Former $Alphabet-A (GOOGL.US)$ employee shares his views on ...

Former $Alphabet-A (GOOGL.US)$ employee shares his views on the infrastructure build-out for AI application and what it means for Big Tech:
- When the expert started working at $Alphabet-A (GOOGL.US)$ , they were still very dependent on $NVIDIA (NVDA.US)$ GPUs. Since then, they have started rolling out their own TPUs, which they now also offer to GCP customers to rent out. In his view, TPUs can be much more efficient than GPUs for AI in some use cases.
- In his opinion, GCP clients and $Alphabet-A (GOOGL.US)$ 's internal products run largely on the same infrastructure. One of $Alphabet-A (GOOGL.US)$ 's goals with Cloud is to achieve a higher utilization of its own infrastructure.
- One of the big constraints with data centers is energy. The expert talks about the actual power grid and power stations available in the area where the data center is built. This becomes even more important as many AI computations can be extremely power-hungry.
- He thinks Big Tech like $Alphabet-A (GOOGL.US)$ , $Meta Platforms (META.US)$ , $Microsoft (MSFT.US)$ , and $Amazon (AMZN.US)$ in 5 years will be less reliant on $NVDA than they are today, as these companies want to control their destiny.
- He believes that TPUs and custom chips will be used for many of the use cases, but for some custom use cases, cloud players will still prefer $NVIDIA (NVDA.US)$ 's GPUs as they are more customizable.