Google is building its own AI chip called Frozen v2 to make Gemini models significantly more power-efficient. The Information reported the news, citing anonymous sources; TechCrunch reports on this story. The chip could be between 6 and 10 times more efficient than Google's current AI chips—measured by tokens generated per unit of power. Expected launch: 2028.
Google neither directly confirmed nor denied the report to TechCrunch. Instead, the company stated:
"Our teams are constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers. While not every project moves into production, this rigorous exploration is central to our full stack approach. By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimized for real-world workloads."
The essentials
- Frozen v2 is Google's internal codename for a new server chip optimized for Gemini models
- Efficiency gain: 6–10x better tokens-per-watt compared to current Google chips
- Availability: 2028 planned
- Alphabet's stock rose roughly 3% after the announcement
The trend: Everyone is building their own chip
Google is not alone. Pressure to reduce dependence on Nvidia is mounting across the AI industry. OpenAI announced its first custom chip in June—an inference processor called Jalapeño. Anthropic is reportedly negotiating a chipmaking partnership with Samsung. The logic is straightforward: running models on proprietary hardware cuts costs, reduces latency, and decreases reliance on Nvidia's market dominance.
For Google, the stakes are higher: the company plans to invest between 180 and 190 billion dollars in its AI strategy. Investors want proof that this money will pay off. News of the more efficient Frozen v2 chip arrived at the right moment—and calmed the markets.
Efficiency as the new selling point
The industry has shifted. While two years ago every new model announcement sparked euphoria, today a different question dominates: How expensive is it to run? Efficiency has become the decisive selling argument—for customers, investors, and markets.
Frozen v2 fits this logic perfectly. A chip that runs Gemini with less power and fewer compute resources lowers Google's operating costs and makes Gemini more competitive as a cloud service. It's not flashy, but it is profitable.
What this means for European companies
The vertical integration of AI—tight coupling of model, software, and hardware—is becoming the new standard. This has implications for European firms investing in AI: those running models in the cloud will depend on providers optimizing their own hardware. This can reduce costs but may also strengthen lock-in effects. At the same time, a new market for specialized chip development is emerging—an opportunity for European hardware companies, if they move fast enough.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
