Google Is Building a New AI Chip That Could Cut Power Costs Tenfold
A chip codenamed 'Frozen v2' is reportedly 6 to 10 times more efficient than Google's current hardware. If it delivers, it could meaningfully reduce what it costs to run Gemini, Google's family of AI models.
Key points
- Google is developing a new server chip, internally codenamed "Frozen v2," with a planned release in 2028.
- The chip could be 6 to 10 times more efficient than Google's existing AI chips, measured by how many text responses it can produce per unit of electricity.
- Google's stock rose roughly 3% on Monday morning after news of the chip became public.
- OpenAI announced its first custom chip, called Jalapeño, in June 2025; Anthropic is reportedly in talks with Samsung over a chipmaking partnership.
Google is quietly designing a new server chip built to run its Gemini AI models more cheaply. Codenamed "Frozen v2" inside the company and first reported by The Information, citing anonymous sources, it's targeted for release in 2028.
The headline figure is a big one. Frozen v2 could produce 6 to 10 times as many AI-generated responses for every unit of electricity compared with Google's current chips. Same electric bill, far more AI work done.
Google didn't confirm the report when TechCrunch asked directly, and it didn't deny it either. "Our teams are constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers," a spokesperson told TechCrunch, adding that not every internal project reaches production.
Running AI is expensive, and those costs flow downstream. Companies that cut their chip bills have more room to keep AI tools free or affordable rather than passing costs on to subscribers or advertisers. We've reported on exactly this pressure since July: as our 17 July story "The Electricity Problem at the Heart of Modern AI" explained, the company that squeezes the most useful work from each watt tends to win.
The wider industry is pushing hard in the same direction. Nvidia currently supplies most of the specialised processors that AI companies need. That dependence worries the big labs: if Nvidia raises prices or supply tightens, everyone from Google to startups feels it. Building their own chips is the most direct way to reduce that exposure.
OpenAI took that step in June 2025, announcing Jalapeño, its first in-house inference chip, a processor designed to generate AI responses rather than train new models. Anthropic is reportedly in early discussions with Samsung about a similar arrangement.
For Google, the financial pressure is acute. Earlier this year the company said it plans to spend between $180 billion and $190 billion on AI infrastructure through 2025. News of Frozen v2 appears to have offered investors some reassurance: Google's stock climbed around 3% on Monday morning following The Information's report, just ahead of the company's quarterly earnings call.
What does this mean for people who use Google products?
Faster, cheaper chip technology generally means AI features reach more people at lower cost. If Frozen v2 performs as described, Google could run Gemini more efficiently, translating into quicker responses in Search or Android without a price rise for users. The 2028 target is still years away, and Google's own statement acknowledged not every internal project makes it to production. That caveat deserves as much attention as the efficiency numbers.
