On November 14, 2026, noted AI researcher Nathan Lambert wrote in Interconnects that Nvidia CEO Jensen Huang's resignation is a major blow to the US open-model ecosystem: over the past two years, Nvidia's funding for open models has largely been a founder-style bet decided by Huang himself, and that money is very likely to dry up now that he is gone. Nvidia's top priority now is stabilizing the company and delivering to customers on schedule — open models are no longer on the list.
Lambert's argument rests on a window of time: the open-model economy is still in its infancy, and these are the crucial years for building use cases, revenue streams, and the motivation to keep training the next generation of models. The good news is that open source has found a way to live — 'ownership of intelligence' and enterprise post-training are a real lane. The hard part is that open source's true long-term value, making intelligence a cheap commodity, is still far from realized, while compute is at its scarcest. Inference startups can't even get chips, because Anthropic and OpenAI are buying up supply; if open-weight models also have to be served at high inference margins, they are far from being a commodity.
The Tighter Compute Gets, the More Closed Labs Win
In this landscape, the side with greater scale and integration wins: closed labs own a more complete chain, from models to data centers to chips. Lambert cites Dylan Patel, CEO of SemiAnalysis, who argues the problem is not money but talent — the very best people are mostly at Anthropic, OpenAI, DeepMind, and Meta, and the rest are scattered across a hundred startups. Thinking Machines Lab's newly open-sourced Inkling 2 Small is one of the best Western open-weight models in a while, yet it still trails Moonshot's Kimi K3.1 Pro by a clear margin; Mira's next funding round is reportedly done at a $40 billion valuation, but even that money is not enough for open source to match the spending pace of the closed frontier.
The bigger worry sits on the regulatory side. Lambert argues that the near-term risk is hasty AI regulation at the state level, followed by national regulation, pushed by people who understand too little about what open source is good for. He writes that since this summer, the US has been one strong Chinese open-weight model away from a crisis; no large-scale attack driven by a popular Chinese open model has happened yet, but it is a matter of time. The open camp needs someone to pick up the phone when a crisis hits and steady the ship — and for the past two years, that person was Jensen Huang.
Lambert's prescription is a single one: incubate a genuinely frontier open-model player in the United States. Who will take that call — Zuckerberg, Jassy, or Satya — is still unanswered; what is certain is that the days of relying on Jensen Huang alone are over, and the funding tap for the open ecosystem is tightening too. There is not much time left for it to grow a new pillar of its own.