Alphabet shares climbed 3% on Monday following reports that the tech giant is developing a radical new server chip designed specifically to supercharge its Gemini artificial intelligence models.
Internally dubbed “Frozen v2,” the experimental chip represents a major shift in Google’s hardware strategy. According to The Information, the chip will permanently embed parts of Gemini’s software architecture directly into the physical silicon. By hardwiring these processes, the chip dramatically reduces the amount of data movement and calculations required to generate answers to user queries.
The efficiency gains could be staggering. Google engineers project that “Frozen v2” could serve between six and ten times more tokens per unit of power than the company’s current state-of-the-art Tensor Processing Units (TPUs).
Rather than replacing the general-purpose TPUs that currently act as the workhorses of Google’s data centers, Frozen v2 is intended to become a specialized branch of Google’s custom-chip portfolio.
In a statement to CNBC, Alphabet emphasized its commitment to pushing hardware boundaries.
“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.”
Easing the Compute Crunch
Google aims to deploy the new chip by 2028, a timeline heavily driven by an intensifying internal compute shortage. The lack of processing power has reportedly fueled internal tensions and even forced Google Cloud to turn away prospective outside business.
The compute deficit is so severe that just last month, Google struck an unprecedented deal to pay SpaceX nearly $1 billion a month to help bridge the processing gap and meet its enterprise cloud commitments.
However, the incredible efficiency of Frozen v2 comes with a major trade-off: a loss of flexibility. Because the chip’s silicon is physically modeled around Gemini’s current architecture, it will only work with future iterations of Gemini if Google maintains that exact same underlying structural design. Due to this rigidity, Google reportedly views the chip as a trial run and doesn’t plan to mass-produce it at the massive scale of its TPUs.
Mounting Competitive Pressure
The news of Frozen v2 arrives at a critical juncture for Google’s AI ambitions.
The company is facing intense, multi-front pressure. The release of its highly anticipated next-generation model, Gemini Pro, has been delayed, and the company has recently lost several senior researchers to rival AI labs.
Furthermore, Google is battling an unexpected surge from overseas competitors. Chinese startups like Moonshot AI—which recently unveiled its new Kimi model—and tech giants like Alibaba are rapidly closing the capability gap. According to recent data, Chinese models now account for an astonishing 45% of AI token usage among American businesses.
As the technological arms race accelerates, Google is also looking toward Washington to help shape the rules of engagement. Google DeepMind CEO Demis Hassabis is on Capitol Hill this week pitching lawmakers on a new regulatory framework. Hassabis is advocating for a FINRA-style, industry-funded watchdog that would be federally overseen and tasked with stress-testing the world’s most advanced AI models for national security risks prior to their public release.ers and more than 4.5 GW of secured power capacity to fuel its long-term expansion.
Leave a Reply