LinkedIn says it will not expand its data-centre capacity or increase GPU spending during its current financial year, despite continuing to introduce more AI features. The decision stands apart from the heavy infrastructure spending by companies such as OpenAI, Meta and Google, and matters because it suggests a large technology company believes efficiency improvements can temporarily substitute for new hardware investment.

The company, which has more than 1.3 billion users, says it has doubled the efficiency of its existing GPUs over the past six months and will aim to keep compute and storage capacity broadly flat until next June. LinkedIn operates its own data centres in Oregon, Texas and Virginia after finding Microsoft’s Azure cloud uneconomic for its scale, while facing rising AI query costs and data storage that had been doubling each year. Executives acknowledge that rapidly changing AI hardware needs and higher memory-chip prices could still challenge the plan, but argue that tighter limits should encourage more disciplined engineering.