While China floods capital into AI chips, the world's smartest money is betting the physics and supply chains won't bend to political will.
The Summary
- Baillie Gifford's Paulina McPadden says TSMC and SK Hynix have near-monopoly positions in advanced chips that China can't easily replicate, even as Beijing mobilizes stock and bond markets to fund AI infrastructure
- The bet: manufacturing moats in leading-edge semiconductors are deeper than capital allocation can solve
- For builders of AI agents and Web4 infrastructure, this means the chip supply chain stays concentrated in Taiwan and South Korea for the foreseeable future
The Signal
Baillie Gifford manages over $200 billion and has a track record of early bets on Tesla, Amazon, and Nvidia. When McPadden says TSMC and SK Hynix hold structural advantages in advanced node manufacturing, she's not making a geopolitical prediction. She's reading the physics and economics of extreme ultraviolet lithography, cleanroom operations, and yield optimization curves.
China's latest push mobilizes state-backed capital markets to bootstrap domestic AI chip production. The money is real. The intent is clear. But capital doesn't compress the learning curve on 3-nanometer processes or replace the tacit knowledge embedded in TSMC's fabs after decades of iteration.
"Near-monopoly positions in advanced semiconductors aren't policy choices. They're the result of compounding technical expertise that money alone can't buy back."
Here's what matters for the agent economy: the compute that powers frontier AI models, the inference chips that run distributed agent networks, and the memory that makes it all work at scale all flow through two chokepoints. TSMC makes the cutting-edge logic chips. SK Hynix dominates high-bandwidth memory for AI workloads. Both sit outside China's direct control, and both have moats measured in years of process refinement, not quarters of R&D spend.
McPadden's position is a bet on manufactured scarcity persisting. If she's right, three things follow:
- AI infrastructure costs stay high and geographically concentrated
- Western and allied AI labs maintain a structural compute advantage
- China's path to AI sovereignty runs through older nodes, less efficient architectures, and higher energy costs per inference
The Implication
If you're building AI agents or infrastructure that depends on cutting-edge inference, your supply chain risk is Taiwan and South Korea, not China. Watch TSMC and SK Hynix capacity announcements, not Beijing's funding pledges. The real leading indicator for Web4 compute availability is fab construction timelines in Tainan and Icheon, not policy white papers.
For asset allocators, McPadden's thesis is a reminder that in hard tech, first-mover advantages compound. The companies that got to 3nm first aren't just ahead. They're learning faster than anyone chasing them.