Marc Andreessen Says the US Needs to Lead Open-Sourced AI

Venture capitalist Marc Andreessen has a clear warning: America needs to get serious about open-source AI or risk ceding control to China.

“Just close your eyes,” the cofounder of VC firm Andreessen Horowitz said in an interview on tech show TBPN published on Saturday. “Imagine two states of the world: One in which the entire world runs on American open-source LLM, and the other is where the entire world, including the US, runs on all Chinese software.”

Andreessen’s comments come amid an intensifying US-China tech rivalry and a growing debate over open- and closed-source AI.

Open-source models are freely accessible, allowing anyone to study, modify, and build upon them. Closed-source models are tightly controlled by the companies that develop them. Chinese firms have largely favored the open-source route, while US tech giants have taken a more proprietary approach.

Last week, the US issued a warning against the use of US AI chips for Chinese models. It also issued new guidelines banning the use of Huawei’s Ascend AI chips globally, citing national security concerns.

“These chips were likely developed or produced in violation of US export controls,” the US Commerce Department’s Bureau of Industry and Security said in a statement on its website.

As the hardware divide between the US and China deepens, attention is also on software and AI, where control over the underlying models is increasingly seen as a matter of technological sovereignty.

Andreessen said it’s “plausible” and “entirely feasible” that open-source AI could become the global standard. Companies would need to “adjust to that if it happens,” he said, adding that widespread access to “free” AI would be a “pretty magical result.”

Still, for him, the debate isn’t just about access. It’s about values — and where control lies.

Andreessen said he believes it’s important that there’s an American open-source champion or a Western open-source large language model.

A country that builds its own models also shapes the values, assumptions, and messaging embedded in them.

“Open weights is great, but the open weights, they’re baked, right?” he said. “The training is in the weights, and you can’t really undo that.”

For Andreessen, the stakes are high. AI is going to “intermediate” key institutions like the courts, schools, and medical systems, which is why it’s “really critical,” he said.

Andreessen’s firm, Andreessen Horowitz, backs Sam Altman’s OpenAI and Elon Musk’s xAI, among other AI companies. The VC did not respond to a request for comment from Business Insider.

Open source vs closed source

China has been charging ahead in the open-source AI race.

While US firms focused on building powerful models locked behind paywalls and enterprise licenses, Chinese companies have been giving some of theirs away.

In January, Chinese AI startup DeepSeek released R1, a large language model that rivals ChatGPT’s o1 but at a fraction of the cost, the company said.

The open-sourced model raised questions about the billions spent training closed models in the US. Andreessen earlier called it “AI’s Sputnik moment.”

Major players like OpenAI — long criticized for its closed approach — have started to shift course.

“I personally think we have been on the wrong side of history here and need to figure out a different open source strategy,” Altman said in February.

In March, OpenAI announced that it was preparing to roll out its first open-weight language model with advanced reasoning capabilities since releasing GPT-2 in 2019.

In a letter to employees earlier this month announcing that the company’s nonprofit would stay in control, Altman said: “We want to open source very capable models.”

The AI race is also increasingly defined by questions of national sovereignty.

Nvidia’s CEO, Jensen Huang, said last year at the World Government Summit in Dubai that every country should have its own AI systems.

Huang said countries should ensure they own the production of their intelligence and the data produced and work toward building “sovereign AI.”

“It codifies your culture, your society’s intelligence, your common sense, your history — you own your own data,” he added.