Alphabet, Google’s parent company, is developing a new server chip designed to run its Gemini AI models more efficiently, as major AI companies race to build custom hardware that reduces costs, improves performance, and lowers dependence on external chip suppliers.<\/p>\n\n\n\n
The chip, internally codenamed Frozen v2<\/strong>, is expected to launch in 2028 and could deliver a significant leap in efficiency for Google’s AI infrastructure.<\/p>\n\n\n\n According to reports, Frozen v2 may be between six and ten times more efficient than Google’s current AI chips, based on the number of AI tokens generated per unit of power consumed.<\/p>\n\n\n\n Improving this metric has become increasingly important as AI companies seek to reduce the enormous energy and infrastructure costs associated with running large language models.<\/p>\n\n\n\n If the reported performance gains are achieved, Frozen v2 could enable Google to serve Gemini models faster while lowering operating costs across its cloud and AI services.<\/p>\n\n\n\n Google did not directly confirm the existence of Frozen v2 but also did not deny the reports.<\/p>\n\n\n\n In a statement, the company said its engineering teams continuously research and test new technologies to improve performance and efficiency for customers.<\/p>\n\n\n\n Google added that not every experimental project reaches production, but emphasized that its strategy of co-designing hardware and software allows its AI systems to be highly optimized for real-world workloads.<\/p>\n\n\n\n Leading AI developers are increasingly investing in proprietary chips as demand for AI computing continues to grow worldwide.<\/p>\n\n\n\n Custom silicon enables companies to optimize hardware specifically for their own AI models, improving efficiency while reducing operational costs and increasing control over critical infrastructure.<\/p>\n\n\n\n The approach has become particularly important as demand for AI computing capacity continues to outpace supply.<\/p>\n\n\n\n One of the primary motivations behind custom AI chips is reducing reliance on Nvidia, which has dominated the AI accelerator market for years.<\/p>\n\n\n\n As AI development accelerates, companies are seeking greater independence from third-party hardware suppliers while improving the economics of deploying increasingly powerful AI models.<\/p>\n\n\n\n A successful launch of Frozen v2 could allow Google to run Gemini more efficiently across Google Cloud and its internal infrastructure while reducing its dependence on external GPUs.<\/p>\n\n\n\n Google is not alone in developing custom AI hardware.<\/p>\n\n\n\n In June, OpenAI unveiled its first custom inference chip, Jalape\u00f1o, while recent reports indicated that Anthropic is discussing a chip development partnership with Samsung.<\/p>\n\n\n\n These initiatives highlight a broader industry trend in which competition extends beyond AI models to include the hardware powering them.<\/p>\n\n\n\n As AI infrastructure expands, energy efficiency has become one of the industry’s most important competitive metrics.<\/p>\n\n\n\n Generating more AI tokens while consuming less electricity can significantly reduce operating expenses for cloud providers and AI companies.<\/p>\n\n\n\n More efficient hardware also enables providers to serve more users using the same data center infrastructure, improving scalability and profitability.<\/p>\n\n\n\n Frozen v2 is part of Alphabet’s broader investment in artificial intelligence infrastructure.<\/p>\n\n\n\n The company plans to spend between $180 billion and $190 billion<\/strong> on AI-related infrastructure, including data centers and computing hardware, as it continues expanding Gemini and Google Cloud.<\/p>\n\n\n\n With such significant capital investments, Alphabet faces growing pressure from investors to demonstrate that its AI spending will translate into stronger financial returns and long-term competitive advantages.<\/p>\n\n\n\n The development of Frozen v2 reflects a major shift in the AI industry.<\/p>\n\n\n\n Competition is no longer centered solely on building more capable language models. Increasingly, success depends on who can design the most efficient chips, operate the most advanced data centers, and deliver AI services at the lowest possible cost.<\/p>\n\n\n\n If Frozen v2 meets expectations, it could strengthen Google’s position in the AI race by improving Gemini’s performance while reducing infrastructure costs and dependence on third-party chipmakers.<\/p>\n","protected":false},"excerpt":{"rendered":" Alphabet is developing a new AI server chip, codenamed Frozen v2, to power Gemini models more efficiently. Expected to launch in 2028, the chip could deliver up to 10 times greater efficiency while helping Google reduce infrastructure costs and lessen its reliance on Nvidia.<\/p>\n","protected":false},"author":37,"featured_media":71318,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[8566],"tags":[],"class_list":["post-71320","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai"],"acf":[],"_links":{"self":[{"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/posts\/71320","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/users\/37"}],"replies":[{"embeddable":true,"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/comments?post=71320"}],"version-history":[{"count":1,"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/posts\/71320\/revisions"}],"predecessor-version":[{"id":71321,"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/posts\/71320\/revisions\/71321"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/media\/71318"}],"wp:attachment":[{"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/media?parent=71320"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/categories?post=71320"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/entarabi.com\/en\/wp-json\/wp\/v2\/tags?post=71320"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}Frozen v2 Could Be Up to 10 Times More Efficient<\/h2>\n\n\n\n
Google Declines to Confirm the Project<\/h2>\n\n\n\n
Why AI Companies Are Building Custom Chips<\/h2>\n\n\n\n
Reducing Dependence on Nvidia<\/h2>\n\n\n\n
OpenAI and Anthropic Join the Chip Race<\/h2>\n\n\n\n
Energy Efficiency Becomes a Competitive Advantage<\/h2>\n\n\n\n
Alphabet Bets Big on AI Infrastructure<\/h2>\n\n\n\n
The Next Phase of AI Competition<\/h2>\n\n\n\n