AI hardware · Layer
Supply constraints and bottlenecks
AI hardware is a chain of dependent steps, and the slowest link sets the pace. Understanding where the binding constraints sit explains why compute is scarce even when one input is plentiful.
Key facts
- The recurring chokepoints are HBM, advanced packaging (CoWoS-class), and grid power, more than raw wafer capacity.
- Each chokepoint has long lead times to expand, so shortages persist for quarters or years, not weeks.
- Concentration compounds scarcity: leading-edge fabrication, EUV, and HBM each rest on very few suppliers.
- Export controls add a policy layer on top of the physical chain, restricting where the most advanced hardware can go.
Where it bottlenecks
The binding constraint rotates between HBM, packaging, and power; relieving one often just exposes the next.
Who dominates it
TSMCSK HynixASMLNVIDIA