Chinese AI models now handle nearly half of U.S. router traffic

Layer3Labs says Chinese-developed models now process as much as 46% of token traffic on a major U.S.-based AI router, upending how American businesses handle sensitive data. The firm warns most companies do not know which model is reading their data because AI features are often routed through vendors, resold, or used outside IT oversight.

Why it matters: - Layer3Labs says a growing share of U.S. business AI traffic is being processed by Chinese-developed models, creating data-sovereignty, compliance, and security risk. - The company says the exposure is often invisible to business owners because they buy an AI feature, not the underlying model. - High shadow AI exposure is linked to an estimated $670,000 breach premium per incident.

What happened: - Layer3Labs said as much as 46% of token traffic on the largest U.S.-based AI model router is now routed to Chinese models. - The firm said U.S. models handled about 70% of that traffic a year ago and are now closer to 30%. - The Miami-based AI implementation firm said it audited dozens of small- and mid-sized business AI stacks and found 70% contained at least one Chinese model the company had not knowingly approved. - Jonathan Teplitsky, Layer3Labs CEO, said most business owners believe their data stays on U.S.-based servers, but large volumes are being run through Chinese-provided endpoints. - Layer3Labs is offering a free AI Stack Sovereignty Check to identify which models are processing company data and flag sovereignty and compliance risk.

The details: - Layer3Labs said routing defaults can send prompts to the cheapest available provider, and those endpoints can be Chinese inference engines that are 60% to 90% cheaper. - The firm cited Andreessen Horowitz estimates that 80% of U.S. startups build on Chinese base models. - Named adopters in that estimate include Airbnb, Uber, and Cursor. - Layer3Labs said 70% of enterprise AI activity occurs outside IT oversight, much of it through Chinese models that are small enough to run on a laptop. - The company said any prompt processed by a Chinese-provider endpoint falls under China’s National Intelligence Law, which can compel data disclosure to the state. - Layer3Labs pointed to a July 2026 House Select Committee on the CCP probe into U.S. firms’ use of these models. - The firm also noted that the NDAA already bars DeepSeek from Defense Department systems. - Mark Boyhous said: