Date: 21-Dec-2025

Hey {{first_name | AI enthusiast}},

This week: We look at the tie-up between Google and Meta to loosen the monopoly that NVIDIA has managed to create with its GPUs and its software ecosystem.

Anthropic further lends credence to this strategy with its investment into Google’s TPUs. It already uses Amazon’s Tranium chips.

Finally, I can’t let go of the video of the dancing robots that have invaded my X feed. Plus a video on the most advanced city in the world- Shenzen.

It is a mixed bag of goodies- I hope you enjoy it!

— Renjit Philip

PS: If you want to check out how to implement AI agents in your business and get more revenue with the same number of employees, speak to me:

Google is making a quiet but important shift in its AI strategy. The focus is moving from faster chips to easier software adoption.

Google is working on TorchTPU to improve how its AI chips run PyTorch. PyTorch is the most widely used framework among AI developers.

TPUs already matter to Google Cloud revenue growth. Adoption stalled because developers faced software friction. Most engineers do not write code that chips execute directly. They rely on PyTorch libraries to abstract hardware complexity.

Nvidia understood this early. Its CUDA software became deeply embedded inside PyTorch workflows. That tight integration helped Nvidia defend its dominance. Performance gains came bundled with developer convenience.

Google took a different path internally. TPUs were optimized around Jax and XLA. That worked for Google teams. It did not work for most customers.

Enter TorchTPU.

The goal is simple: let PyTorch users adopt TPUs without rewrites. Google is committing more resources than in past attempts. Some parts of the software may be open sourced. Enterprise customers have been clear. Switching frameworks adds time, cost, and risk.

Google is also working closely with Meta, PyTorch’s steward. Meta wants lower inference costs and more supplier choice.

TPUs are now sold directly into customer data centers. AI infrastructure leadership reports directly to the CEO. The AI chip war is evolving. Software ecosystems now shape hardware winners.

Microsoft’s dominance in the nineties and the 2000s was partly because of its great developer tools like Visual Studio /VS Code that helped developers to build software and make money. Apple does the same with its App Store (although it is more directly extractive from the app revenues).

Whoever fits into developer habits fastest gains the edge.

Anthropic is scaling compute as a strategic priority, not an operational detail.
Model capability now depends directly on access to large, reliable infrastructure.

The planned expansion includes up to one million TPUs on Google Cloud.
This signals confidence in TPU price performance and energy efficiency. The investment runs into tens of billions of dollars. Frontier AI development now requires capital at infrastructure scale.

Capacity coming online in 2026 exceeds one gigawatt. AI labs are starting to resemble large scale energy consumers.

Anthropic’s customer base is expanding rapidly. Large enterprise customers grew nearly seven times in one year and we covered this earlier in an edition.

Anthropic is keep to avoid single vendor dependency. It actively balances TPUs, Trainium, and Nvidia GPUs. This diversified stack strengthens negotiating leverage.
It also protects against supply and pricing shocks.

So what? Google gains a flagship workload for its TPU roadmap. Anthropic gains long-term capacity without exclusive lock-in.

Amazon remains Anthropic’s primary training partner. They have a joint “Project Rainier” that reinforces long horizon compute planning. Remember that Amazon had invested billions into Anthropic.

The frontier AI race is shifting. Compute access, energy efficiency, and supplier diversity now define advantage.

This just caught my eye. Now I can pretend that AI is involved in all of these innovations, and that is why I included this video.

That is only part of the reason as you will see! Shenzen is indeed the city of the future.