AMD closed above $1 trillion in market value for the first time Monday, becoming the fourth U.S. chipmaker to cross that threshold, as a broad chip-stock rally sent Intel up more than 12% and Arm up more than 17% on no company-specific news from any of the three. The catalyst was investor reaction to Meta's newly launched Muse AI agent topping Apple's App Store, which revived bets that agentic AI inference — not just training — will drive CPU demand alongside GPU demand. The move is a reminder that a single hyperscaler product's reception can now reprice an entire sub-sector of chip stocks overnight, with the Philadelphia Semiconductor Index notching its fifth straight session of gains. (Bloomberg)
Private Companies
Astrusfundraise
Astrus closed a $12M Series A led by Caffeinated Capital, with Khosla Ventures returning after backing the Kitchener-Waterloo startup's seed round, to commercialize a reinforcement-learning system that automates physical design for analog chips — the manual, months-long layout step that can cost hundreds of millions on advanced nodes. The round funds the move from R&D into customers' hands, targeting one of chip design's last unautomated bottlenecks as demand for custom analog and mixed-signal silicon grows alongside AI accelerators. (BetaKit)
Public Markets
Intel led the session's individual gainers, up more than 12% for a fifth straight day of gains, as investors extended Monday's Muse-driven rally into a broader re-rating of CPU exposure across the sector. (Bloomberg)
Alibaba unveiled its Zhenwu V900 AI chip — 3x its predecessor's performance — alongside a $53B, three-year plan to reach 20GW of global data-center capacity by 2032, the clearest sign yet China's largest cloud provider is building its own AI full stack. (Bloomberg)
Emerging
Power for the next AI wave: Google agreed to fund capacity "uprates" at Southern Co. subsidiary Georgia Power's Vogtle and Hatch nuclear plants, adding roughly 96 megawatts of new supply through a new financing tariff as data-center demand strains the grid. The deal treats existing nuclear capacity, not new construction, as the fastest lever for matching AI power demand to supply — cheaper and faster than the small modular reactors several other hyperscalers have instead committed to build from scratch. (Bloomberg)
Faster inference over light: A new paper from Lightmatter researchers, Scaling Inference Prefill with High-Radix Photonic Interconnects, found that high-radix photonic interconnects cut prefill latency 2.1–3.2x in high-batch regimes and up to 5.8x in communication-limited configurations as context windows scale into the millions of tokens. The result, honored as best paper at IEEE Hot Interconnects 2026, is evidence that the scale-up fabric linking GPUs — not just the chips themselves — is becoming the binding constraint on how fast reasoning and agentic models can serve long-context requests. (arXiv)