Why open source has become an American national priority

When President Trump published the US action plan for AI last week, many were surprised to “encourage open-source and open AI”, as one of the main priorities of the administration. The White House raised what was once a very technical subject in an urgent national concern – and a key strategy to win the AI race against China.

The emphasis put by China on open source, also underlined in its own action plan published shortly after the United States, makes the Open-Source race imperative. And the soft power which is accompanied by more open models from China makes their recent leadership even more remarkable.

When Deepseek-R1, a powerful model of large open source language (LLM) outside China, was published earlier this year, he did not come with a press tour. No flashy demos. No linked speeches. But it was open weights and open sciences. Open weight means that anyone with the right skills and IT resources can execute, reproduce or make a model theirs; Open Science shares some of the tips behind the development of the model.

In a few hours, researchers and developers seized it. In a few days, it has become the most appreciated model of all time By hugging the face – with thousands of variants created and used in large technological companies, research laboratories and startups. More striking, this adoption explosion occurred not only abroad, but but in the United States For the first time, the American AI was built on Chinese foundations.

Deepseek was not the only one

In one week, the US stock market – feeling the tremor – took a fall.

It turns out that Deepseek was only the act of opening. Dozens of Chinese research groups now push the borders of Open Source, not only sharing powerful models, but data, code and scientific methods behind them. They move quickly – and they do it in the open air.

Meanwhile, American companies – many of which were the pioneers of the Revolution of Modern AI – are closing more and more. Highlight models like GPT-4, Claude and Gemini are no longer released so as to allow manufacturers to control more control. They are accessible only via chatbots or APIs: closed interfaces that allow you to interact with a model but do not see how it works, recycle or use it freely. The weights, training data and model behavior remain owners, closely controlled by some technological giants.

It is a dramatic reversal. Between 2016 and 2020, the United States were THE World leader in Open Source. Google’s research laboratories, Openai, Stanford and elsewhere have published revolutionary models and methods that have laid the foundations for everything we now call “AI”. The transformer – The “T” in Chatgpt – was born from this open culture. An embroidered face was created at that time to democratize access to these technologies.

Now the United States slips and the implications are deep.

Scientists, startups and American institutions are increasingly pushed to rely on Chinese open models because the best American models are locked behind the APIs. While each new open model emerges from abroad, Chinese companies like Deepseek and Alibaba strengthen their positions as fundamental strata in the World Ecosystem of AI. The tools that feed the next generation of products, research and infrastructure of America AI come from more and more overseas.

And at a deeper level, there is a more fundamental risk: every advancement of the AI – including the most closed systems – is built on open foundations. Owner models depend on open research, processors’ architecture to training libraries and evaluation frames. But more importantly, the opening increases the speed of a country in the construction of the AI. It feeds rapid experimentation, reduces obstacles to entry and creates composition innovation.

When the opening slows down, the entire ecosystem follows. If the United States is late today in open source, it can be late in AI.

Get away from the AI black box

This does not only count for innovation, but for security, science and democratic governance. The open models are transparent and verifiable. They allow governments, educators, health establishments and small businesses to adapt to AI to their needs, without locked outbuildings or black box dependencies.

We need more and better open source models developed in the United States. American institutions that are already pressure for opening must rely on their success. Meta’s Llama’s Llama family has led to tens of thousands of variations on the embraced face. The Allen Institute for AI continues to publish excellent fully open models. Promising startups like the black forest build open multimodal systems. Even Openai suggested that he could soon release open weights.

With more public and political support for open source AI, as shown in the American AI action plan, we can restart a decentralized movement that will ensure American leadership. It is time for the American community of AI to wake up, to drop the “open story is not sure” and to return to its roots: open science and open-source AI, propelled by an unequaled community of border laboratories, large technologies, startups, universities and non-drunk.

We can restart a decentralized movement which will ensure American leadership, built on opening, competition and scientific research, and empower the next generation of manufacturers. If we want the AI to reflect democratic principles, we must build it outdoors. And if the United States wants to lead the AI race, it must lead the Open Source race.

Clément Delangue is the co-founder and CEO of Hugging Face.

Daily information on business use cases with VB daily

If you want to impress your boss, VB Daily has covered you. We give you the interior scoop on what companies do with a generative AI, from regulatory changes to practical deployments, so that you can share information for a maximum return on investment.

Read our privacy policy

Thank you for subscribing. Discover more VB newsletters here.

An error occurred.