LinkedIn says GPU efficiency will hold its AI compute footprint flat

LinkedIn plans to keep its compute, storage, and GPU investment roughly flat this fiscal year even as it launches more generative AI features, WIRED reports. Executives say the company doubled GPU efficiency over the past six months.

That choice stands out while many large technology companies are racing to build data centers and secure scarce chips. LinkedIn says the constraint is intentional: it wants engineers to be more disciplined about how they use existing hardware.

The plan could still change if AI hardware demands rise faster than expected, but LinkedIn says it has already considered surging memory-chip prices. The company’s fiscal year runs through next June.

The broader signal is that production AI is entering a cost-control phase. For large platforms, the winning strategy may not be unlimited infrastructure spending, but better scheduling, model selection, caching, and engineering practices that make each GPU do more useful work.