Jen-Hsun Huang recently responded to the threat posed by Google's TPU. This was triggered by a report from Morgan Stanley, which predicts that Google will sell 12 million TPUs by 2027-28, nearly double the previous estimate. According to their Algorithm, every 500,000 units could generate $13 billion in revenue for Google.

The numbers are frightening. But how does Jen-Hsun Huang see it?

He said that we have been competing with application-specific integrated circuits (ASICs) for many years. Google's TPU is one of those things. Yes, Google does quite well, but what's the problem? ASICs are specialists and indeed strong in specific scenarios, but once the track changes, they are at a loss. NVIDIA takes the generalist route.

The CUDA architecture is like an open operating system. It can not only run large model training but also covers fields such as industrial design, new drug development, and engineering simulation. In these markets, specialized chips cannot get a foothold. This is the power of the platform—just like the App Store, where developers gather, the ecosystem naturally develops.

Google's TPU is more like an internal workshop for its own empire, serving its cloud services and AI applications. And what about Nvidia? Its shadow is everywhere. Cloud vendors, server suppliers, enterprise data centers, edge computing nodes, all are using its solutions. This is an open arsenal, where anyone can come to procure weapons.

Google's strategy essentially aims for a full-chain vertical integration from chips to applications, becoming self-sufficient. But the problem arises — can this closed model truly compete with an open platform that has a massive network effect?

Looking back at history, what are the odds of a closed system against an open ecology? Nvidia is selling not just chips, but an entry ticket to the entire ecosystem.