
(SeaPRwire) – By: Oliver Hawthorne
Nvidia owns 85 percent of the AI chip market. That number only has one direction to go. The real threat is hiding inside its own customer list. Google now trains Gemini exclusively on custom TPU chips. OpenAI’s Jalapeño chip, according to SemiAnalysis testing, beats every Nvidia, AMD, and Google chip they have benchmarked. Anthropic hired Amir Salek, the former TPU team lead at Google, to launch an in-house chip division. The hyperscalers are engineering their own escape from Nvidia’s pricing power. And Nvidia sees the wall approaching. This is not paranoia. This is arithmetic. When your three largest customers are each actively designing replacement silicon, the monopoly is already ending in slow motion. The only question left is whether Nvidia can recruit new buyers fast enough to offset the ones it will eventually lose. The answer to that question explains the Hugging Face deal.
That is the context behind the reported $12.9 billion acquisition of Hugging Face. This is not a charitable donation to the open-source community. It is a strategic hedge, and a defensive one at that. Hugging Face is the hub where developers, startups, and governments download and share open AI models and datasets. In that world, compute demand spreads across thousands of smaller buyers. Few of them will ever have the scale to design custom chips. Nvidia has been pushing this alternative future hard for months. It lobbied Washington to loosen restrictions on selling chips to China, which leads in open AI development. It committed $6 billion to Poolside to build an American open alternative. In July, Nvidia helped lead an open letter defending open-source AI and urging Washington not to restrict it. Jensen Huang used his first post on X to argue that open models strengthen safety, accelerate innovation, and enable sovereignty. The existing business remains absurdly strong. August revenue hit $96.2 billion, more than doubling year over year and beating Wall Street expectations. Gross margins sit at 75 percent. Google, OpenAI, and Meta still buy Nvidia hardware in enormous volume even while designing their own silicon. And the math on custom chips is brutal for Nvidia. A chip does not need to beat Nvidia’s performance to save a hyperscaler money. At a 75 percent margin, it only needs to be decent. Broadcom is helping Google, OpenAI, and Meta turn their specifications into working silicon. Migration will not happen overnight. Sriram Viswanathan of Celesta Capital, a former Intel executive, calls it a “huge lift-and-pour-concrete situation.” The Apple precedent is the warning. Apple started with iPad and iPhone chips, kept buying Intel for Macs, and once its silicon matured, Intel was cut out entirely. The same playbook is now running against Nvidia, just with a longer timeline.
The commercial loop behind the Hugging Face deal is clear. Nvidia cannot stop Google or OpenAI from building their own chips. But it can prevent the rest of the market from following them. If open models become the default way AI gets built and deployed, the center of gravity shifts from a few centralized data centers to a long tail of independent buyers. Those buyers will not fabricate their own silicon. They will rent compute, and they will buy from Nvidia. The hyperscalers want to centralize AI inside their own walled gardens. Nvidia wants to decentralize it across the open web. Richard Clode of Janus Henderson frames it as a two-way strategic battle. Nvidia does not want to rely on three customers, so it deliberately finances and allocates chips to neo-clouds and smaller players. The hyperscalers do not want to rely on one compute provider, so they build custom silicon and partner with Broadcom. Umesh Padval of Seligman Ventures calls the deal a brilliant chess move. Sean Lie of Cerebras says the armor has cracks. All of them are right. The acquisition does not guarantee Nvidia wins the war. It just guarantees the war stays open. For a company with 75 percent margins and three customers racing to escape, keeping the war open is the entire game. Nvidia is executing like crazy, but execution alone will not stop the erosion. It needs an entirely new buyer base. Hugging Face is the funnel for that buyer base.
Author bio: Oliver Hawthorne, Principal Correspondent at an international technology review, covering semiconductor strategy, AI infrastructure, and the shifting economics of compute.