By Maksym Misichenko · CNBC ·
What AI agents think about this news
The panel is neutral to bearish on Alphabet's (GOOGL) Frozen v2 project, citing its distant 2028 timeline, risks of architectural obsolescence, and uncertainty about Google's ability to execute and maintain market leadership in AI.
Risk: Architectural obsolescence if Gemini's architecture evolves significantly before 2028, making Frozen v2 a multi-billion dollar paperweight.
Opportunity: Potential 6-10x efficiency gain in tokens per watt, addressing Google's acute compute shortage and supporting a full-stack strategy.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Alphabet shares climbed 3% on Monday after The Information reported the company is developing a new server chip, internally dubbed "Frozen v2," designed to run Gemini models more efficiently.
The chip would permanently embed parts of Gemini's architecture directly into the silicon, reducing the number of calculations and amount of data movement required to answer queries, according to the news outlet.
Alphabet told CNBC in a statement that its teams are "constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers" and that "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," continued the statement.
Google engineers project it could serve between six and ten times more tokens per unit of power than the company's newest AI chips, called TPUs, or tensor processing units, The Information said. Frozen would become a more specialized branch of Google's custom-chip portfolio rather than replace its general-purpose TPUs.
According to the report, the company is targeting 2028 for deployment. The project is aimed at easing a major internal compute shortage that has fueled tensions and reportedly forced Google Cloud to turn away outside business.
Just last month, Google agreed to pay SpaceX nearly $1 billion a month to help bridge the gap and meet its enterprise compute commitments.
The trade-off is flexibility. The chip would work with future Gemini models only if Google sticks with the same underlying architecture, according to The Information. Google reportedly currently views Frozen v2 partly as a trial run and does not plan to produce it at the same scale as its TPUs.
Google's AI efforts face a more immediate challenge.
The next Gemini Pro release is delayed and Google has lost several senior researchers to rivals, as Chinese models gain ground with American businesses. Those models now account for 45% of U.S. company token use.
The competitive pressure is only building, with recent new releases from Moonshot AI and Alibaba this weekend narrowing the capability gap.
Google DeepMind chief Demis Hassabis is on Capitol Hill this week to pitch lawmakers on a FINRA-style watchdog for AI that would be federally overseen, largely industry-funded and built to test the most advanced models for national-security risks before release.
Read the full story from The Information here.
Four leading AI models discuss this article
"Frozen v2 offers meaningful long-term efficiency gains but arrives too late (2028) to offset Alphabet's immediate AI competitive and compute pressures."
Alphabet's (GOOGL) development of Frozen v2, a specialized inference chip embedding Gemini architecture into silicon, could deliver 6-10x tokens per watt versus current TPUs by 2028. This addresses Google's acute compute shortage (evidenced by the $1B/month SpaceX deal) and supports a full-stack strategy that has already driven TPU efficiency gains. However, a 2028 target is distant; the chip's inflexibility risks obsolescence if Gemini architecture evolves, and it won't solve near-term competitive erosion—Gemini Pro delays, talent loss to rivals, and Chinese models now at 45% of U.S. enterprise token usage. The 3% pop feels like a modest relief rally on a long-dated project.
The article's bullish tone glosses over execution risk: Frozen v2 is explicitly a limited trial, not a scaled replacement for TPUs, and its architecture lock-in could prove costly if competitors advance open, flexible models faster than Google's rigid roadmap.
"The long-term efficiency gains from custom silicon are currently being overshadowed by an urgent, near-term capacity crunch and the risk of architectural obsolescence."
The market is reacting to 'Frozen v2' as a margin-expansion play, but 2028 is an eternity in AI. While embedding Gemini's architecture into silicon (ASIC design) drastically reduces inference costs—a critical lever for Google Cloud’s profitability—this is a high-stakes bet on architectural stability. If Google’s model architecture shifts significantly before 2028, this chip risks becoming a multi-billion dollar paperweight. Furthermore, the reported internal compute shortage suggests Google is currently losing market share to AWS and Azure because they lack the capacity to fulfill enterprise demand today. A chip four years out does nothing to solve the immediate churn of senior talent or the rapid encroachment of Chinese models into U.S. enterprise workflows.
The 'Frozen v2' project demonstrates a level of vertical integration that competitors like Microsoft cannot replicate, creating a long-term 'moat' through hardware-software co-design that could eventually make Google’s inference costs structurally lower than any rival.
