Alphabet shares rose 3% on Monday following reports from The Information that Google is developing a specialized server chip, internally codenamed “Frozen v2,” to accelerate its Gemini AI models. By embedding Gemini’s architecture directly into silicon, the chip aims to improve efficiency and reduce the compute shortages currently straining Google Cloud’s enterprise operations.
The Technical Design of “Frozen v2”
Unlike Google’s general-purpose Tensor Processing Units (TPUs), the “Frozen v2” chip is designed for a singular, specialized purpose: running Gemini models. According to The Information, the chip architecture permanently embeds parts of the Gemini model directly into the hardware. This design minimizes the data movement and number of calculations required to process queries, which are traditional bottlenecks in AI inference.
Google engineers estimate the chip could process six to ten times more tokens per unit of power compared to the company’s current TPUs. While this offers significant efficiency, it creates a trade-off in flexibility. Because the architecture is hard-coded, the chip would only function with future Gemini iterations that share the same underlying design.
Did you know?
Google is currently prioritizing a “full stack” approach to AI. By co-designing hardware and software, the company aims to optimize systems for specific real-world workloads rather than relying solely on off-the-shelf components.
Addressing the Compute Shortage
The development of “Frozen v2” is a direct response to an internal compute scarcity that has limited Google’s ability to meet external demand. The pressure has been significant enough that Google Cloud has reportedly turned away potential business due to capacity constraints. In a move to bridge this gap, Google recently agreed to pay SpaceX nearly $1 billion a month to secure necessary computing resources.
Alphabet addressed these development efforts in a statement to CNBC, noting that teams are “constantly researching and experimenting with new innovations” to maximize performance. The company emphasized that while not every project reaches production, this rigorous exploration is a core component of its strategy to integrate hardware and software from the ground up.
Market Competition and Strategic Challenges
Google faces significant headwinds in the AI sector, including a delayed release for the next version of Gemini Pro and the departure of several senior researchers. These internal pressures are compounded by external competition from Chinese AI models, which now account for 45% of U.S. company token use, according to The Information.
Recent releases from firms like Alibaba and Moonshot AI have further narrowed the capability gap, forcing Google to accelerate its hardware roadmap. Meanwhile, Google DeepMind chief Demis Hassabis has been engaging with lawmakers on Capitol Hill this week. He is advocating for a FINRA-style watchdog for AI—a federally overseen, industry-funded body designed to test advanced models for national-security risks prior to their public release.
Pro Tip: When evaluating AI infrastructure stocks, analysts often look for companies that control their own hardware supply chains, as this reduces dependency on external semiconductor manufacturers during periods of high demand.
Frequently Asked Questions
- What is the “Frozen v2” chip? It is a specialized, custom-designed server chip from Google intended to run Gemini models with higher power efficiency than standard TPUs.
- When will the chip be deployed? According to reports, Google is targeting 2028 for the deployment of this technology.
- Will this replace Google’s TPUs? No. The project is viewed as a specialized branch of Google’s custom-chip portfolio rather than a replacement for general-purpose TPUs.
- Why is Google facing a compute shortage? Rapid demand for AI services has outpaced existing infrastructure capacity, leading the company to seek external partnerships while developing custom hardware solutions.
Are you tracking the shift in AI infrastructure? Subscribe to our newsletter for the latest updates on enterprise compute and cloud technology trends.
Worth a look