Here's something Wall Street seems to be missing: while Intel and AMD have been ripping higher on the "CPUs for AI" narrative, Google has quietly been building its own chips for years—and they're not just for AI training.
Everyone knows Google makes its own TPUs (Tensor Processing Units) to avoid paying the Nvidia tax on AI workloads. But the company also designed a custom CPU called Axion, based on ARM architecture, specifically to stop paying the Intel tax for running its massive data centers. These chips power YouTube, search, cloud services—basically everything that makes Google money.
The logic is simple: if you're running millions of servers, you don't want to hand over a chunk of your margin to Intel every time you need to scale. By designing chips in-house, Google controls its costs, optimizes performance for its own workloads, and doesn't have to wait for Intel's roadmap to catch up with its needs.
Amazon figured this out years ago with its Graviton processors for AWS. Microsoft has been working on custom silicon for Azure. But Google has been quieter about it, even though it's arguably further along than either of them.
So why isn't the stock reflecting this? Part of it is that Google doesn't sell these chips to anyone else. Intel and AMD get credit because their revenue comes directly from chip sales. Google's advantage shows up as better cloud margins and lower infrastructure costs—benefits that don't make for flashy headlines but add up to billions over time.
The other issue is that Wall Street loves a simple story. "CPUs are important for AI" is easy to understand. "Google vertically integrated its entire compute stack to reduce dependency on external suppliers" doesn't fit in a tweet.
