Google is working on a new AI chip designed to make Gemini more efficient
TechCrunch Lucas Ropek ● Covered by 19 sources
Google's reportedly building a new AI chip called Frozen v2, due in 2028, to run Gemini way more efficiently. It could be 6-10x more power-efficient than current chips—big if it holds up.
Based on reporting by TechCrunch, Lucas Ropek — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Alphabet is apparently cooking up a new server chip, internally called Frozen v2, aimed squarely at making its Gemini models cheaper to run. The Information broke the news citing anonymous sources, and the headline number is striking: six to ten times the efficiency of Google's current AI chips, measured in tokens generated per watt. Google hasn't confirmed the report by name, but it hasn't shot it down either. The company's statement to TechCrunch leaned on familiar language about "co-designing hardware and software" and "rigorous exploration," which is corporate-speak for yes, we're probably doing this, but we're not ready to say so officially.
The timing matters more than the chip itself right now. Google said earlier this year it plans to spend somewhere between $180 billion and $190 billion on its AI buildout, a number big enough to make even patient investors nervous. Efficiency gains are the industry's current favorite answer to the obvious follow-up question: how does any of this spending actually pay off? A chip that squeezes out several times more tokens per unit of power isn't just a technical flex, it's a direct rebuttal to anyone wondering whether Alphabet's AI bet is turning into a money pit.
This also fits a broader pattern that's become impossible to ignore. OpenAI unveiled its own inference chip, Jalapeño, back in June. Anthropic has reportedly been in talks with Samsung about a chipmaking partnership. Everyone with serious AI ambitions is trying to loosen Nvidia's grip on the supply chain, both because custom silicon can be tuned for a company's specific workloads and because relying entirely on one supplier in a chip shortage is a genuinely bad position to be in. Frozen v2, whenever it actually ships in 2028, would be Google's latest move in that direction, following its long-running TPU line.
Investors, at least for a day, seemed to like what they heard. Alphabet's stock ticked up roughly 3% Monday morning after the report landed, just ahead of the company's earnings call later this week. Whether that bump holds depends on whether Frozen v2 turns into something real rather than another internal project that quietly dies before production, which Google itself admitted happens with "not every project."
My take — AI-written commentary, not fact-checked reporting
A 2028 release date for a chip getting hyped in late 2025 is doing a lot of narrative work for a stock price that needed good news before earnings. I'd bet Frozen v2 is real, because Google's TPU track record is legitimately strong, but three years is an eternity in this industry and Nvidia isn't exactly standing still either.
Read more about this at: TechCrunch