Google plans to deploy Frozen V2 chips by 2028, escalating the AI hardware arms race

Google is building a specialized AI chip called Frozen V2 that could be 6 to 10 times more efficient than its current Tensor Processing Units. The chip, designed specifically to supercharge its Gemini AI models, is targeted for deployment as early as 2028.

The report, originating from The Information on July 20, 2026, outlines a project that goes beyond the typical chip upgrade cycle. Rather than building general-purpose silicon, Google is hardwiring optimizations directly into the chip’s architecture that are tailored for Gemini workloads.

Google’s decade-long custom silicon strategy

This isn’t Google’s first rodeo with custom chips. The company launched its first-generation TPU back in 2016, making it one of the earliest big tech firms to go the bespoke silicon route rather than relying entirely on NVIDIA GPUs.

Since then, Google has iterated through multiple TPU generations, deploying them across its cloud services and internal systems. Search, YouTube, and virtually every Google product that touches machine learning runs on some form of custom hardware.

The Frozen project represents a philosophical evolution in that approach. Previous TPU generations were designed to handle a broad range of AI workloads. Frozen V2, by contrast, appears to embed model-specific optimizations directly into the hardware, with Gemini’s neural network architecture baked into its circuitry, which is why the efficiency gains are so dramatic compared to general-purpose alternatives.

This mirrors a broader industry trend. Amazon has its Trainium chips, Meta has been developing custom inference accelerators, and Microsoft has Project Maia.

What this means for the compute economy and crypto markets

Decentralized compute networks like Render, Akash, and io.net have built their entire value proposition on the idea that GPU and TPU resources are scarce and expensive. A dramatic efficiency leap from Google could reshape that calculus.

Investors in AI-adjacent crypto tokens should watch two things closely. First, whether Google makes Frozen V2 available through Google Cloud or keeps it exclusively for internal use. Second, how NVIDIA responds, as any counter-move could reshape the GPU economics that decentralized compute networks depend on.

Google has not officially confirmed the Frozen V2 project. The deployment target remains as early as 2028.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.