Key insights
- The author argues NVIDIA's open-source software strategy (Dynamo, Nemotron, GR00T, Isaac) is creating a deep ecosystem lock-in, similar to CUDA. This ensures NVIDIA's dominance in future trillion-dollar industries like physical AI, robotics, and autonomous systems. The market is underpricing NVIDIA by focusing solely on near-term GPU demand, missing the long-term strategic advantage. This is bullish for NVDA and the broader tech sector.
People are still debating whether NVIDIA's valuation is justified based on data center GPU demand. I think that's the wrong lens entirely. GTC 2026 made something much bigger visible, and it is something that already happened before.
In 2006, NVIDIA released CUDA with developer tools, libraries, documentation — all of it, no charge. A generation of researchers and engineers built their careers on CUDA. Universities taught it. Companies standardized on it. By the time competitors realized what had happened, the switching cost wasn't a price — it was a decade of institutional knowledge that couldn't be replicated.
GTC 2026 celebrated CUDA's 20 yearsi.
Dynamo 1.0 — the inference operating system for AI factories — is free and open source, and it boosts Blackwell GPU performance by 7x. Nemotron models are open. GR00T for robotics is open. Isaac simulation frameworks are open. The Nemotron Coalition is co-building frontier models with Mistral, Perplexity, LangChain and others, and open sourcing the results.
NVIDIA is once again being generous with software, and for exactly the same reason as before.
They're enrolling the next generation.
The robotics engineers building on Isaac today are the computer vision researchers who built on CUDA in 2012. The autonomous vehicle teams standardizing on DRIVE Hyperion are the deep learning labs that standardized on cuDNN in 2014. NVIDIA isn't giving away software — they're making sure that when physical AI, robotics, and autonomous systems become trillion-dollar industries, every engineer in those fields learned on NVIDIA tools, every model was trained on NVIDIA infrastructure, and every company's stack runs natively on NVIDIA hardware.
Competitors can read the Dynamo source code. What they can't do is compress 15 years of ecosystem compounding into a product cycle. By the time a competitor reaches parity on one layer, NVIDIA has already moved two levels higher.
The market prices NVIDIA on near-term GPU demand. That's a legitimate short-term lens, and it'll drive volatility. But the actual thesis is this: NVIDIA is laying the infrastructure foundation for every physical AI breakthrough of the next decade — robots, autonomous vehicles, orbital data centers, distributed edge compute across 5G networks — and they're doing it the same way they captured deep learning: by making their platform the path of least resistance for every serious developer and researcher on the planet.
That's not a GPU company with a good product cycle. That's a toll booth on the next industrial revolution.
I know it is not cheap, but could NVIDIA be considered a value investment based on its moat and strong cash sheet? Munger supported investing in good business at fair value, could this be one of those cases?