Google is working on a new AI chip designed to make Gemini more efficient
Original reporting by TechCrunch
"Frozen v2" refers to Alphabet’s internally developed server chip, reportedly designed to dramatically improve the operational efficiency of its Gemini AI models. Slated for release around 2028, this new chip is projected to be between six and ten times more efficient than Google’s current AI hardware, as measured by tokens generated per unit of power. This significant leap in efficiency aims to directly address the immense computational demands of cutting-edge AI. While Google offered a non-committal response to the report, emphasizing its continuous research into hardware-software co-design, the news highlights a critical industry trend.
The Custom Chip Race Across the artificial intelligence landscape, major players are increasingly investing in proprietary chip development. This strategic shift is driven by a multifaceted need: to optimize the performance and energy consumption of their large language models, alleviate global AI computing capacity shortages, and reduce reliance on dominant third-party chipmakers like Nvidia. Such efficiency has become a key selling point as concerns about AI spending have tempered initial market euphoria, pushing companies to demonstrate clear returns. Competitors such as OpenAI, with its Jalapeño inference processor, and Anthropic, reportedly exploring a partnership with Samsung, are pursuing similar paths. For Alphabet, whose ambitious AI strategy involves expenditures nearing $190 billion, demonstrating tangible returns on these massive investments is paramount. The prospect of Frozen v2’s substantial efficiency gains quickly assuaged investor concerns about these vast outlays, leading to a notable 3% rise in the company’s stock following the report and boosting confidence ahead of its earnings.
Alphabet’s rumored “Frozen v2” chip represents more than just a technological upgrade; it signifies a strategic imperative for Google in the fiercely competitive AI landscape. By aiming for a six to ten-fold increase in efficiency for its Gemini models by 2028, Google is directly addressing the immense computational costs associated with advanced AI. This internal development underscores the company’s commitment to optimizing its vast AI investments, providing a clear pathway for these expenditures to translate into tangible operational gains and stronger investor confidence, as evidenced by the recent positive market reaction.
Redrawing the Hardware Map
This move by Google is emblematic of a significant industry-wide trend: major AI developers are increasingly designing their own custom silicon. Following similar initiatives from OpenAI and Anthropic, the push for proprietary chips is driven by a dual need to reduce dependence on external suppliers, notably Nvidia, and to achieve highly specialized performance optimized for specific AI workloads. This vertical integration, where hardware and software are co-designed from the ground up, promises to redefine the economics and capabilities of AI. In the coming years, this strategic shift will likely accelerate innovation in chip architecture, foster greater efficiency across the AI ecosystem, and ultimately lower the barriers to entry for advanced AI development, ensuring more sustainable and scalable growth for the technology. The era of generic, off-the-shelf AI computing is rapidly giving way to a new paradigm of bespoke, highly specialized hardware.
Frequently asked questions
- What is Google doing to make its AI models more efficient?
- Google's parent company, Alphabet, is developing a new server chip internally known as "Frozen v2." This custom hardware is specifically designed to power its Gemini AI models more efficiently. Expected around 2028, the chip aims to be significantly more efficient than existing AI accelerators, potentially generating six to ten times more tokens per unit of power. This initiative is part of a broader strategy to optimize AI operations and manage substantial investment costs.
- Why are major AI companies developing their own custom chips?
- Major AI companies are increasingly developing custom chips to enhance the efficiency of their proprietary AI models. This strategy helps them reduce operational costs, address global shortages in AI computing capacity, and lessen their dependence on third-party chip manufacturers like Nvidia, which has historically dominated the market. Custom hardware allows for tighter integration with their software, leading to highly optimized systems for specific AI workloads and improved performance.
- When is Google's new "Frozen v2" AI server chip expected to be released?
- Google's new custom AI server chip, reportedly named "Frozen v2," is anticipated to be released around 2028. This chip is designed to power the company's in-house Gemini AI models and aims for substantial efficiency improvements. Sources suggest it could be six to ten times more efficient than Google's current AI chips, measured by tokens generated per unit of power. Its development signifies Google's ongoing investment in optimizing its AI infrastructure.