The AI infrastructure paradox: over-ordering and under-building
Reported by TechRadar1 source ↓
Memory supply is tight as AI developers seek capacity, while data center connection applications outpace what grids can accommodate. The contrast shows why proposed projects are not the same as funded, operational demand.
Key points
- TrendForce put first-quarter 2026 conventional DRAM contract-price increases at roughly 93% to 98%
- UK grid connection applications rose from 41GW to 125GW in less than a year
- Ofgem says data center projects accounted for at least 80GW of the increase
- SK Hynix warned that high-bandwidth memory shortages may persist until 2030
- Ofgem proposed fees and milestones to test whether grid applicants are ready to build
Details
TechRadar reports that memory suppliers have shifted production toward server memory and high-bandwidth memory as AI demand grows, tightening supplies for other buyers. New fabrication capacity takes years to build and qualify, and TrendForce expects most near-term supply growth to come from process improvements rather than major increases in wafer capacity.
The article contrasts that physical constraint with proposed data center demand. In Britain, grid connection applications rose from 41GW to 125GW in less than a year; Ofgem attributes at least 80GW of the increase to data center projects. The regulator has proposed commitment fees and additional milestones, including evidence of procurement, to distinguish prepared projects from speculative queue positions.
TechRadar notes that developers may seek more capacity early because land, power and equipment are scarce, but applications do not establish that projects have customers, financing or a route to power. The article also says supplier investment in AI companies can support purchases of computing infrastructure, making it harder to distinguish demand funded by end customers from demand supported by capital within the supply chain.
Written from coverage as of
Images
- TechRadar