LinkedIn has announced it will keep its GPU investment flat for the next fiscal year, choosing to hold its compute and storage footprint steady rather than expanding AI data centers. The company says it has doubled the output of its existing GPUs over the past six months through accumulated improvements in utilization, model distillation and workload allocation.
How LinkedIn Doubled GPU Output
Erran Berger, LinkedIn's engineering CTO, described the goal as keeping the compute footprint flat while shipping more intensive AI features into production. The efficiency gains came from a combination of improvements rather than any single innovation. The company focused on:
Raghu Hiremagalur, LinkedIn's CTO for infrastructure, said the cost of serving each query had been climbing steadily while stored data was doubling annually. This trend, he said, was unsustainable. The efficiency push reads less like a strategic choice about the AI market and more like a company that saw its cost curve and decided to bend it.
A Contrast With Microsoft's Spending Spree
Despite being part of Microsoft, LinkedIn's approach diverges sharply from its parent company. Microsoft recently closed its fiscal year with $41 billion in capital expenditure in one quarter alone, adding 31 data centers and expecting to spend more than $50 billion in the current quarter. The company's spending is overwhelmingly driven by Azure customer demand, particularly capacity contracted to supply OpenAI, rather than internal product workloads.
LinkedIn's compute needs are a rounding error compared with Microsoft's total. Still, the subsidiary's decision to hold GPU investment flat offers an alternative perspective when a large consumer platform can add generative AI features for a year without adding hardware. This runs counter to the prevailing assumption that AI features and capacity growth are inseparable.
Why This Matters
If LinkedIn's approach proves sustainable, it could weaken the argument that AI product ambition requires proportional growth in hardware. This matters for other companies facing similar cost pressures. The industry currently rewards announcements of capacity rather than efficiency, with the four largest US hyperscalers committing roughly $600 billion to $700 billion in capital expenditure for the calendar year. LinkedIn's strategy challenges that norm and could push more organizations to prioritize optimization over raw expansion. The immediate impact is on LinkedIn's own cost structure, but the broader implication is a potential shift in how the industry thinks about AI infrastructure.
What It Means for LinkedIn's Infrastructure
This year's plan is made possible by a decision that was initially seen as a retreat. In 2022, LinkedIn shelved its plan to migrate infrastructure onto Azure under a project codenamed Blueshift, citing difficulties with tooling and Azure's own demand pressures. Instead, the company committed to its own data centers in Oregon, Texas and Virginia. Hiremagalur now argues that owning the full stack is precisely what enables the efficiency gains, because the company can instrument every layer and treat efficiency as a standing investment rather than a one-off cost-cutting exercise. The approach is not without risk, but it positions LinkedIn as an outlier in a field dominated by daily capex announcements.



