- Alphabet is developing an inference chip code-named 'Frozen v2' that hardwires Gemini's architecture into silicon, first reported by The Information [1] - Google engineers project the chip will deliver 6-to-10x more AI tokens per unit of power than the company's latest TPUs [2] - Deployment is targeted for 2028; the chip complements rather than replaces existing TPU infrastructure [3] - GOOGL shares rose approximately 3% on Monday following the report, trading around $356 [4] - A Google spokesperson cautioned that 'not every experimental project ultimately becomes a commercial product'
[3] Alphabet is developing a next-generation AI inference chip, internally code-named "Frozen v2," that embeds portions of its Gemini model architecture directly into silicon hardware, according to a report by The Information published Monday. The chip represents a departure from Google's existing tensor processing units by hardcoding model logic into the chip itself rather than running it as software [1].
Google engineers project the chip will serve between six and ten times more AI tokens per unit of power consumption than the company's newest TPUs, a significant efficiency leap that addresses mounting pressure from ballooning AI infrastructure costs [2]. Alphabet's capital expenditures hit $35.7 billion in the first quarter alone, and the company has guided full-year 2026 capex to between $180 billion and $190 billion
.
[1]GOOGL shares climbed roughly 3% on Monday following the report, trading at approximately $356.23, according to market data. Analysts maintain a consensus Buy rating on the stock with an average price target of $429.67 [4].
The Chip #
Frozen v2 takes a fundamentally different approach from Google's general-purpose TPUs. Rather than relying on programmable silicon that runs AI models as software, the chip permanently embeds elements of Gemini's neural network architecture into the hardware itself. This reduces the amount of computation and data movement required to generate AI responses during inference — the process of serving predictions and outputs to users [2].
The design draws conceptual parallels to Apple's strategy of tightly coupling hardware and software, though applied to AI model serving rather than consumer devices. Engineers can update the system through new model weights without redesigning the silicon, but the underlying architecture must remain consistent [5].
The trade-off is flexibility. Because Gemini's architecture is baked into the hardware, the chip would only work with future Gemini models if Google maintains the same underlying model structure. A major architectural shift would risk making the specialized chips obsolete [1].
Why It Matters #
The project is driven by severe AI compute shortages within Google. Computing capacity gaps have caused internal resource conflicts and forced Google Cloud to turn away enterprise customer orders, a direct revenue constraint at a time when AI demand is surging across the industry [5].
With Alphabet's annualized capex run rate approaching $190 billion, efficiency gains of 6-to-10x per watt on inference workloads represent potentially enormous cost savings. Inference — serving AI responses to hundreds of millions of users — accounts for a growing share of Google's total compute spend as Gemini is integrated across Search, Workspace, and Cloud products [1].
The chip also signals the broader hyperscaler trend of developing proprietary silicon to reduce dependence on Nvidia's GPUs. Amazon has pursued a similar strategy with its Trainium and Inferentia chips, while Microsoft has developed Maia for AI workloads [2].
Scale and Timeline #
Alphabet currently views Frozen v2 partly as a trial run and does not plan to produce it at the same scale as its TPUs, which are manufactured by the millions and deployed across Google's global data center fleet [1]. The chip is targeted for deployment around 2028, though the design remains in development.
A Google spokesperson cautioned that 'not every experimental project ultimately becomes a commercial product,' framing the effort as a research exploration rather than a committed product roadmap [3]. The company declined to provide additional technical details.
Market Reaction #
Alphabet shares rose approximately 2.9% on Monday to around $356.23 following the report. The stock carries a consensus Buy rating from Wall Street analysts, with an average price target of $429.67, implying roughly 20% upside from current levels [4].
Recent analyst upgrades have been bullish: TD Cowen has a $475 target, Oppenheimer at $445, and JPMorgan at $460 [4]. The stock trades at roughly 26.4x forward earnings, reflecting a premium valuation tied to expectations that AI investments will drive long-term revenue growth
.
[4]The chip report adds to a string of AI infrastructure developments that have supported Alphabet's stock this year, even as investors have questioned the pace of the company's capital spending. Alphabet's full-year 2026 capex guidance of $180 billion to $190 billion represents more than double the prior year's spending [1].
Companies mentioned #
Further sources #
[1] CNBC: Alphabet stock pops on report it's developing a more efficient AI chip ↗ [2] QuiverQuant: Google Reportedly Developing 'Frozen v2' AI Chip to Boost Gemini E… ↗
[[3] Yahoo Finance: Alphabet stock gains on report of next-generation AI chip project ↗](https://finance.yahoo.com/technology/ai/articles/alphabet-stock-gains-report-next-141621968.html)
[[4] Benzinga: Why Is Alphabet Stock Gaining Monday? ↗](https://www.benzinga.com/markets/tech/26/07/60557742/why-is-alphabet-stock-gaining-monday)
[5] TradingKey: Technical Details of Google's New AI Inference Chip 'Frozen v2' Rev… ↗
The stories that matter, in one email. Free — unsubscribe anytime.