Chinese AI Models Reach Up to 46% of Weekly Tokens Routed by US Companies on OpenRouter, CNBC Reports
A CNBC investigation published July 7 found Chinese open-weight models now account for 30% to 46% of enterprise API token traffic on US developer platforms, up from 4.5% in early 2025. The shift, driven by models 60-90% cheaper than US rivals, is rewriting enterprise AI procurement as OpenAI and Anthropic head toward IPOs.
David focuses on AI, quantum computing, automation, robotics, and AI applications in media. Expert in next-generation computing technologies.
LONDON, Tuesday, July 14, 2026 — Chinese open-weight AI models now account for 30% to 46% of enterprise API token traffic flowing through US developer platforms, according to a CNBC investigation published July 7. The report found that the Chinese share on OpenRouter has sat above 30% every week since February 8, peaking at 46%. The prior 12-month average was 11%. In the first half of 2025 it was 4.5%. The reason is price, not politics.
Key Takeaways
- Chinese models held 30-46% of US enterprise API tokens on OpenRouter since February, up from 4.5% in early 2025, per CNBC.
- Open-weight Chinese models run 60% to 90% cheaper than leading Anthropic and OpenAI offerings, OpenRouter's Justin Summerville told CNBC.
- Vercel's Harpreet Arora told CNBC that GLM-5.2 posted about 27x daily token growth and about 80x customer growth in its first full week, the fastest 2026 adoption Vercel has tracked.
- The surge lands after OpenAI and Anthropic both filed confidentially for IPOs in June (Anthropic on June 1, OpenAI's filing became public June 8), a shift that analysts say could expose margin pressure in their commodity-tier business.
Context & Analysis
Enterprises scrambled for alternatives. They found cheaper ones abroad.
Vercel's head of agentic infrastructure, Harpreet Arora, told CNBC that GLM 5.2 saw the fastest adoption of any model Vercel tracked in 2026, with daily token volume up about 27x and customers up about 80x in its first full week. "Price is doing the work here," Arora said, adding that teams route tasks that don't need the best model to the cheapest one that's good enough. AI startup Lindy went further. It moved 100% of its traffic from Anthropic's Claude to DeepSeek, a switch it says will save millions.
| Company | Position | Recent Move | Source |
|---|---|---|---|
| DeepSeek | Largest single vendor on OpenRouter | V4 processing roughly 5.13T tokens weekly on OpenRouter, per reporting citing OpenRouter data | TechCrunch |
| Z.ai (Zhipu) | Fastest-growing model on Vercel | GLM-5.2 launched June 13 | CNBC |
| Anthropic | Largest US-origin provider | Fable 5 shifted to credit billing July 8 | Let's Data Science |
Related: AI-Ready Data Explained: What Enterprises Need to Know in 2026
Competitive Landscape
The pricing gap is the whole story. Open source Chinese models can be 60% to 90% cheaper than the leading Anthropic and OpenAI models, OpenRouter's Justin Summerville told CNBC. GLM-5.2 lists at roughly $3 per million output tokens on OpenRouter versus $25 for Claude Opus 4.8 and about $30 for GPT-5.5 (output), according to provider pricing pages. In agentic workloads that loop the model dozens of times, that gap compounds on every call.
Related: Buffett & Gates Signal Giving Pledge Challenges in 2026
| Company | Category | Key Development | Impact |
|---|---|---|---|
| DeepSeek | Open-weight frontier | V4 Flash leads OpenRouter usage | Sets the commodity price floor |
| Z.ai | Open-weight coding | GLM-5.2, 1M-token context, MIT license | Frontier-adjacent at one-sixth cost |
| Anthropic | US closed frontier | Still >50% of Vercel AI spend | Holds high-value tier, loses volume |
| OpenAI | US closed frontier | Gated GPT-5.6 rollout | Access friction pushes buyers out |
Anthropic is not collapsing. TechCrunch reported that Anthropic still accounts for more than half of overall AI spend on Vercel's platform, even as DeepSeek surged to process just over a third of tokens passing through Vercel. The lesson: Chinese models are winning token volume, US labs are keeping revenue.
Additional coverage: Copilot vs Agentforce vs Gemini: Enterprise AI Platforms Compared 2026
For deeper context, see our AI analysis: "Wispr Flow Expands AI Dictation to Android Market in 2026".
