Alphabet shares climbed about 3% on Monday following reports that Google is developing a new server chip, internally named Frozen v2, designed to run its Gemini AI models with significantly higher power efficiency. The chip aims to embed parts of Gemini's architecture directly into silicon, reducing the calculations and data movement required for inference. Engineers estimate Frozen v2 could serve six to ten times more tokens per unit of power than Google’s current TPUs, though it will not replace them and would remain a specialized branch focused on Gemini workloads. A potential launch is targeted for 2028, pending continued development.
Meanwhile, Chinese startup Moonshot AI released its open-weight model Kimi K3, boasting 2.8 trillion parameters and a one-million-token context window. Analysts at Stifel suggest the low-cost model could benefit cybersecurity and AI infrastructure firms, including CrowdStrike, Cloudflare, and Palo Alto Networks, as cheaper AI drives higher consumption – a phenomenon aligned with Jevons Paradox. Memory chip makers SK Hynix and Samsung Electronics are seen as direct winners given the model’s massive memory requirements, while Nvidia and AMD face renewed questions about premium pricing amid a flood of cheaper alternatives. The development intensifies global AI competition, with Google facing pressure from both U.S. enterprise demand and Chinese model adoption, reportedly reaching 45% of company token usage.
