Most investors can’t buy it. I own it in my Roth IRA.

To understand why this matters, you have to go back to Google’s TPU.

Years ago, Google realized traditional processors weren’t optimized for the massive matrix calculations required by AI. Instead of relying entirely on third-party chips, they built their own custom AI accelerator:

The Tensor Processing Unit (TPU).

Google initially used TPUs internally to power products like Search, YouTube, and Gemini. Then they started renting access through Google Cloud.

Today, some of the world’s leading AI companies, including Anthropic, use Google’s TPU infrastructure to train and run frontier models.

This new venture with Blackstone expands access and use of that infrastructure.

Think of it as a CoreWeave-style platform built around Google’s proprietary technology stack instead of rented NVIDIA hardware.

CoreWeave built a massive business by acquiring NVIDIA GPUs and renting them to AI companies.

The challenge is that everyone is competing for the same underlying hardware.

The new Google + Blackstone platform is attempting to build an AI infrastructure business around technology that only Google can provide.

If NVIDIA is the dominant AI compute ecosystem today, TPUs are one of the few credible alternatives operating at global scale.

For certain workloads, there are compelling reasons customers prefer them:

• Designed specifically for large-scale AI computations
• Potential advantages in performance and power efficiency
• Lower infrastructure costs for certain training and inference workloads

As AI scales, power availability and compute costs are becoming just as important as raw performance.

Public markets are already assigning enormous valuations to AI infrastructure companies built on rented hardware.

This business is being built around Google’s proprietary TPU ecosystem, backed by Blackstone’s capital and infrastructure expertise.

Most investors can’t access that opportunity. Our clients can within their Roth IRAs.