Half of Enterprise Tokens Already Route Through China

U.S. companies are routing nearly half their enterprise AI tokens through Chinese models — not because Washington lost the capability race, but because gating America's frontier labs made cheap open weights the rational default.

The numbers:

• Chinese-origin models have held 30%+ of U.S. enterprise token volume on OpenRouter every week since February 8, peaking at 46%, per CNBC's July 7 investigation citing platform data.
• The prior 12-month average was 11%. In the first half of 2025 it was 4.5%.
• Combined U.S. model share on OpenRouter collapsed from roughly 70% to 30% over the past year, per Bloomberg and Exponential View data cited by OfficeChai.
• DeepSeek is OpenRouter's single largest vendor at 17.6% of routed tokens — more than Google (12.5%), Anthropic (14.8%), or OpenAI (8.4%) — per June platform data analyzed by Forkast.

Why now:

• Open-source Chinese models run 60% to 90% cheaper than leading Anthropic and OpenAI offerings, Justin Summerville of OpenRouter told CNBC. DeepSeek V4 Flash costs $0.14 per million input tokens versus $5.00 for GPT-5.5, per Forkast pricing data.
• Vercel's Harpreet Arora told CNBC: "Price is doing the work here." GLM-5.2 saw daily token volume grow 27x and customer count 80x in its first full week after launch.
• AI startup Lindy moved 100% of managed agent traffic from Claude to DeepSeek in June. CEO Flo Crivello told CNBC: "You could see that cost curve go down, like, crash to the ground" — saving millions within months.
• DeepSeek topped Ramp's June trending foundational-LLM expense index, per OfficeChai — meaning corporate cards, not sandboxes.

The policy irony:

• Washington spent June gating GPT-5.6 Sol behind a federal guest list and briefly killing Anthropic's Fable 5 and Mythos 5 globally.
• On July 9, OpenAI launched GPT-5.6 publicly — but Sol's most sensitive capabilities remain gated behind federal clearance.
• France24 reported July 9 that developers who relied on closed U.S. models felt the disruption firsthand. Stems Labs co-founder Haitham Mengad on losing Fable: "It's almost like a drug."
• Hugging Face's Yacine Jernite told CNBC: "There is a real risk that users get stuck having to choose between performant but expensive US proprietary models whose price and accessibility can quickly fluctuate, or using Chinese models as the only feasible alternative whenever they want to control costs or own their AI stack."

Congress noticed — too late:

• On July 8, CNBC reported House Homeland Security and China Select Committee chairs are probing growing Chinese model adoption — letters went to Cursor (building Composer 2 on Kimi, now subject to Musk's $60 billion acquisition) and Airbnb.
• A committee aide told CNBC lawmakers are examining whether America has "a sufficient open-weight AI strategy" so companies aren't forced to choose between "expensive or restricted U.S. models and cheap, capable PRC-developed alternatives."
• Rep. Andy Ogles said in June: "When the cheap, capable, easy option for an AI model is Chinese, the rest of the world will build on it."
• Brookings fellow Kyle Chan told CNBC it's "ultimately impossible to ban China's open-source AI models because their model weights are available freely on the internet" — and that restrictions "could enter into first amendment speech issues."

Convina's view: Washington thought it was securing the frontier. OpenRouter's receipts show it was pricing Chinese models into the default stack. You cannot gate Sol behind a guest list, watch Fable go dark for two weeks, and expect American enterprises to keep paying frontier premiums for workloads that Chinese open weights handle at a tenth of the cost. Congress is now probing the adoption it helped create. The fix isn't another export order — it's an American open-weights strategy that doesn't force every CFO to expense DeepSeek.