
AI’s infrastructure side quest
Blackstone says it’s teaming up with Google to create a new U.S.-based company aimed at meeting surging demand for Tensor Processing Units, or TPUs. Translation: instead of just selling the shovels in the AI gold rush, they’re helping build the mine, the roads, and probably the snack bar too.
The new venture is supposed to offer efficient data center capacity, operations, networking, and Google Cloud’s TPU compute-as-a-service setup. That matters because AI training and inference don’t magically happen in the cloud ether. They need real estate, power, networking, and a lot of very expensive hardware.
Why investors should care
For Blackstone, this is another bet that digital infrastructure will keep getting more valuable as AI demand scales. If the thesis holds, BX gets exposure to a growing slice of the compute economy without having to pretend it’s suddenly a chip company.
For Google, the move gives TPU demand a more direct on-ramp and could help expand the ecosystem around its custom AI chips. In other words, it’s a way to make sure the infrastructure beneath AI doesn’t buckle under the hype.
Big picture
This is the kind of deal that tells you where the real bottlenecks are: not just models and software, but power, buildings, networking, and compute capacity. The AI boom may look like a software story, but a lot of the money is still hiding in the plumbing.
