Silicon Valley’s hottest AI startups are increasingly building their products on free, open-source AI models developed in China, signaling a subtle but significant shift in the U.S. artificial intelligence landscape. Once dominated by proprietary systems from OpenAI, Anthropic, and other American tech giants, the AI frontier is now seeing Chinese models like DeepSeek’s R1 and Alibaba’s Qwen gain traction for their affordability, flexibility, and growing sophistication.
Experts note that these open models, once considered far behind, are rapidly approaching the performance of leading U.S. systems, offering startups a practical alternative to expensive closed-source AI. As more American companies adopt these tools, questions arise about the future of domestic innovation and whether Silicon Valley’s reliance on foreign open models will reshape the global AI race.
Read More: Xsolla and HP Join Forces to Cultivate the Next Generation of Global Game Talent
U.S. Startups Embrace Open Chinese Models
Earlier this year, Misha Laskin, a theoretical physicist and machine learning engineer who contributed to some of Google’s most advanced AI models, surveyed the state of artificial intelligence in America—and grew concerned. Laskin observed a growing trend among American AI companies: the adoption of free, customizable, and increasingly powerful “open” AI models. The twist? Most of these models were being developed in China, and they were rapidly catching up to U.S. competitors.
“These models were not that far behind the frontier,” Laskin said. “In fact, they were surprisingly close. The ones that are coming now—well, they’re palpably close to the frontier.”To provide an American alternative, Laskin founded Reflection AI, a startup recently valued at $8 billion, focused on creating open-source AI models to compete with the Chinese systems gaining traction in Silicon Valley.
“You’re starting to see glimpses of open-model companies actually driving the frontier of intelligence in China—and overall, the frontier of intelligence,” he added.
Chinese Models Gain Traction in Silicon Valley
Over the past year, a growing number of U.S. AI startups have turned to Chinese open-source AI models as the backbone of their products. While American companies continue to lead at the cutting edge of AI capabilities, Chinese systems have become an attractive option: they are cheaper, more customizable, and now sufficiently powerful for many practical applications.
NBC News interviewed over 15 AI startup founders, machine-learning engineers, investors, and industry experts. Their consensus: while U.S. models still set the pace in terms of raw performance, Chinese open models increasingly serve as a practical and cost-effective foundation for innovation.
Cost, Speed, and Flexibility
Michael Fine, head of machine learning at Exa—a $700 million AI-focused search company backed by Lightspeed Venture Partners and Nvidia—explained the operational advantages of Chinese models. Running these models on Exa’s own infrastructure is often faster and less expensive than using larger, proprietary systems like OpenAI’s GPT-5 or Google’s Gemini.
“What often happens is we’ll get a feature working with a closed model and realize it’s too expensive or too slow,” Fine said. “Then we ask, ‘What levers do we have to make this faster and cheaper?’ That usually means replacing the closed model with the equivalent open model and running it on our own hardware.”
Open vs. Closed AI Models
Chinese AI models, such as DeepSeek’s R1 and Alibaba’s Qwen, are free to use and labeled “open-source” or “open-weight,” meaning anyone can download, modify, and operate them. In contrast, American systems like OpenAI’s GPT models or Anthropic’s Claude remain proprietary, accessible only via controlled cloud platforms.
Historically, U.S. closed-source models outperformed both domestic and Chinese open alternatives. Even well-resourced projects, like BloombergGPT—a tool developed by Bloomberg using internal open-source models—lagged behind OpenAI’s offerings in financial intelligence.
Closing the Gap
However, Chinese AI companies have rapidly advanced in the past year. Open-source products from DeepSeek and Alibaba now approach—or in some cases match—the performance of leading closed American models, according to Artificial Analysis, an independent AI benchmarking firm.“The gap is really shrinking,” said Lin Qiao, CEO of Fireworks AI and co-creator of PyTorch, the leading framework for AI model training.
He emphasized that Chinese open-source models are becoming increasingly competitive with U.S. closed-source alternatives. The shift toward open Chinese AI raises critical questions for the U.S. industry. With billions invested in companies like OpenAI and Anthropic under the assumption that American AI will dominate globally, the growing reliance on foreign open-source models could reshape both innovation strategies and competitive dynamics.
Frequently Asked Questions
What is open-source AI?
AI that anyone can download, modify, and run on their own systems.
Why are U.S. startups using Chinese models?
They’re cheaper, faster, and more customizable than U.S. proprietary models.
How do Chinese models compare to American AI?
They are rapidly improving and now approach the performance of top U.S. systems in many areas.
Does this mean U.S. AI is falling behind?
Not entirely; U.S. companies still lead frontier research, but open models are increasingly competitive.
Are there security concerns?
Yes—foreign models can raise data privacy, IP, and regulatory risks.
What does this mean for AI’s future?
AI is becoming more global and flexible, challenging the dominance of proprietary U.S. systems.
Conclusion
As Silicon Valley increasingly leans on free Chinese AI models, the U.S. AI landscape is at a pivotal moment. Open-source systems from China are no longer just alternatives—they are practical, powerful, and cost-effective tools driving innovation. While American companies continue to push the frontier of AI research, the growing adoption of foreign open models highlights a strategic shift: success may depend as much on accessibility and adaptability as on sheer technological superiority.