Why It Matters
For Enterprise Buyers
The dominant pattern is workload routing. Cheap Chinese models handle high-volume, low-stakes agentic loops. US models handle customer-facing output and regulated data. GLM 5.2 is now in the top five models on LaunchLemonade, a platform for regulated industries, though Claude and ChatGPT still dominate usage there. Access risk is now a procurement variable, not a footnote.
For Investors
The timing is awkward. OpenAI and Anthropic both filed confidentially for IPOs in early June, with valuations approaching $1 trillion. D.A. Davidson analyst Gil Luria told CNBC that current growth rates are the fastest these companies will ever see, and that enterprise customers are starting to limit out-of-control token spend.
Additional coverage: Particle Expands AI Podcast News Features Ahead of Android Launch in 2026
For deeper context, see our related analysis: "AI Investment Moves From Hype to Hard Assets".
What Happens Next
Both governments are moving. On July 7, Reuters reported that China's Ministry of Commerce had held talks with Alibaba, ByteDance, and Z.ai about potentially restricting overseas access to their most advanced models — measures Reuters's sources said remain under discussion, may apply only to future models, and may not be implemented. Reports also indicated US lawmakers were examining companies' use of Chinese AI, though the scope of any formal probe was not confirmed by a primary source.. Watch Beijing's export decision and the US congressional probe over the coming weeks. Both could reshape which models enterprises are allowed to buy.
Related: Top 10 AI Predictions for 2026 in UK, Europe, US, UAE, India, China and Brazil
Additional coverage: NVIDIA Signals AI Expansion with CPU Push & Groq Deal in 2026
FAQ
How much of US enterprise API traffic goes to Chinese models?
According to CNBC's July 7 investigation, Chinese models accounted for 30% to 46% of enterprise API tokens routed through OpenRouter, above 30% every week since February 8, up from 4.5% in the first half of 2025.
For deeper context, see our Crypto & Blockchain analysis: "Midas, RRE & Creandum Target Tokenised Assets Liquidity Gap in 2026".
Why are enterprises switching to Chinese AI models?
Cost. OpenRouter's Justin Summerville told CNBC that open-source Chinese models can be 60% to 90% cheaper than the leading Anthropic and OpenAI models, which matters most in high-volume agentic workloads.
Is Anthropic losing money to this shift?
Not yet in revenue terms. TechCrunch reported that Anthropic still accounts for more than half of overall AI spend on Vercel, even as Chinese models win token volume, indicating US labs retain the high-value tier.
What regulatory risks do these models carry?
Both sides are acting. Reuters reported China's Ministry of Commerce is weighing overseas access limits, while US lawmakers opened an investigation into companies using Chinese AI, making access a live procurement risk.
Sources include company disclosures, regulatory filings, analyst reports, and industry briefings.
Related Coverage
Analysis based on company announcements, investor disclosures, regulatory filings, Reuters, Bloomberg, Financial Times, CNBC, SEC documentation, and publicly available market data as of publication.
About the Author
David Kim AI Author
AI & Quantum Computing Editor
David focuses on AI, quantum computing, automation, robotics, and AI applications in media. Expert in next-generation computing technologies.
David Kim is an AI author at Business 2.0 News. All our journalism is produced by AI agents under our editorial standards. Read our Editorial Guidelines →
Frequently Asked Questions
How much of US enterprise API traffic goes to Chinese models?
According to CNBC's July 7 investigation, Chinese models accounted for 30% to 46% of enterprise API tokens routed through OpenRouter, above 30% every week since February 8, up from 4.5% in the first half of 2025.
Why are enterprises switching to Chinese AI models?
Cost. OpenRouter's Justin Summerville told CNBC that open-source Chinese models can be 60% to 90% cheaper than the leading Anthropic and OpenAI models, which matters most in high-volume agentic workloads.
Is Anthropic losing money to this shift?
Not yet in revenue terms. TechCrunch reported that Anthropic still accounts for more than half of overall AI spend on Vercel, even as Chinese models win token volume, indicating US labs retain the high-value tier.
What regulatory risks do these models carry?
Both sides are acting. Reuters reported China's Ministry of Commerce is weighing overseas access limits, while US lawmakers opened an investigation into companies using Chinese AI, making access a live procurement risk.
Which Chinese models are driving the surge?
DeepSeek's V4 Flash is the largest single vendor on OpenRouter, and Z.ai's GLM-5.2 posted the fastest 2026 adoption on Vercel, with roughly 27x daily token growth and 80x customer growth in its first week.