Google is reportedly developing a new server chip designed to make its Gemini AI models run more efficiently. The project, known internally as Frozen v2, is expected to arrive around 2028 and could deliver a major leap in performance per watt.
According to the report, the chip may be six to 10 times more efficient than Google's current AI hardware, based on how many tokens it can generate for each unit of power. Google did not directly confirm the details, but said its teams are continuously testing new ideas to improve speed, efficiency, and system integration across its hardware and software stack.
The move reflects a broader shift in the AI industry, where major companies are increasingly designing their own chips to support in-house models and reduce dependence on external suppliers. Efficiency has become a central priority as firms look for smarter ways to scale AI infrastructure.
Google's strategy also comes as other AI leaders pursue similar paths. OpenAI recently introduced its own custom chip, while Anthropic has been linked to chipmaking discussions with Samsung. These efforts show how custom silicon is becoming a key part of the next phase of AI development.
For Alphabet, the timing matters. The company has committed to large-scale AI investment, and stronger chip performance could help support that long-term vision. Following the report, Alphabet shares moved higher, signaling investor interest in the potential gains from more efficient AI hardware.
As custom chips become more advanced, they may shape a future where AI systems are faster, leaner, and more widely deployable across everyday digital experiences.
