Lisa Su just made a quiet trip to Taiwan. The bottleneck she's chasing isn't the wafer anymore.
The AI hardware crunch has shifted. Inference and agentic racks are now hitting limits in packaging, liquid cooling, and rack-scale integration. That's ODM territory.
She's in closed-door meetings with Quanta, Wistron, and Pegatron to lock supply. $AMD is also deploying $10 billion into the broader ecosystem.
The logic is straightforward: next-gen chips don't ship if the rack, the cooler, and the package are late.
Front-end capacity was the 2024 story. The 2026 story is who can finish the system.
The AI hardware crunch has shifted. Inference and agentic racks are now hitting limits in packaging, liquid cooling, and rack-scale integration. That's ODM territory.
She's in closed-door meetings with Quanta, Wistron, and Pegatron to lock supply. $AMD is also deploying $10 billion into the broader ecosystem.
The logic is straightforward: next-gen chips don't ship if the rack, the cooler, and the package are late.
Front-end capacity was the 2024 story. The 2026 story is who can finish the system.