When NVIDIA buys out the entire home of open source AI, what does "open" even mean here?
Those of you who play with AI surely know what Hugging Face is? If you don't know, this is basically a place that almost anyone who works with AI has visited at least once: a huge repository where developers upload and download open source AI models, meaning models whose construction is public, anyone can download and use or customize, unlike closed models like GPT or Claude. And according to technology news, Nvidia has agreed to buy Hugging Face for $12.9 billion, although the final agreement has not been officially signed and is still likely to fall apart. Actually, after reading the whole story, I see that this is not a simple matter of "big company buying small company". There is an interesting layer of contradiction underneath that you should know before considering this as just a normal sale.
Why Nvidia wants to buy a platform that is almost free
Hugging Face doesn't make a lot of money, this year's revenue is said to be about $150 million, a small number compared to the $12.9 billion price people are expected to pay. So what does Nvidia buy? The answer lies in the fact that almost the entire open source model is running on Nvidia hardware. While large closed AI labs like OpenAI, Google, Amazon and even Anthropic are developing their own chips to become less dependent on Nvidia, a healthy open source ecosystem is what keeps customers with Nvidia hardware, because the more open model options, the harder it is for the market to move away from the chip running behind all those models. Holding Hugging Face, Nvidia not only holds the hardware but also holds the gateway where most open models are distributed.
Just a year ago, Hugging Face rejected NVIDIA, now it agrees to sell itself for a higher price.
The second layer: the "open" story is being politicized
Reading further, I discovered another layer of context that is also worth mentioning. Hugging Face CEO Clem Delangue has publicly sided with the wave of campaigns supporting Nvidia's open source model this year, while Washington considers tightening regulations on the open-weight model, which also publicizes the trained parameters, not just the source code. The reason is partly due to the fact that Chinese laboratories such as Moonshot AI launched the Kimi K3 model with the same performance as leading American models with much cheaper operating costs, causing US officials to worry about both competition and national security. Nvidia, along with Hugging Face and 24 other companies, even signed a joint letter calling on the US government to support rather than restrict the open model.
China's low-cost model with high performance poses great political risks for the US To put it bluntly, when a company both advocates for "open" policies and prepares to be acquired by a closed hardware company, the line between advocating for ideals and advocating for the commercial interests of that hardware company becomes quite blurred.
Open question: does open source still retain its true meaning?
Looking a little broader, this is not a single deal. This same month, Stripe also acquired OpenRouter, a platform that helps users choose between many different AI models, for more than $7 billion, even though the company was only valued at $1.3 billion in May. Looking at the two deals side by side, a clear pattern is being formed: the middle layer infrastructure of open source AI, places that were originally considered neutral, are being grouped together by giants with their own commercial interests. Question for those of you who still download models from Hugging Face to run at home or for work: when the place that stores and distributes most of the industry's open source models is owned by the very company that sells the hardware that runs those models, does "open source" still retain its neutral meaning as people still imagine, or is it now just a customer retention strategy wrapped in the ideal of free and open?