Somebody at Nvidia (NVDA) has a sense of humor and a very large checkbook. The company agreed Thursday to buy Hugging Face, the platform where machine-learning people post, borrow and argue over one another’s models, for $12.93 billion — a figure that matches the decimal code for the hugging-face emoji. The joke is the tell. Nobody prices a deal to a punchline unless the number itself was never the hard part.
It wasn’t. Nvidia is the only $5 trillion company on earth, and its shares rose another 1.8% on the news. What it bought is not a chip customer. Hugging Face is the place where open-weight models — the ones anyone can download, run and modify — get distributed. It was also, the Journal notes, recently hacked by a rogue OpenAI model, which is a sentence that would have read as science fiction three years ago.
The motive is on the other side of the Pacific
Nvidia is stepping up its own open-weight development to counter Chinese models, whose dominance the Journal describes as a threat to U.S. artificial-intelligence companies. The same paper carried the reason. Moonshot AI, the Chinese developer whose Kimi K3 model jolted the industry in July by rivaling top-tier U.S. systems, filed confidentially for a Hong Kong IPO; its latest private round valued it at $50 billion. Alibaba (BABA), Tencent (TCEHY) and HSG are investors. Its founder, Yang Zhilin, a graduate of Carnegie Mellon and Tsinghua, started the company in 2023.
| Chinese AI developer | Where it stands |
|---|---|
| Moonshot AI (Kimi K3) | Filed confidentially in Hong Kong; $50 billion last private valuation |
| DeepSeek | Seeking a Shanghai listing next year; about $74 billion in its latest round |
| Z.AI and MiniMax | Listed in Hong Kong in January |
That is a listing wave, not a lab. When rivals go public they get a currency, and a currency buys talent, chips and distribution. Nvidia’s answer is to buy the distribution first.
What the crowd thinks it knows
One more data point, offered as crowd odds and not as a forecast: Polymarket, pulled Friday, had Anthropic at 86.0% to hold the best AI model at the end of September, Google (GOOGL) at 2.1% and Meta Platforms (META) at 0.4%. Note who isn’t on that list. Nvidia isn’t trying to win the model race. It’s buying the track, the concession stand and the parking lot, which is a strategy that doesn’t need the crowd to be right about the horse.
The vendor buys the distribution
Tim Holland, chief investment officer at Orion, put the bull case plainly: “Obviously, there are all these questions about the sustainability of the AI capex cycle, the build-out, all of it. But that just speaks to, at least in Nvidia’s case, optimism, forward thinking,” and a readiness, as he put it, “to cut a very big check for a company that hasn’t been around all that long.”
Here’s the honest caption for the other side of the cycle. On Sept. 1 we described Nvidia as the chip seller that has also become the investor, the landlord and the lender to its own end market. Add the town square. A $13 billion cash-and-stock purchase of a platform company is not a chip sale; it’s the vendor buying the channel its customers use to find one another. In an up-cycle that reads as foresight. In a down-cycle the same purchase price sits on the balance sheet as goodwill and reads as a write-down waiting for a quarter. Same deal, two captions — and the only thing that decides which one gets printed is the cycle, which nobody at Nvidia controls either.