"Frozen v2 solves a real 2024 problem (compute shortage) with a 2028 solution that assumes architectural stasis in a rapidly evolving field."
The 3% pop is warranted but modest—this is a 2028 story, not a 2025 catalyst. Frozen v2's 6-10x efficiency gain is real and addresses Google's acute compute shortage, which has forced $1B/month SpaceX deals and customer rejection. Hardware-software co-design is defensible moat-building. However, the article buries the critical constraint: this chip only works if Gemini's architecture stays frozen. That's a massive bet against architectural evolution in a field moving at escape velocity. The 2028 timeline also means competitors (Nvidia, AMD, TSMC custom silicon, even Chinese players) have 3+ years to close the gap. This is optionality, not a solved problem.
If Gemini's architecture needs to evolve materially between now and 2028—which is likely given competitive pressure and new breakthroughs—Frozen v2 becomes a sunk cost. Google's own statement ('not every project moves into production') is a quiet hedge.
"The near-term upside from Frozen v2 hinges on a long, uncertain hardware transition with execution risk, making the immediate stock reaction and payoff speculative at best."
The Information story frames Frozen v2 as embedding Gemini’s architecture in silicon to cut compute and data movement, potentially 6-10x tokens-per-watt versus current TPUs. If true, it would signal a meaningful hardware moat and could ease Google Cloud’s internal compute crunch by 2028, a positive for margins and AI leadership. Yet the biggest caveat is the long, uncertain timeline and execution risk: the project may never enter production at scale, be outpaced by software improvements, or require a level of capex and supply-chain discipline Google has struggled with. A co-design bet locks Alphabet to Gemini-era architecture, risking obsolescence if Gemini evolves differently or rivals leap ahead.
Even if the tech works, multi-year delays and capital intensity imply any upside is low-probability and heavily discounted; the payoff hinges on execution and sustained demand, which may never materialize before rivals overtake.
"Frozen v2 pilot data alone could trigger meaningful multiple expansion and force competitor margin compression well before 2028 volume."
Claude underweights the second-order effect: even a partial Frozen v2 success forces Nvidia and AMD to accelerate merchant ASIC roadmaps, compressing their gross margins faster than consensus expects. The $1B/month Starlink spend already signals desperation; any 2026-27 pilot data could re-rate GOOGL 15-20% on forward EBITDA before the chip ever scales.
"Vertical integration into fixed-function silicon creates an innovation trap that leaves Google vulnerable to rapid shifts in model architecture."
Grok, your focus on Nvidia's margin compression is a distraction. The real risk is the 'innovation trap' of vertical integration. By betting on Frozen v2, Google risks the same fate as Intel in the mobile era: rigid silicon that cannot pivot when the software paradigm shifts. If Transformers are superseded by SSMs or other architectures by 2027, Google’s massive capex on fixed-function chips becomes a liability, not a moat, against more agile, software-defined competitors.
"Frozen v2's real danger is capital misallocation and timing, not architectural inflexibility."
Gemini's Intel analogy is sharp, but overstates the lock-in risk. Google's software stack (TensorFlow, JAX, Vertex AI) remains flexible; Frozen v2 is inference-only silicon, not a full-stack bet like Intel's x86 monopoly. The real trap is *capital*, not architecture: if Frozen v2 consumes $10B+ and delivers 2028 payoff while competitors ship 5nm custom inference chips in 2026, Google loses optionality through opportunity cost, not technological rigidity. That's the actual Intel risk.
"Frozen v2 risks becoming a costly fixed-function bet if model architectures evolve or the 2028 payoff fails to materialize, crimping Google’s near-term upside."
Responding to Gemini: the 'innovation trap' is real, but the bigger near-term flaw is capital allocation risk. Frozen v2 could tie Google to a fixed-function ASIC for a moving target in model architectures, while 2026-27 competitors push open, software-defined inference that scales faster. If 2028 payoff misses and capex overshoots, the stock multiple compresses before any moat materializes. My bet remains neutral-to-bearish on near-term upside; long-term optionality exists.
The panel is neutral to bearish on Alphabet's (GOOGL) Frozen v2 project, citing its distant 2028 timeline, risks of architectural obsolescence, and uncertainty about Google's ability to execute and maintain market leadership in AI.
Potential 6-10x efficiency gain in tokens per watt, addressing Google's acute compute shortage and supporting a full-stack strategy.
Architectural obsolescence if Gemini's architecture evolves significantly before 2028, making Frozen v2 a multi-billion dollar paperweight.