Google Reportedly Developing "Frozen v2" AI Chip, Expected to Launch in 2028 with Up to 10× Higher Energy Efficiency

Alphabet is reportedly developing a next-generation AI server chip internally, a project said to carry the codename “Frozen v2.” According to a report from The Information, which cited anonymous sources, the chip is intended to improve the efficiency of running Google’s Gemini models and could arrive as early as 2028.

The most notable claim tied to the project is its projected energy performance. If development progresses as expected, Frozen v2 may deliver 6 to 10 times more tokens per unit of energy than Google’s current AI chips. That kind of gain would be significant at a time when AI companies are under growing pressure to expand computing capacity while keeping power use and costs under control.

Google says experimentation does not guarantee a product launch

In response to the report, Google told TechCrunch that it continuously explores and tests new technologies aimed at improving performance and efficiency. At the same time, the company made clear that not every research effort ultimately becomes a mass-produced product.

Google also highlighted its broader approach to infrastructure design, emphasizing a full-stack development model in which hardware and software are built together. That strategy, the company said, allows it to create more tightly integrated systems that are optimized for real AI workloads rather than relying on isolated component upgrades.

Viewed in that context, Frozen v2 appears to be more than just a standalone chip initiative. It reflects Google’s larger push to strengthen the underlying infrastructure that supports its AI ambitions.

The AI race is moving beyond models alone

As generative AI adoption accelerates, demand for computing power continues to rise across the industry. That surge is pushing major technology companies to invest more heavily in custom silicon, both to improve runtime efficiency and to reduce dependence on external suppliers.

Nvidia still holds a dominant position in the AI chip market through its GPU ecosystem, but pressure is building as more companies look inward and begin designing their own hardware. The trend is no longer limited to model development; it increasingly includes chips, data center architecture, and end-to-end system design.

Recent moves across the sector reflect that shift:

• OpenAI introduced its first custom inference chip, Jalapeño, in June.

• Anthropic has also reportedly been in talks with Samsung regarding chip manufacturing cooperation.

These developments suggest that competition in AI is broadening. The contest is no longer just about who has the most capable model, but also about who can build the most efficient and scalable infrastructure to support those models.

Why chip efficiency matters more than ever

Alphabet has previously said it plans to invest between $180 billion and $190 billion to support its artificial intelligence strategy. As spending on AI infrastructure grows, efficiency is becoming a more important measure of whether those investments can produce strong long-term returns.

That is one reason reports about Frozen v2 drew market attention. After the story surfaced, Alphabet shares rose by about 3% in early trading on Monday, suggesting that investors are watching closely for signs that the company’s AI spending could translate into durable competitive advantages.

What Frozen v2 signals for Google

With AI models becoming larger and more resource-intensive, custom chips are increasingly central to how major tech companies plan to manage cost, performance, and control over their computing ecosystems. For Google, a project like Frozen v2 would fit directly into that broader strategic need.

If the chip reaches production on the reported timeline, it could become an important part of Google’s effort to improve the economics and performance of Gemini-related workloads. But for now, the project remains in the exploration and testing phase, based on Google’s own comments.

That means key questions remain unresolved, including whether Frozen v2 will ultimately enter mass production and when it might formally launch. Even so, the reported project underscores a clear industry reality: in the next phase of AI competition, hardware efficiency and infrastructure depth may matter almost as much as the models themselves.