Silicon Valley Startups Increasingly Build Products on Free Chinese AI Models

American artificial intelligence startups are increasingly building their products on free, open-source Chinese AI models that now rival expensive proprietary systems from OpenAI and Anthropic, according to interviews with over fifteen founders, engineers, and industry experts conducted by NBC News.

The shift represents a significant evolution in the global AI landscape, where Chinese companies like DeepSeek and Alibaba have released open-source models that closely approach or match the performance of leading American closed systems across many use cases, while offering substantial cost advantages and greater customization flexibility.

Misha Laskin, a former Google machine learning engineer who founded $8 billion startup Reflection AI, expressed concern about the trajectory earlier this year. Chinese open-source models were "surprisingly close to the frontier," he noted, adding that newer releases are "palpably close to the frontier" of AI capabilities.

Independent benchmarking by Artificial Analysis confirms the convergence. DeepSeek's R1 and Alibaba's Qwen now closely approach or match leading American closed models in many domains, according to tracked performance metrics.

Lin Qiao, CEO of Fireworks AI and co-creator of PyTorch, the dominant AI training framework, said bluntly: "The gap is really shrinking."

Michael Fine, head of machine learning at $700 million AI search company Exa, explained the economic reality driving Chinese model adoption. When features prove too expensive or slow using GPT-5 or Gemini, his team asks: "What levers do we have to make this faster and cheaper?"

"That usually means replacing the closed model with the equivalent open model and then running it on our own infrastructure," Fine said.

Jerry Liu, founder of productivity app Dayflow, estimates roughly 40 percent of users now choose open-source models over proprietary alternatives. For tasks like screen description, Alibaba's Qwen performs as well as GPT-5, he said, while costing dramatically less. Closed-model usage can cost Dayflow up to $1,000 per person, making cheaper alternatives essential for viability.

Developer resources also favor Chinese models. Antonio Vespoli, co-founder of browser agent startup Circlemind AI, noted that Chinese models now dominate online developer resources with abundant training guides and community support. Among the twenty top models used by coding app Kilo Code, seven are Chinese systems, six of which are open-source.

Airbnb CEO Brian Chesky confirmed in October 2025 that his company "heavily" relies on Qwen.

Beyond cost savings, open-source models offer privacy advantages through on-device processing. Liu emphasized this appeal: "Would I use a product where my entire screen was beaming up to some random guy's cloud? Hell no."

Charles Zedlewski, chief product officer at Together AI, said developers increasingly start from open models and customize them with proprietary data, adding capabilities unavailable in existing systems.

The trend raises questions about American AI dominance. Investors have wagered tens of billions on OpenAI and Anthropic, betting on continued American leadership. Growing Chinese model adoption challenges those assumptions.

Security concerns persist. The US Center for AI Standards and Innovation released a September 2025 report finding weakened safety protocols and increased pro-Chinese outputs in DeepSeek models compared to American alternatives. A recent White House memo accused Alibaba of supporting China's military, which Alibaba called "complete nonsense."

Some observers suggest Chinese progress stems from copying American foundational work rather than independent innovation.

Recognition of the challenge has prompted action. OpenAI released its first open-source model in August 2025 after five years, explicitly citing the importance of American open-source development for "democratic AI."

The Allen Institute launched its ATOM Project—American Truly Open Models—declaring that "America has lost its lead in open models" and cannot "cede such a critical piece of the ecosystem to any nation."

Reflection AI's $8 billion valuation reflects investor belief in American open-source alternatives.

As Nathan Lambert, senior research scientist at the Allen Institute, concluded: "The balance of power has been shifting rapidly in the last 12 months."

The trajectory suggests American AI companies face a strategic inflection point: adapt their business models to compete with free, capable alternatives, or risk losing developer mindshare to Chinese ecosystems. With billions invested in proprietary closed-source development, the rise of Chinese open-source models challenges fundamental assumptions about AI's commercial future and America's technological leadership.