Key Takeaways:
Anthropic's safety testing placed Kimi K2.7 alongside GPT-5.5 and its own Opus 4.8, while Coinbase and SpaceX-backed Cursor have already put Chinese models into production.
Chinese AI models have crossed a threshold in the US enterprise market, winning production deployments at Coinbase and powering a startup acquired by SpaceX for $60 billion, as cost-conscious American companies seek alternatives to frontier American labs.
"We're experimenting with defaulting to open-weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task," Brian Armstrong, chief executive officer of Coinbase, said in a post on X.
Coinbase's AI spending fell nearly 50% from its peak after switching default models to Chinese open-weight alternatives, according to a graph Armstrong shared. Token usage reached one of the highest levels in company history even as costs declined. The crypto exchange also routes prompts to appropriate models based on difficulty, improved caching to lift hit rates from 5% to 60%, and keeps context lean by starting new sessions between tasks.
The shift signals a reordering of the AI competitive landscape. Anthropic's internal safety testing of its newly restored Claude Fable 5 found that eight models — including Kimi K2.7, GPT-5.5, and its own Opus 4.8 — could all identify the same software vulnerabilities that triggered US export controls on Fable 5. The finding effectively placed a Chinese model in the same tier as frontier Western systems, validated by the lab widely considered the industry leader.
The Kimi Commercial Flywheel
The validation extends beyond benchmarks. Cursor, the AI coding assistant that SpaceX agreed to acquire on June 16 for about $60 billion in an all-stock deal, uses Kimi K2.5 as the base model for its Composer feature. SpaceX shares surged roughly 16% on the announcement, briefly making it the fourth-largest US company by market capitalization.
Fireworks AI, a Kimi partner that provides inference infrastructure, saw its valuation climb from $4 billion to roughly $15 billion over about seven months. Its customer roster includes Cursor, Perplexity, Notion, and Uber.
Kimi's own financial trajectory mirrors the adoption curve. Its annual recurring revenue tripled in three months, with overseas user growth of 400%, according to company disclosures. API revenue accounts for about 70% of total revenue, a structure similar to Anthropic's own business mix. The company employs roughly 300 people and generates revenue from more than 200 countries, with overseas income approaching half the total.
Valuation Gap and the Anthropic Parallel
Kimi's pre-money valuation stands at about $31.5 billion, according to public filings — roughly 3% of Anthropic's latest valuation and below Zhipu AI's market valuation on the Hong Kong Stock Exchange. The gap between Kimi's revenue trajectory and its valuation creates a structural tension for investors.
The company's growth pattern — product-led, developer-driven, API-centric — closely tracks Anthropic's own path from $100 million to $1 billion in ARR during 2024. Kimi crossed $100 million in ARR in March 2026, putting it roughly where Anthropic stood two years ago.
For investors, the question is whether the valuation discount reflects a genuine risk premium for Chinese technology companies operating in a US-dominated market, or an opportunity to buy an Anthropic-like growth curve at a fraction of the price. The answer depends on whether US enterprises continue adopting Chinese models at the current pace — and whether geopolitical tensions escalate further.
This article is for informational purposes only and does not constitute investment advice.
