Google's Secret 'Frozen v2' Chip Could Slash Gemini Power Costs Tenfold by 2028

Alphabet, the parent company of Google, is reportedly deep into development of a next-generation AI server chip that could dramatically improve how efficiently its Gemini family of models runs at scale. The chip, known internally as 'Frozen v2,' is expected to reach deployment sometime in 2028, according to a report from The Information citing anonymous sources familiar with the project. If those figures hold, the chip could generate between six and ten times more tokens per unit of power than the AI chips Google currently operates — a potentially transformational leap in computational efficiency.

Google declined to officially confirm or deny the existence of the Frozen v2 project when approached by TechCrunch, but its statement was notably careful rather than dismissive. The company said its teams are 'constantly researching and experimenting with new innovations to deliver maximum performance and efficiency,' adding that its hardware-software co-design philosophy is central to how it approaches AI infrastructure. That kind of measured non-denial often signals that a project is real but not yet ready for public announcement.

The broader context here matters enormously. AI companies across the industry have been racing to design proprietary silicon as a way to cut costs, reduce bottlenecks, and loosen their dependence on Nvidia, whose dominance in AI chip supply has left major players in a vulnerable position. OpenAI announced its own custom inference chip, dubbed Jalapeño, in June, while Anthropic has reportedly been in talks with Samsung about a new chipmaking partnership. Google joining this arms race with a chip this ambitious puts it firmly in competition at the hardware layer, not just the model layer.

The timing of the Frozen v2 revelation is also notable given Alphabet's financial context. The company has committed to spending between $180 billion and $190 billion as part of its broader AI buildout — a figure that has made investors nervous about whether returns will justify the scale of expenditure. News of the more efficient chip appears to have offered some reassurance, with Alphabet's stock rising roughly 3% on Monday morning following the report, ahead of the company's earnings release later in the week.

For users and enterprise customers, a chip this efficient could translate into faster, cheaper, and more widely available access to Gemini-powered services. AI inference costs remain one of the key friction points limiting how aggressively companies can deploy large language models at scale. If Frozen v2 delivers even half of its projected efficiency gains, it could meaningfully change the economics of AI deployment — not just for Google, but as a benchmark that pressures the entire industry to follow suit.