Alphabet Inc. (GOOGL) shares surged Monday after a report revealed Google is developing a radical new server chip that permanently embeds key parts of its Gemini AI model directly into silicon, a design engineers estimate could deliver up to ten times the power efficiency of its current custom processors.<\/p>\n
The project, code-named \u201cFrozen v2,\u201d represents a sharp departure from the general-purpose philosophy behind Google\u2019s existing Tensor Processing Units (TPUs). By hardwiring Gemini\u2019s architecture into the chip itself, Google aims to slash the computational steps and data movement required to answer AI queries, a move designed to ease a severe internal computing shortage that has reportedly forced Google Cloud to turn away major enterprise deals.<\/p>\n
According to a report by The Information citing people familiar with the matter, Google is targeting deployment as early as 2028. The chip is not intended to replace the company\u2019s TPU lineup but rather to establish a new, highly specialized branch of custom silicon engineered exclusively for Gemini workloads.<\/p>\n
Google engineers estimate that Frozen v2 could process six to ten times more AI tokens per unit of power than the company\u2019s latest generation of TPUs. A Google Cloud spokesperson acknowledged the experimental nature of the work, stating that teams \u201cconstantly research and experiment with innovations to provide users and customers with the best performance and efficiency,\u201d while adding that \u201cnot every project moves into production.\u201d<\/p>\n
The news sent Alphabet\u2019s Class A shares up as much as 3.7% in intraday trading before settling at a 1.5% gain to close at $351.99. Class C shares (GOOG) climbed roughly 3.9% at their peak. The rally reflects investor optimism that proprietary infrastructure investments can strengthen Google\u2019s competitive position as the AI arms race intensifies.<\/p>\n
Hardware-Software Co-Design Takes Center Stage<\/p>\n
The concept behind Frozen v2 reportedly originated with Jeff Dean, Google\u2019s chief scientist. The name itself derives from the idea of \u201cfreezing\u201d parts of Gemini\u2019s decision-making logic into the chip\u2019s physical structure. Unlike a TPU or an Nvidia (NVDA) GPU, which must load and interpret model instructions at runtime, Frozen v2 would come with Gemini\u2019s blueprint baked in from the start.<\/p>\n
This approach eliminates layers of processing overhead. In a conventional setup, a general-purpose accelerator must constantly fetch model weights, execute calculations, and shuffle data between memory and processing cores. Frozen v2 collapses those steps by embedding portions of the model\u2019s architecture and weights permanently in hardware.<\/p>\n
The trade-off is flexibility. The chip will only work with future Gemini versions if Google maintains the same underlying model architecture. If the company overhauls Gemini\u2019s design, the hardwired elements could become obsolete. Sources indicated that Google currently views Frozen v2 as a trial run and has no plans to manufacture it at the same scale as its TPU fleet.<\/p>\n
\u201cBy co-designing hardware and software from the ground up, we ensure deep system integration optimized for real workloads,\u201d the Google spokesperson said, describing the company\u2019s full-stack strategy.<\/p>\n
A Response to Crippling Compute Shortages<\/p>\n
The Frozen v2 initiative is not merely a research curiosity. It is a direct response to what insiders describe as a debilitating shortage of AI computing capacity inside Google. The company does not have enough power to simultaneously train next-generation models and serve the rapidly expanding volume of Gemini queries.<\/p>\n
That crunch has triggered internal friction over resource allocation and, according to the report, forced Google Cloud to decline business from external customers. In a striking illustration of the pressure, Google agreed last month to pay Elon Musk\u2019s SpaceX nearly $1 billion per month to lease data center capacity, ensuring it could meet commitments to enterprise clients.<\/p>\n
Global data center capacity remains tight, and energy costs continue to rise. By squeezing dramatically more performance out of each watt, Frozen v2 could help Google stretch its limited compute infrastructure further without waiting for new facilities to come online.<\/p>\n
The project also fits into a broader industry trend. Virtually every major AI developer is now pursuing custom silicon to reduce reliance on Nvidia and optimize hardware for specific model architectures. Google launched its first TPU in 2016 and has since released multiple generations, most recently the seventh-generation Ironwood TPU in late 2025. The company has also begun supplying TPUs to external cloud customers, including a multi-billion-dollar agreement with Meta Platforms (META) and a deal to provide one million TPUs to Anthropic for future Claude models.<\/p>\n
Market Context and Upcoming Earnings<\/p>\n
The Frozen v2 disclosure lands just ahead of Alphabet\u2019s second-quarter earnings report, scheduled for July 22. Wall Street expects adjusted earnings per share of approximately $2.88 to $2.90 on revenue of roughly $117 billion, representing year-over-year growth of more than 20% on both metrics.<\/p>\n
Investors will be listening closely for commentary on AI infrastructure spending and the return on those investments. The company\u2019s core advertising business is expected to show strength, driven by increased marketer spending on Search and YouTube. Google Cloud, which grew 63% year-over-year in the first quarter to $20.03 billion, could post another solid quarter if enterprise AI demand continues to accelerate.<\/p>\n
Analyst consensus on TipRanks rates GOOGL a Strong Buy, based on 29 Buy and five Hold ratings, with an average price target of $437.79 implying roughly 24% upside from current levels.<\/p>\n
Yet the Frozen v2 news also highlights persistent challenges. Bloomberg reported last week that Google has delayed the launch of its next-generation Gemini Pro model after it fell short of internal goals, particularly in coding capabilities. The company has also lost several senior AI researchers to competitors, though Alphabet\u2019s first-quarter results\u2014including EPS of $5.11 against a $2.63 estimate\u2014suggest the brain drain narrative may be overstated.<\/p>\n
Google merged its Brain and DeepMind units under Demis Hassabis, creating what supporters call the densest concentration of AI PhDs in the world. The company has also structured multi-million-dollar retention packages that match startup equity while offering access to internal TPU clusters that smaller rivals cannot replicate.<\/p>\n
A Long-Term Bet with Near-Term Limits<\/p>\n
For all its promise, Frozen v2 cannot solve Google\u2019s immediate compute problems. With a 2028 target, the chip will not alleviate the power constraints and capacity bottlenecks the company faces today. And the rapid pace of AI model evolution means the hardware could be outpaced by the time it reaches production.<\/p>\n
Still, the project signals that Google is willing to bet on a future where foundation model architectures stabilize enough to justify locking them into silicon. If that bet pays off, the efficiency gains could reshape the economics of running large-scale AI inference.<\/p>\n
\u201cThe fact that we own frontier models and own the silicon really helps us stay ahead of the curve,\u201d CEO Sundar Pichai said on the company\u2019s first-quarter earnings call, a statement that now appears to preview the Frozen v2 philosophy.<\/p>\n","protected":false},"excerpt":{"rendered":"Alphabet Inc. (GOOGL) shares surged Monday after a report revealed Google is developing a radical new server chip…\n","protected":false},"author":2,"featured_media":112779,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[9],"tags":[1483,7718,57585,2408,132,2194,1430,58,658,141,4281],"class_list":["post-112778","post","type-post","status-publish","format-standard","has-post-thumbnail","category-google","tag-alphabet","tag-demis-hassabis","tag-frozen-v2","tag-gemini","tag-google","tag-google-cloud","tag-google-gemini","tag-nvidia","tag-spacex","tag-sundar-pichai","tag-tpu"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/112778","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=112778"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/112778\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/112779"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=112778"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=112778"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=112778"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}