The Sequence Opinion #831: NVIDIA Is Quietly Building the Operating System of AI
From Atoms to Agents
There’s a pattern in computing that repeats every few decades. A hardware company ships a breakthrough chip. Developers write software for it. Then the hardware company realizes the software layer is where the real lock-in lives, and starts building it themselves. Intel did it with compilers. Apple did it with iOS. NVIDIA is doing it right now with AI — except they’re doing it at a scale and speed that makes the previous examples look quaint.
At GTC 2026 last week, Jensen Huang stood on stage for two hours and barely talked about GPUs. He talked about an operating system for AI factories. He talked about agentic AI platforms. He talked about inference serving frameworks and enterprise security stacks and robot foundation models. The GPU — the thing that made NVIDIA a trillion-dollar company — was almost an afterthought. Buried under seven new chips, five rack-scale systems, and a software ecosystem so vertically integrated it makes Apple look open.
This is the thesis: NVIDIA has gone from selling the best pickaxes in the gold rush to owning the mine, the refinery, the supply chain, and the storefront. And the most underappreciated part of this transformation is the software. To see why, you have to look at what shipped across two consecutive GTCs — 2025 and 2026 — and understand how the releases fit together into a coherent platform play.