12 hours ago
946
Alphabet (NASDAQ: GOOG) shares climbed roughly 3% after news emerged that Google is developing a new server chip, internally called Frozen v2, which is designed to run its Gemini AI models more efficiently.
Class C shares closed at $346.12 on Friday, which put the intraday move at roughly $11 per share.
According to reports, the new server chip is expected to be released as soon as 2028.
What is Frozen v2 supposed to do?
Frozen v2 is designed to permanently embed parts of Gemini’s architecture directly into the dedicated chip, reducing the number of calculations and amount of data movement required to answer queries.
According to Google engineers, it could save between six and ten times more tokens per unit of power than the company’s newest AI chips.
In simpler terms, rather than relying on a general-purpose chip to figure out how to run Gemini at runtime, the chip would have Gemini’s blueprint baked in from the start, cutting down on power consumption and generating faster responses.
Additionally, the chip would be separate from Google’s tensor processing units (or TPUs) and would reduce the amount of data the chip has to move around, making it faster at responding to queries.
However, a major issue with the design is that the chip would only work with future Gemini models if Google sticks with the same underlying architecture.
At the moment, Google sees Frozen v2 as a trial run and does not plan to produce it at the same scale as its TPUs. The more of Gemini that gets locked into hardware, the more efficient the chip becomes, but the harder it becomes to update when the model evolves. AI models evolve fast.
Where does Frozen v2 fit in Google’s broader hardware strategy?
Google launched its first-generation Tensor Processing Unit (TPU) back in 2016, which made it one of the first tech giants to build custom silicon rather than rely on Nvidia GPUs. Since then, Google has released multiple TPUs, applying them across its cloud services and within its internal systems.
With Frozen v2, Google would be adding a new computing option alongside TPUs and Nvidia GPUs, which work best for flexible tasks that require constant model updates. Rather than changing its current hardware strategy, Google has simply added a new option for situations where efficiency is crucial, and the model design is much more stable.
Why investors care
Investors consider Google’s hardware strategy as a long-term competitive advantage rather than a way to not rely on third-party hardware. In November 2025, the company released its 7th generation Ironwood TPU, which served as an alternative to Nvidia’s AI GPU.
Around the same time, Anthropic announced plans to include Google cloud technologies, including one million TPUs to power future Claude models, which greatly boosted confidence in Google’s in-house silicon ecosystem.
The momentum also placed Google as a trusted alternative to Nvidia in AI computing, causing Alphabet to briefly overtake Nvidia as the world’s most valuable company during after-hours trading in May 2026 before dropping back later in the week.
While the investment growth sounds promising, there is still one major roadblock. The chip is not expected to be released till 2028, meaning it simply cannot solve the power constraints and data center capacity issues that Google faces today.
Additionally, the chip’s design may be outpaced by the rapid evolution of AI models, which might cause it to lose its competitive edge by the time it is released.
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.
English (US)