Companies turn to Chinese AI models to cut costs

There's a quiet migration happening inside corporate IT budgets, and it's not the one Washington wanted. According to Sam Bresnick, a research fellow at Georgetown's CSET, companies worldwide are increasingly routing AI workloads to Chinese models instead of paying premium rates for Anthropic or OpenAI systems.

The logic is unglamorous but hard to argue with. Bresnick, speaking to the Financial Times, framed it as a simple cost question: why pay a premium when Chinese models are generally workable for a large share of what enterprises actually need? Most business AI use isn't cutting-edge reasoning or novel research. It's summarizing documents, drafting emails, classifying support tickets. For that kind of volume work, a model that's 90 percent as good but a fraction of the price wins on spreadsheets alone.

Open-weight availability adds another layer of appeal. Companies that want to run models on their own infrastructure, tweak them for specific tasks, or avoid vendor lock-in have found Chinese labs more willing to release weights openly than their American counterparts, who've mostly kept their frontier models locked behind APIs. That flexibility matters more to enterprise buyers than marketing decks tend to suggest.

None of this means Chinese models have caught up at the very top end. But the gap has narrowed enough, and the price difference is large enough, that a lot of procurement decisions are starting to tilt east. Bresnick's comments land at a moment when the AI race is often described purely in terms of who has the smartest model. The market, at least for now, seems to be answering a different question: who has the cheapest model that still does the job.