AI Data Centers Hit Bottlenecks in Billions of Tiny Parts as Racks Need 10,000s of Links
Updated
Updated · Bloomberg · May 12
AI Data Centers Hit Bottlenecks in Billions of Tiny Parts as Racks Need 10,000s of Links
11 articles · Updated · Bloomberg · May 12
Bottlenecks are spreading across AI data centers beyond Nvidia chips, with shortages and production strain emerging in memory chips, battery systems and other small but critical hardware.
Each new AI chip generation demands more supporting components, pushing server racks from dozens of fiber-optic connections to tens of thousands and sharply raising cooling and power requirements.
Those systems depend on precision-built parts—tiny motors, fans, liquid-cooling channels and copper plates—where microscopic flaws can shut equipment down.
The surge in demand for billions of such components is creating a new choke point for AI infrastructure, while also opening major opportunities for specialized suppliers.
With key parts sold out for years, is the multi-trillion dollar AI infrastructure boom headed for a sudden halt?
As supply chains for tiny parts falter, must we fundamentally rethink the physical design of artificial intelligence itself?
Beyond the AI chip race, what is the hidden environmental cost of the billions of tiny components fueling it?