To Beat China, Embrace Open-Source AI

 

“China is making strides in open-source artificial intelligence. Eighty percent of developers worldwide who use open-source AI tools are building with Chinese models, according to an estimate by our colleague Martin Casado, general partner at Andreessen Horowitz. Research from our firm and OpenRouter shows a significant increase in the use of Chinese open models last year, reaching in some weeks a high of 30% of all AI usage.

 

In January, Alibaba's Qwen family surpassed 700 million downloads to become the most widely adopted open-source AI system on the planet.

 

This didn't happen because China out-engineered the U.S. It happened because U.S. policymakers spent two crucial years treating open-source AI as a threat.

 

After ChatGPT launched in late 2022, a wave of proposals assailed open development as dangerous. Prominent voices compared open-source AI models to nuclear weapons. Legislators floated broad licensing regimes and restrictions aimed specifically at open releases. California's SB 1047 would have required open-source developers to monitor and control downstream uses of their models. That clashes with one of the primary features of open source, namely that anyone can build on it. The message to American developers was clear: Choose open source, and you may face regulatory risk that your proprietary competitors won't.

 

Predictably, fewer American developers chose open source. And while American open-source development stalled amid regulatory uncertainty, Chinese open models filled the vacuum.

 

This matters both for fostering competition and for creating a healthy, trusted AI ecosystem. Open-source models let independent experts test systems, identify risks and build tools that benefit the entire ecosystem. They lower barriers to entry for startups and researchers, and they speed development by allowing users to experiment and improve systems together.

 

The policy tide has since turned. The National Telecommunications and Information Administration concluded that the marginal risks of open models don't justify restrictions. The Bureau of Industry and Security reached a similar conclusion on export controls. The current White House has been supportive, putting forth an AI Action Plan that encourages open-source development. There is now bipartisan consensus that open-source AI matters for American competitiveness.

 

That shift in tone matters. But to close the gap with China, policymakers need to go further. The U.S. government should demonstrate it is invested in the success of open-source AI development. Two moves would do that.

 

First, the government should practice what it preaches. When federal agencies build AI tools, they should release them under open-source licenses whenever possible. This would extend existing policies for traditional software. The Obama administration's "open by default" policy required agencies to release source code to the public. Congress codified this principle in the Share IT Act, which President Biden signed in December 2024. And this past January the General Services Administration updated its Open Source Software Policy to require that new custom code be developed in public repositories from day one.

 

These policies reflect the simple logic that when the public funds the development of software, the public should benefit from it. If agencies are building AI tools with taxpayer dollars, defaulting to open release also strengthens the broader ecosystem, giving startups and researchers building blocks they can't afford to create from scratch. This also tells the market that the era of treating open-source AI with suspicion is over.

 

Second, the federal government should stop favoring proprietary vendors in procurement. Current processes disadvantage open-source providers through requirements that name proprietary products while failing to take notice of the total cost of ownership and ignoring the problem of vendor lock-in. These aren't intentional bans on open source. They are defaults that quietly favor incumbents.

 

Solving this problem doesn't require a mandate that agencies always buy open source. It simply means ensuring that agencies give open-source solutions a fair look. They should consider the benefits of interoperability and open application programming interfaces. They should evaluate long-term costs -- including support, upgrades, and data migration -- for which open-source tools often have a significant advantage over proprietary alternatives. And they should stop writing tenders that are compatible only with proprietary solutions.

 

This is a pro-competition reform, not a subsidy. The Pentagon has recognized for years that open-source solutions deserve evaluation on equal footing. Extending that principle to AI procurement across the federal government would simply level a tilted playing field.

 

The question facing policymakers is no longer whether open-source AI will define the next phase of global AI development. It is whether the open layer that the world builds on will be American or Chinese. Policy uncertainty held American open-source developers back while the competition surged ahead. Closing that gap requires the government to show up as a builder, a buyer and a champion.

 

---

 

Mr. Ramaswamy is chief legal and policy officer and Mr. Perault head of AI policy at Andreessen Horowitz.” [1]

 

1. To Beat China, Embrace Open-Source AI. Ramaswamy, Jai; Perault, Matt.  Wall Street Journal, Eastern edition; New York, N.Y.. 18 Apr 2026: A13.