AI must leave the data center and work in the real world. Jensen Huang
Cloud subscriptions cannot fund the US $6.7 trillion required for AI data-center expansion. Every AI inference has a real energy cost, and the economics of cloud-only AI are structurally negative. Only physical-world deployment can generate the recurring revenue and margin needed to sustain artificial intelligence over time.
Hardware capacity in California is world-class — but heavily focused on design and largely made overseas. “From a maker room to a Chinese factory with almost nothing in between”
California can (re)build hardware — but not at the scale, speed, and breadth demanded by the AI explosion. The ability exists, but it is concentrated in a handful of hyperscalers. The new reshoring wave requires a broader ecosystem: robotics, mid-stage manufacturing, and distributed industrial capacity.
Democratized reshoring requires distributed robotics, mechatronics, and manufacturing capacity available beyond the hyperscalers. Projected Physical-AI reshoring in the US represents a US$ 100–300B/year opportunity by 2035 — supported by the combined growth of robotics (Deloitte, BCG), edge AI devices (McKinsey), and TCO-based reshoring potential (~US$ 200B/year; AMT). Given that California already produces ~14% of US GDP and remains the #1 manufacturing state (BEA), the Bay Area can credibly anchor 10–20% of this wave — reaching US$ 3–8B/year as early as 2027.
The Magnet is our answer.
The (Un)Perfect Valley
Articles
