Alphabet, Google's parent company, is designing a new server chip, internally dubbed "Frozen v2," intended to help its in-house Gemini AI models run more efficiently. According to a report in The Information, the chip is slated for release in 2028 and could be between six and 10 times more efficient than Google's existing AI chips, measured by the number of tokens generated per unit of power. Google did not directly confirm the report but also did not deny it, telling TechCrunch that its teams constantly research and experiment with new hardware and software innovations.<\/p>\n
Frozen v2 represents a shift from running Gemini as software on general-purpose AI hardware to hardwiring parts of the model directly into the silicon. The goal is a tighter integration between model and chip, in which the accelerator is designed around a specific model family rather than designed to run any model. The Information, citing anonymous sources, reported that engineers expect the chip to deliver six to 10 times the efficiency of Google's current Tensor Processing Units (TPUs) when serving AI queries, with that gain measured in tokens generated per unit of power.<\/p>\n
Two pressures are converging on the project. First, AI companies are trying to ease internal compute crunches as demand for inference continues to rise. Second, the industry is working to reduce its dependence on Nvidia, which has historically dominated the AI chip market and left major AI makers reliant on its hardware. Building custom accelerators addresses both problems at once, cutting the power and cost per response while reducing exposure to external supply.<\/p>\n
Google is not alone in this push. In June, OpenAI announced its first custom chip, an inference processor called Jalape\u00f1o. Earlier this month, it was reported that Anthropic was discussing a new chipmaking partnership with Samsung. The pattern is consistent: each major AI lab is moving toward tighter hardware and software integration.<\/p>\n
Investors have pressed Alphabet on the scale of its planned AI investments. Earlier this year, Google said it plans to spend between $180 billion and $190 billion on the infrastructure needed to support its AI strategy. With that level of capital committed, the company faces pressure to demonstrate that the spending will pay off in measurable efficiency gains. News of the Frozen v2 project appeared to ease some of those concerns, with Alphabet's stock climbing roughly 3% on Monday morning following publication of The Information's report, ahead of the company's earnings report later in the week.<\/p>\n
Google did not confirm the existence of the Frozen v2 chip by name, but it did not dispute the report either. A spokesperson described the company's approach to hardware and software in general terms: "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."<\/p>\n
The Information reported that the chip is not expected to arrive until 2028, making it a longer-horizon effort rather than an immediate upgrade to Google's inference fleet. In the near term, Google will continue to rely on its existing TPUs and on Nvidia hardware for much of its AI compute. Frozen v2, if it reaches production, would arrive as part of a broader shift in the industry toward specialized silicon designed for specific model families.<\/p>\n
Frozen v2 is the reported internal name for a new AI chip Alphabet is designing to make its Gemini models more efficient. According to The Information, engineers expect it to be six to 10 times more efficient than Google's current TPUs when serving AI queries, measured by tokens generated per unit of power.<\/p>\n
The Information reports that Frozen v2 is slated to be released sometime in 2028, making it a longer-term project rather than a near-term upgrade to Google's AI infrastructure.<\/p>\n
The push for custom AI chips is part of a broader industry effort to reduce dependence on Nvidia, which has historically dominated the AI chip market. Other labs are following a similar path: OpenAI announced its first custom inference chip, called Jalape\u00f1o, in June, and Anthropic has reportedly been discussing a new chipmaking partnership with Samsung.<\/p>\n