Why Silicon Valley Is Quietly Turning to Free Chinese AI Models

Silicon Valley

America’s AI boom is in full swing. Startups are hitting sky-high valuations, investors are pouring in billions, and new AI products launch almost daily. But behind the scenes, a surprising shift is taking place: a growing number of Silicon Valley companies are quietly building their technology on top of free, open-weight Chinese AI models.

For Misha Laskin a theoretical physicist and machine-learning engineer who helped shape Google’s most powerful AI systems this trend set off alarm bells. Earlier this year, as he surveyed the U.S. AI landscape, something felt off. More American AI founders were choosing China-made open models instead of expensive U.S. alternatives.

And those Chinese systems, Laskin noted, were no longer lagging by a wide margin.
“They were surprisingly close to the frontier,” he said. “And the ones coming now they’re palpably close.”

Concerned, Laskin launched Reflection AI, an $8 billion startup aimed at creating a strong American open-source competitor to China’s rapidly advancing models.

“You’re starting to see open-model companies in China pushing the frontier of intelligence,” he warned.

Why Silicon Valley Is Turning to Chinese AI

Over the past year, open-weight Chinese models have surged in capability while remaining free, customizable, and easy to download a combination that appeals strongly to cash-conscious startups. NBC News interviewed more than 15 engineers, founders, researchers, and investors. Their consensus: U.S. companies like OpenAI and Anthropic still lead at the cutting edge, but for many real-world needs, Chinese models are now “good enough” and far cheaper.

Models like DeepSeek’s R1 and Alibaba’s Qwen are:

  • Free to use
  • Highly customizable
  • Lightweight enough to run on local hardware
  • Increasingly competitive with U.S. closed models

At Exa, a $700 million AI search company backed by Nvidia and Lightspeed, head of machine learning Michael Fine says the economics are straightforward:

“We test a feature with a closed model, realize it’s too slow or too expensive, and then replace it with an equivalent open model,” Fine explained. Running Chinese models on Exa’s own hardware is significantly cheaper and faster than tapping into OpenAI’s GPT-5 or Google’s Gemini.

This pattern is spreading across the industry.

Startups Choosing Open Chinese Models – and Why It Matters

Many founders told NBC that Chinese models now rival U.S. models for tasks like:

  • Summaries
  • Image description
  • Productivity workflows
  • Reasoning tasks on smaller datasets

Jerry Liu, founder of the app Dayflow, estimates 40% of users choose Chinese open-source models inside his app. For describing screenshots and performing simple tasks, he says Alibaba’s Qwen performs nearly as well as GPT-5 – but costs a fraction.

More importantly, Qwen can run locally on a user’s device, offering a privacy advantage.
“Would I use a product where my entire screen was streaming to some random cloud?” Liu said. “Hell no.”

Beyond cost and privacy, open models thrive because of their growing developer ecosystem – tutorials, GitHub repos, communities, tools, and pre-built resources that speed up development.

China’s Open-Source Strategy – and Why It’s Working

Unlike the U.S., where AI is dominated by private companies and closed systems, China has leaned heavily into open-source development. The Chinese government has openly endorsed open-source as an AI priority. President Xi Jinping has called for international cooperation on open technologies, and national economic bodies have pledged support for open-model ecosystems.

Chinese companies release new models at rapid speed. This year:

  • Alibaba released a new model roughly every 20 days
  • Anthropic averaged a new model every 47 days

And China’s AI progress is no accident, says Nathan Lambert of the Allen Institute for AI:

“The Chinese are genuine innovators. The balance of power has shifted rapidly.”

Some analysts argue Chinese labs have benefited from building on U.S. research, but others point out that Chinese models are now improving at a pace that can no longer be dismissed as “copying.”

Despite the momentum, many companies hesitate to use Chinese systems due to:

  • Geopolitical concerns
  • Regulatory uncertainty
  • Perception risks with enterprise buyers
  • U.S. government warnings

A recent report from the U.S. Center for AI Standards flagged safety issues in DeepSeek models. A White House memo also raised concerns about Alibaba’s ties to China’s military claims Alibaba called “complete nonsense.”

There’s also the belief among some investors that closed U.S. models are simply more powerful and offer a more robust ecosystem.
As Menlo Ventures partner Tim Tully argued:

“The tooling is better. The productivity is better. The agent frameworks are better. Closed models just work better.”

Is America Losing the Open-Source Race?

Some AI experts now believe that America’s lack of powerful open-source models is a national security issue. Meta once led the U.S. open-source movement with Llama, but even Mark Zuckerberg has hinted that not all future “superintelligence” models will be open.

Meanwhile, Chinese open-source models continue to surge in adoption.

But the tide may be turning.

  • In July, the White House announced a plan to “encourage open-source and open-weight AI.”
  • In August, OpenAI released its first open-source model in five years, saying American models must remain competitive to “expand democratic AI.”
  • In November, the Allen Institute released Olmo 3, a major new U.S. open-source model.
  • Researchers launched ATOM “American Truly Open Models” calling open models critical to national AI leadership.

“If we want to lead in the AI era, we cannot cede this part of the ecosystem,” said Lambert, who founded ATOM.

The Bigger Question: Who Will Shape the Future of AI?

The quiet adoption of Chinese open-weight models across Silicon Valley reveals a deeper tension within the AI world:

  • Closed vs. open
  • Expensive vs. free
  • Domestic vs. global models
  • Corporate dominance vs. public access

Chinese models are rising fast and reshaping the AI strategies of American startups. The United States is now reckoning with whether it can afford to let China dominate the global open-source AI ecosystem.

Because whichever country leads in open models may end up shaping not just the AI industry but the future of technology itself.

Leave a Reply

Your email address will not be published. Required fields are marked *