Google’s custom AI chip for Gemini: Frozen v2 targets 2028
Alphabet is developing a custom AI chip codenamed “Frozen v2” to run its Gemini workloads. The chip is planned for a 2028 launch and targets 6 to 10 times better token output per unit of power than Google’s latest Tensor Processing Units (TPUs).
Investors reacted quickly: Alphabet shares rose about 3% after the report, signaling renewed market confidence in Google’s AI infrastructure push. The company’s broader strategy is to cut inference costs using a purpose-built ASIC (application-specific integrated circuit), rather than relying on general-purpose accelerators.
Google’s chip effort is not new. It has been designing TPUs since 2015, and it recently deployed the seventh-generation TPU “Ironwood” in late 2025. “Frozen v2” is positioned as the next step from general AI acceleration toward chips optimized for a specific model family.
Competitors are also pursuing bespoke hardware. OpenAI and Anthropic are developing custom AI chips, driven by concerns that Nvidia GPUs can be expensive, supply-constrained, and sometimes suboptimal for particular inference workloads.
For investors, Alphabet’s AI capex outlook remains central: the article cites $180–$190 billion in planned capital expenditure focused on AI infrastructure. The potential upside is higher efficiency and lower per-token inference cost, but the main risk is execution—custom silicon is costly and slow, and “Frozen v2” must outperform what Nvidia, AMD, and others deliver by 2028.
Bottom line for the “custom AI chip for Gemini” thesis: efficiency gains could strengthen Alphabet’s AI economics, while timing and competitive performance remain key variables.
Neutral
This is mostly a tech-sector infrastructure story, not a direct crypto protocol or token catalyst. While Alphabet’s custom AI chip for Gemini could improve AI economics and lift sentiment toward large-cap tech, it does not change crypto network fundamentals, liquidity, or stablecoin flows in the near term.
That said, there can be a mild second-order effect: if AI infrastructure spending remains strong, it supports risk appetite in broad tech equities, which can indirectly influence crypto through correlation during market-wide “risk-on” periods. Historically, major AI hardware announcements (GPU/TPU-related) tend to move equity sentiment more than they move crypto by themselves.
Short term: the ~3% share reaction is unlikely to translate into sustained crypto volatility because there is no mention of crypto assets, exchanges, or blockchain integrations.
Long term: if reduced inference costs lead to faster Gemini deployment, the broader AI industry could grow, which may support demand for compute and data infrastructure. But the timeline (2028 target launch) is too distant to be a direct trading trigger for crypto markets today. Overall impact is best categorized as neutral.