The birthplace of American tech innovation is quietly outsourcing one of its most critical capabilities. Silicon Valley startups are increasingly swapping expensive US-built AI models from OpenAI and Anthropic for cheaper, open-source alternatives developed by Chinese labs like DeepSeek, Alibaba, and Moonshot AI.

According to Martin Casado from Andreessen Horowitz, about 20–30% of startups use open-source models, and roughly 80% of those use Chinese open-source models.

The math is hard to argue with

Chinese AI models are increasingly winning over startups and developers as businesses seek to reduce soaring AI costs, raising questions about whether adoption could become as important as raw model capability in the global AI race.

Chinese AI models can cost 60% to 90% less than leading US frontier models while delivering comparable performance across many common tasks, according to OpenRouter data and related reports.

Many firms now use premium models only for demanding workloads while relying on cheaper models for everyday tasks such as document summaries, customer support and data processing.

Pricing remains a major advantage. DeepSeek’s flagship model costs around $0.87 per million output tokens, compared with roughly $25 for Anthropic and $30 for OpenAI. This makes it an increasingly attractive option for budget-conscious developers.

Developer platforms are already seeing the shift. Vercel reported DeepSeek’s share of AI traffic rose from under 1% to 17% since May, while OpenRouter said the model became its most popular offering after usage doubled between January and June. Chinese models from Xiaomi, MiniMax and Tencent also posted steady gains, largely at the expense of Google and OpenAI.

The trend has already influenced customer behavior. Startup Lindy said switching from Anthropic to DeepSeek saved millions of dollars after AI costs surpassed employee costs, while firms such as Airbnb and Anysphere have incorporated Chinese models into their AI stacks.

Why this matters beyond Silicon Valley

Andreessen Horowitz has publicly acknowledged the trend, noting that many startups are now pitching products built on Chinese open-source foundations.

Most startups aren’t going all-in on a single provider. The emerging playbook is a hybrid approach, combining Chinese models for cost-sensitive tasks with US models for situations where specific capabilities or compliance requirements demand them.

What this means for investors

US AI companies that have built their businesses around premium API pricing face a genuine pricing squeeze from below. When an open-source alternative delivers 85-90% of the performance at a fraction of the cost, most startups will take that deal.

The geopolitical dimension adds uncertainty. US regulators have already shown willingness to restrict Chinese technology access in semiconductors. If Chinese AI models become the backbone of American startup infrastructure, policymakers may feel compelled to act through export controls, data sovereignty rules, or outright bans, creating sudden disruption for companies that have built their stacks on these models.