Invest in Google for its Unrivaled Vertical Integration in AI

Google is the only one with the model, software, hardware and everything around it. Sure Microsoft has a chip, but no one talks about Maia, and Amazon has a great set of chips with Graviton and Trainium, but they don't have their own models. That leaves you with Google, who has integrated to create a best in class product and will continue to do so.

Google Next was last week and there was a crazy amount good content showing just how well positioned they are.

These are just a few of my top takeaways. I've put all the transcripts together at https://arrugialabs.com/podcast-research

You need to login, but there is no paywall for anything. I also had AI put together some cheat sheets if you don't want to read through the transcripts.

1. TPU 8t for massive scale pre-training and TPU 8i for low latency inference means Google has recognized that AI workloads have diverged and moving away from one size fits all is going to improve performance. 8i networking topology is reducing latency by up to 50% for real-time reasoning tasks. 2. Google is actively transitioning into the "agentic era" by developing autonomous AI agents that can solve complex, multi-step problems 3. To support these massive computational demands, Google plans to invest between $175 billion and $185 billion in capital expenditures this year 4. An AI coding agent named AlphaEvolve successfully proposed a highly efficient circuit design that Google integrated directly into the silicon of its new TPUs. This seems huge to me as they are using AI to improve its own chips. 5. General-purpose CPUs are predicted to make a major comeback because they are required to orchestrate the massive sandboxes and virtual machines that AI agents use. 6. Advancements in AI computational power are accelerating scientific discovery to the point where 10 years of research could soon be completed in a single year