The AI infrastructure paradox: over-ordering and under-building
What memory fabs and speculative data center queues reveal about the durability of AI demand.
The AI infrastructure market is experiencing a paradox of over-ordering and under-building, with memory shortages persisting despite high demand. Memory prices have surged due to supplier prioritization of server applications and cloud providers, while data center developers have been applying for grid connections at a rapid pace.
However, the actual installed capacity may lag behind the announced capacity due to the time-consuming nature of constructing fabrication plants and the uncertainty surrounding model efficiency and architectural changes. While memory and storage companies continue to invest in US manufacturing and research, the slow arrival of new capacity highlights the disconnect between declared demand and irreversible investment.
Brief written by urgent.news; may contain errors — check the original.
Source: techradar.com