Google's Frozen v2 chip boosts Alphabet 3%, revives semiconductor bull case

On Monday, Alphabet stock climbed about 3% following a report from The Information that Google is developing Frozen v2, a new AI chip designed to run its Gemini models up to 10 times more efficiently. The innovation works by baking key parts of the Gemini model directly into the silicon, which reduces data movement and the power consumption required per inference (the step where an AI produces an answer).

The technical leap is significant relative to Google's historical progress. Google engineers believe Frozen v2 could serve six to 10 times more tokens—the small chunks of text AI models produce—per unit of power compared to Ironwood, Google's seventh-generation TPU. Ironwood itself only doubled the performance per watt of its predecessor, making a 6–10× gain a substantial jump. However, the chip comes with a constraint: it will only work with future Gemini models if Google maintains the same core design, and it will support existing TPUs built since 2016 rather than replace them. The launch is targeted for 2028.

Google's move reflects acute competitive pressure in AI infrastructure. The company faces a compute shortage that has reportedly forced Google Cloud to reject customer business. In response, Google agreed last month to pay SpaceX nearly $1 billion（約1600億円） a month to bridge the capacity gap. Amazon, Microsoft, and Meta are all pursuing similar strategies by building their own AI chips, underscoring the industry-wide shift toward vertical integration of hardware and software.

The timing of Alphabet's announcement proved fortuitous for semiconductor bulls. Chip stocks had just closed their worst week in more than a year. The VanEck Semiconductor ETF (SMH) fell 8.9%—its worst week since April 2025, when it lost 15%—and the Philadelphia SE Semiconductor Index (SOX) dropped nearly 10%. Fear stemmed from multiple sources: China's Kimi K3, an open-weight reasoning model released by startup Moonshot AI, sparked concerns about an AI shakeout, and large funds had spent weeks pulling back from Nvidia following cracks in the memory-stock rally.

Wall Street's response has been mixed but constructive. Morgan Stanley analyst Joseph Moore called the memory selloff an attractive entry point and favors Nvidia (NVDA) and Broadcom (AVGO), while Mizuho analyst Vijay Rakesh agreed in a client note, stating: "We believe there is a lot of gas left in the tank as AI capex … and [gigawatt] power installations ramp well beyond 2028–29E … all driving semiconductor demand with a wide supply gap." JPMorgan's Mislav Matejka expects buyers to return soon, arguing that new chip supply will not arrive before 2028. However, Evercore's Mark Lipacis projects another 10% to 15% of downside ahead, though his data indicates the SOX typically bounces 36% within 20 weeks of corrections like the current one. All eyes now turn to Alphabet's Wednesday earnings report, which will reveal whether Google is backing its chip ambitions with commensurate capital spending.

Alphabet's announcement of the Frozen v2 chip arrives at a pivotal moment for semiconductor stocks. The sector just endured its worst week in more than a year, with the VanEck Semiconductor ETF falling 8.9% and the Philadelphia SE Semiconductor Index dropping nearly 10%, driven by concerns about an AI slowdown following China's Kimi K3 release and a broader pullback in memory-stock valuations. Against this backdrop, Google's commitment to a next-generation chip with 6–10× efficiency gains signals that major cloud providers view AI infrastructure expansion as essential through 2028 and beyond, not a temporary phenomenon.

Wall Street's response has been divided but cautiously optimistic. Morgan Stanley's Joseph Moore and Mizuho's Vijay Rakesh both called the memory selloff an attractive entry point for investors, while Rakesh emphasized that AI capex and power-installation ramps will drive semiconductor demand "well beyond 2028–29." JPMorgan's Mislav Matejka expects buyer appetite to return soon, citing the fact that new chip supply from competitors will not arrive before 2028. Even Mark Lipacis at Evercore, who projects another 10% to 15% downside, acknowledges that the SOX typically bounces 36% within 20 weeks of corrections of this magnitude. The critical test comes Wednesday, when Alphabet's earnings report will reveal whether management plans to fund the Frozen v2 roadmap with the capital spending needed to make it credible.

No comments yet. Be the first to share your thoughts!