8 hours ago
5
Key Takeaways
- Alphabet shares climbed 1.51% on Monday following revelations about a specialized server processor dubbed “Frozen v2”
- The processor would hardwire elements of Gemini’s design directly into hardware to reduce energy consumption
- Internal projections suggest 6–10x improvements in tokens processed per unit of energy versus existing TPUs
- Google is aiming for a 2028 rollout, though final specifications remain under development
- The project addresses severe internal capacity constraints that have resulted in Google Cloud declining customer engagements
Shares of Alphabet (GOOGL) advanced 1.51% during Monday’s trading session after The Information disclosed that Google is working on a proprietary server processor known as “Frozen v2.”
The processor aims to execute Gemini AI workloads with greater efficiency by integrating specific elements of the model’s framework directly into the chip architecture. This design strategy minimizes computational overhead and reduces data transfer requirements when processing user requests.
$GOOGL is developing a new AI chip that could run Gemini models 6 to 10 times more efficiently than its latest TPUs.
The chip, internally called “Frozen v2,” would bake parts of Gemini’s architecture directly into silicon, reducing data movement and the number of decisions… pic.twitter.com/O6dhl6twdm
— Wall St Engine (@wallstengine) July 20, 2026
According to internal estimates cited in the report, Google’s engineering teams believe the processor could deliver six to ten times more tokens per watt of energy consumption compared to the company’s existing tensor processing units.
In a statement to CNBC, Google indicated its teams are “constantly researching and experimenting with new innovations” and emphasized its practice of co-designing hardware and software “from the ground up” to maximize performance for production environments.
Addressing Critical Capacity Constraints
The initiative responds to significant internal infrastructure bottlenecks. These capacity issues have generated friction within Google and allegedly forced Google Cloud to reject prospective enterprise clients.
As recently as last month, Google committed to paying SpaceX approximately $1 billion monthly to supplement its capacity and fulfill enterprise service agreements.
Frozen v2 is intended to complement Google’s current TPU portfolio rather than serve as a replacement — functioning as a purpose-built solution for targeted applications.
The limitation is adaptability. The processor would remain compatible with subsequent Gemini iterations only if Google maintains architectural consistency. Google currently regards this as an experimental initiative and doesn’t intend to manufacture it at the volume of its standard TPUs.
The company is working toward a 2028 deployment timeline, though engineers continue refining the technical specifications.
Growing Competitive Challenges
This chip development emerges as Google confronts substantial AI sector challenges. The upcoming Gemini Pro version has encountered setbacks after failing to meet internal performance benchmarks, Bloomberg reported.
Google has also experienced departures of multiple key researchers to competing organizations in recent months.
Chinese AI platforms are expanding their presence among American enterprises, now representing 45% of token consumption by U.S. companies. Recent launches from Moonshot AI and Alibaba over the weekend are further closing the performance divide.
Google DeepMind CEO Demis Hassabis appeared before Congress this week advocating for a FINRA-inspired federal regulatory body for AI, primarily industry-funded, that would evaluate sophisticated models for national security implications prior to public deployment.
Alphabet shares extended gains by 0.97% in extended trading hours on Monday.
The post Alphabet (GOOGL) Stock Gains as Google Develops Specialized AI Chip for Gemini appeared first on Blockonomi.
English (US) ·