Revolutionizing Chips: Nvidia's Localized Blackwell Breakthrough

Daily Nvidia Brief — October 18, 2025

“AI factories are the new data centers.” — Jensen Huang

  1. First U.S.-made Blackwell wafer unveiled with TSMC in Arizona
  • What’s new: Nvidia and TSMC showed the first wafer for Blackwell chips made in the U.S., at TSMC’s Phoenix fab. (reuters.com)
  • Bigger vision: Localizing advanced chip supply strengthens Nvidia’s “AI factory” agenda—bringing compute, power, and manufacturing closer together for scale and resilience. (reuters.com)
  • AI impact: More onshore capacity should ease GPU shortages and speed rollout of next‑gen training and inference, benefiting the whole ecosystem. (reuters.com)
  1. Huang: Nvidia’s advanced‑chip share in China has fallen to “zero”
  • What’s new: Jensen Huang said Nvidia’s position in China’s advanced AI‑chip market has dropped from ~95% to zero amid export curbs; Nvidia has stopped including China in guidance. (scmp.com)
  • Bigger vision: Nvidia is leaning into diversified global AI factories and sovereign builds to offset the China gap, keeping its growth tied to many regions instead of one. (cnbc.com)
  • AI impact: Splintered supply chains push parallel AI stacks (China vs. rest‑of‑world), shaping where models are trained and deployed. (cnbc.com)
  1. CoreWeave + Poolside plan a 2‑GW “AI factory” in West Texas using Nvidia GB300s
  • What’s new: CoreWeave will supply 40,000+ Nvidia GB300 NVL72 GPUs starting December 2025 to power Poolside’s Horizon campus, targeting 2 GW over time. (barrons.com)
  • Bigger vision: This aligns with Nvidia’s full‑stack, partner‑led buildout—pairing chips, networking, and energy‑aware campuses to industrialize AI compute. (wsj.com)
  • AI impact: Purpose‑built, power‑secure sites lower costs and expand global training capacity, accelerating progress in code‑generation, agents, and multimodal models. (wsj.com)
  1. Robotics momentum: open Newton physics engine and Isaac GR00T updates
  • What’s new: Nvidia released the open‑source Newton physics engine in Isaac Lab and advanced its GR00T robot foundation model and Omniverse libraries; leading robotics groups are adopting them. (nvidianews.nvidia.com)
  • Bigger vision: Beyond data centers, Nvidia’s vision extends to “physical AI”—standardizing the software stack so robots learn in sim and work reliably in the real world. (nvidianews.nvidia.com)
  • AI impact: Better sim‑to‑real tools and open models should cut development time and widen access, pushing robotics from pilots to everyday operations. (nvidianews.nvidia.com)
  1. Reinforcing the AI networking moat (follow‑through from this week)
  • What’s new: Meta and Oracle are standardizing on Nvidia’s Spectrum‑X Ethernet to link massive GPU clusters with higher efficiency. (investor.nvidia.com)
  • Bigger vision: Own the “nervous system” of the AI factory—from GPUs to switches—so customers can stitch millions of GPUs into one giant computer. (investor.nvidia.com)
  • AI impact: Faster, steadier networks reduce training times and costs, bringing larger models to production sooner. (investor.nvidia.com)

— End of brief —