Sources: Google is developing a specialized server chip, informally dubbed “Frozen v2”, that integrates its Gemini AI model blueprint into the silicon, for 2028

Google is working on a new server chip that would directly integrate the blueprint of its Gemini AI model …

The Information 2026-07-20

Context & Ripple Effects

Google has been pairing its Gemini roadmap with in-house infrastructure: Trillium was introduced as the chip powering Gemini 2.0, while Gemini 2.0 was positioned for testing in Search and AI Overviews. Frozen v2 extends that progression from running a model on custom hardware to designing silicon around the model itself.

The reported 2028 timeline matters because it makes hardware architecture a longer-term dependency of Gemini’s product and model roadmap, rather than a separate data-center procurement decision.

First-order effects

Google’s chip and Gemini teams would need to co-design Frozen v2 around the model blueprint, making the future server-chip roadmap more tailored to Gemini workloads.
The move creates a successor path beyond the Trillium generation, but it does not represent a deployed product: the reported chip remains a 2028 development effort.

Second-order effects

A model-specific chip can make changes to Gemini’s architecture more consequential for infrastructure planning, since model advances and hardware design choices must be validated together over a multiyear cycle.
If successful, tighter software-hardware integration could improve Google’s control over the cost and performance profile of Gemini deployment across its services, relative to relying on more general-purpose compute.

Third-order effects

The effort points toward AI infrastructure competition shifting from acquiring accelerators to owning the co-design loop among models, chips, and data-center software.
That strategy also raises switching costs: as models become optimized for proprietary silicon, AI platforms may become more differentiated by their internal infrastructure stacks, though the payoff depends on Frozen v2 reaching production and delivering the intended gains.

The trend: Frontier AI providers are increasingly treating custom silicon and model architecture as a single strategic system rather than separate layers of the stack.

Discussion

@zephyr_z9 @zephyr_z9 on x
So, this chip makes a lot of sense for real-time voice models and stuff that requires extremely low latency
@jukan05 Jukan on x
So Google is essentially developing Taalas-like chips that bake the model weights directly into the silicon? [image]
@cxcarroll @cxcarroll on x
This is very well-suited for latency-sensitive applications like voice. While it appears that frontier models may become somewhat commoditized, producing silicon specific to models is going to be reserved for just a couple of the big players.
@scottw_grizzle Scott Willis on x
Interesting solution to the compute problem.
@kawzinvests @kawzinvests on x
Taalas has a public demo called Chat Jimmy running their hardwired Llama 3.1 8B at around 17,000 tokens per second. It's honestly insane. You hit enter and the full response is just there. About 0.03 seconds. It streams faster than you can physically read.
@amir Amir Efrati on x
🔥chip news: Google planning to bake LLM architecture into new AI server chips (non-TPUs) to run Gemini 6-10x more efficiently [image]
@kimmonismus @kimmonismus on x
Google may be preparing to freeze parts of Gemini's architecture directly into silicon. Informally called “Frozen v2,” the chip reportedly targets 6-10× more tokens per watt than Google's newest TPUs. Deployment is planned for as early as 2028. The motivation is immediate: [image…
@stocksavvyshay Shay Boloor on x
$GOOGL is developing a new AI chip called “Frozen v2” that could run Gemini models nearly 10x more efficiently than its latest TPUs. The chip is targeted for 2028 and would hardwire parts of Gemini to improve speed and efficiency while easing Google's compute constraints. [image]
@andrewcurran_ Andrew Curran on x
Google is developing a new chip named ‘Frozen v2’ specifically designed to run the Gemini family of models more efficiently. The new hardware is expected to arrive in 2028. This was originally reported by The Information. Gemini probably helped design this. [image]

Chronicles