I just realized something very important in the modern chip industry. There was an exciting panel discussion at GTC where Bill Dally from Nvidia talked about something that changed the way we think about chip design entirely.

It all started with some really frightening numbers — previously, transferring a library of standard cells containing thousands of cells required a team of 8 engineers working for 10 full months. Now? A single GPU processor working overnight, and the task is done. The results are even better than human design in terms of efficiency and power consumption.

But the truth runs deeper than headlines. Nvidia didn’t use a random black box — there are advanced tools developed over years. The NB-Cell software relies on reinforcement learning, and there are large internal language models called Chip Nemo and Bug Nemo trained on everything from Nvidia’s history from G80 to Blackwell. This means a new employee can access 20 years of expert engineering experience with a single click.

But the smartest part of the strategy came afterward. In December 2025, Nvidia invested $2 billion in Synopsys — one of the largest chip design tool companies worldwide. They signed an agreement to deeply integrate Nvidia’s technology into full Synopsys tools. Shortly after, Cadence and other companies announced they are developing AI-powered tools on Nvidia GPUs.

The frightening numbers? Synopsys tools are 30 times faster on Blackwell, other tools are 20 times faster, and 12 times on other processors. The difference is huge.

Here’s where the real problem comes in. In the past, chip design tools ran equally well on Intel and AMD processors. The future is completely different — if you want the fastest tools, you have to buy Nvidia cards only. Imagine you’re an engineer at a competing company wanting to design a chip that outperforms Blackwell — you open the fastest design tool and find it runs optimally only on Nvidia processors. Either accept a design cycle twice as slow, or buy a large fleet of Nvidia cards to design a chip aimed at beating Nvidia itself.

The strategy is broader than that. Nvidia has covered every stage of the supply chain from design to manufacturing using AI. Chip Nemo handles front-end design, NB-Cell manages intermediate optimizations, EDA tools are tied into a $2 billion investment, and even optical calculations in manufacturing are done on Nvidia processors.

The bitter irony: any competitor wanting to beat Nvidia will find that all the tools needed to win are owned or optimized for Nvidia. The person you want to beat is giving you all the tools you need to try.

And the local Chinese companies trying to enter the GPU market? The situation is much worse. Most still use licensed Synopsys and Cadence tools, losing billions annually, but the market still considers them a “local Nvidia” with huge market valuations. The gap between valuation and reality is very frightening.