BY 4.0), arena.ai scrape as the per-board fallback.

Generative-AI services ranked by 1.1.1.1 DNS resolver traffic (Cloudflare Radar) – actual visits, not search-term or app-chart proxies. Chinese services highlighted.

Best Artificial Analysis Intelligence Index score by lab country, reconstructed monthly from the current catalogue's release dates (a model AA delists drops out of history; window starts 2024 because AA re-bases the index across major versions). Lag is Epoch's us-vs-china statistic computed on AA's index – Epoch's own Capabilities Index put the average at 7 months since 2023. The price ratio compares the cheapest blended $/M within 3 index points of each side's frontier – the quality-adjusted version of the flagship price gap above. Source: Artificial Analysis, free Data API.

Quarterly accelerator output in H100-equivalents (median estimates, wide confidence intervals), US designers (NVIDIA, AMD, Google, Amazon) vs Chinese (Huawei, Cambricon), from Epoch AI's chip-sales dataset (CC BY 4.0). The series ends at the last quarter with estimates for both sides – Chinese estimates lag Nvidia's by a couple of quarters. Domestic designs are not China's whole compute inflow: Epoch separately estimates a median ~660k H100e were smuggled into China through end-2025 (90% CI 290k–1.6M), roughly a third of China's stock – so this panel understates what Chinese labs can actually train and serve on.

Why the Chinese labs play the open-weights game this page measures: they are competing on a fraction of the private capital. Figures from the Stanford AI Index (Economy chapter; cross-check: ETO's Country Activity Tracker, Crunchbase-derived). The big caveat is structural, not statistical: China's state guidance funds put roughly $184B into AI firms over 2000–2023 and none of it counts as private investment, so the true capital gap is much narrower than the ratio says. Annual, hand-refreshed each April when the AI Index ships – the year is in the label so a stale figure looks stale.

Polymarket leadership markets resolve on arena.ai rankings; Kalshi is a CFTC-regulated exchange with deeper AI books, fetched by the daily updater (no CORS API). These are the market's own probabilities that Chinese labs take or hold the frontier – the cleanest forward-looking read in this monitor. Rows tagged AUTO were found by searching Polymarket's active markets rather than pinned by id, so the monthly "best Chinese AI model" cohort keeps refreshing as each one resolves; only markets above $5k volume are admitted, which excludes the unnamed placeholder outcomes that sit at 50¢.

The market above prices a US move to cut public access to a major Chinese model; this is the safety record such a move would cite. Letter grades and grade points from the FLI AI Safety Index (semiannual, 37 indicators across six domains; two US labs shown as anchors), plus the headline of Concordia AI's quarterly Frontier AI Risk Monitor. Both are hand-keyed, low-cadence and methodology-churning between editions – compare within an edition, not across. Context only – no index weight.

Prices from the site's shared feed (market-data.json, auto-refreshed server-side). Drawdown is vs the 6-month high; orange sparklines are below it, green at it – red stays reserved for "Chinese labs" everywhere else on this page. Correlation caveat: the AI Crash Monitor's lead-lag study finds no statistically significant lead from prediction-market signals to NVDA/SOXX yet – treat this page as a market-pricing monitor, not a validated forecast.

Enterprise switching is underway. CNBC (2026-07-07): Coinbase cut AI spend nearly 50% by moving ~1,200 internal agents to GLM-5.2 and Kimi K2.7; Lindy migrated fully off Claude to DeepSeek; Uber capped engineer AI budgets at $1,500/month; Citi reportedly restricted access to the most expensive OpenAI/Anthropic tiers. Chinese models price 60–90% below US flagships for near-frontier quality.

The capex asymmetry is the exposure. The US hyperscalers out-spend China's major platforms by an order of magnitude – measured live below (UBS put a wider Chinese set, ByteDance included, at ~$57B for 2025, which is the framing, not the number). If cheap open-weight models keep compressing inference pricing, the revenue side of that capex – and the margin structure of proprietary US model vendors – is what gets repriced. NVDA's China datacenter revenue already fell from ~$4.6B/yr to ~zero under export controls, so the remaining channel is margin pressure and demand-mix shifts, not lost China sales.

The counter-thesis is live too. 2026 has produced no repeat of the Jan-2025 DeepSeek selloff: NVDA beat earnings through DeepSeek V4's launch window and made new highs after V4-Pro. Chinese labs remain export-control constrained, and US frontier models still hold the #1 spots on quality leaderboards. Watch this page for the divergence, not the conclusion.