LinkedIn Holds AI Data Center Spending Flat — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- LinkedIn will keep its GPU investment steady and its compute and storage footprint flat for the fiscal year that began last month and ends next June, with CTO Erran Berger calling it "a pretty bold statement to make in today's world"
- LinkedIn doubled GPU efficiency over the past six months through techniques including model distillation, CPU-GPU workload balancing, and reworking software on Nvidia processors to handle tasks larger than they were designed for, saving roughly $24 million — or 1,100 GPUs running around the clock — over the past year
- LinkedIn runs its own data centers in Oregon, Texas, and Virginia after moving away from Microsoft Azure in 2022, with infrastructure CTO Raghu Hiremagalur citing GPU utilization "north of 95 percent" on the training side
- CTOs Berger and Hiremagalur framed the flat spending as a deliberate constraint to push engineering teams to be more creative, though Hiremagalur acknowledged server prices have jumped threefold in recent months
- Gartner analyst Chirag Dekate said enterprises are shifting from "a buy-more era to a do-more era" but warned that at some point companies must either compromise on AI ambitions or relax IT spending freezes
- HP board member Songyee Yoon, managing partner at Principal Venture Partners, called LinkedIn's move "encouraging," saying "the companies that win will not simply be the ones that spend the most on infrastructure"
Why it matters: With 1.3 billion users and $18 billion in annual sales, LinkedIn is the largest business yet to publicly break from the AI data center building boom championed by OpenAI, Meta, and Google. Its $24 million in efficiency savings is modest in scale but validates a "tokenomics" discipline play — extracting more compute from existing hardware rather than buying more chips — even as Gartner's Dekate warned such constraints cannot hold indefinitely.
Ask SkimNews
