Google reportedly developing ultra-efficient AI chip to power Gemini as race for custom silicon intensifies

Alphabet is reportedly preparing a major leap in its artificial intelligence infrastructure with a new in-house processor designed to make its Gemini models significantly more efficient, as technology companies increasingly view custom silicon as a strategic advantage in the global AI race.

According to a report by The Information, Google is developing a server chip internally known as Frozen v2, with plans to bring it into production in 2028. People familiar with the project told the publication that the processor is expected to outperform Google's existing AI chips by a substantial margin, potentially delivering between six and ten times greater efficiency when measured by the number of AI-generated tokens produced for each unit of power consumed.

While Google did not comment on the specific project, it acknowledged that exploring new hardware designs remains a central part of its long-term AI strategy.

"Our teams are constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers," a Google spokesperson told TechCrunch. "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."

More from Tech

The reported development highlights a growing trend among major AI companies to design specialised chips tailored to their own software rather than relying entirely on third-party hardware. Custom processors can improve performance, reduce operating costs and provide companies with greater control over the computing infrastructure required to train and deploy increasingly complex AI models.

The push also comes as demand for AI computing continues to outstrip supply. Nvidia remains the dominant supplier of AI accelerators, but its commanding market position has encouraged several leading technology companies to invest heavily in alternative hardware platforms to reduce dependence on a single vendor.

Alphabet is far from alone in that effort. OpenAI revealed its first custom inference chip, named Jalapeño, in June, marking an important step in developing its own AI hardware ecosystem. More recently, reports suggested that Anthropic has been exploring a potential chipmaking partnership with Samsung as it expands the infrastructure supporting its Claude models.

Efficiency has become an increasingly important benchmark across the AI industry. As companies spend billions of dollars expanding data centres and purchasing advanced hardware, investors are placing greater emphasis on whether those investments can generate sustainable financial returns rather than simply delivering more computing power.

That scrutiny has been particularly pronounced for Alphabet. Earlier this year, Google outlined plans to invest between $180 billion and $190 billion as part of its broader AI expansion strategy, underscoring the scale of capital being committed to infrastructure, research and cloud capabilities.

News of the reported Frozen v2 project appeared to reassure investors ahead of the company's earnings announcement later this week. Shares of Alphabet rose around 3% in Monday trading following publication of The Information's report, suggesting the prospect of a more efficient AI processor was viewed positively by the market.

Although the chip remains several years away from launch, the reported project underlines how competition in artificial intelligence is extending well beyond software. Increasingly, success in the AI era will depend not only on building capable models, but also on developing the specialised hardware needed to run them efficiently, economically and at scale.