LinkedIn Freezes Data Center Spending as AI Efficiency Gains Offset Rising Costs

LinkedIn is charting an unusual course in the race for AI infrastructure. While rivals like OpenAI, Meta, and Google pour billions into new data centers, the professional network has committed to keeping its GPU investments and compute footprint essentially flat over the next fiscal year—a move executives describe as bold given today's hardware-hungry landscape.

The decision hinges on a striking efficiency milestone: over the past six months, LinkedIn has found ways to do twice as much work with its existing graphics processors. That breakthrough allowed the company to avoid major capital outlays despite surging prices for computing components, even as it prepares to ship new generative AI features.

Doubling Down on Efficiency#

LinkedIn's engineering teams pursued optimization at every stage of the AI pipeline. Raghu Hiremagalur, the company's chief technology officer for infrastructure, noted that the platform initially faced an untenable trajectory: each query was costing more, and data storage was doubling annually. Something had to change.

The company deployed multiple tactics:

• Model distillation: Trained smaller, cheaper models from larger ones. A single compact model now handles job recommendations by learning from two bigger models, maintaining quality while cutting operational costs.
• GPU utilization: Achieved north of 95 percent usage efficiency through better workload allocation and measurement tools that track how much compute individual teams consume.
• Processor optimization: Rewrote foundational software on Nvidia processors to handle larger tasks, and shifted some work from expensive GPUs to cheaper CPUs.
• Inference improvements: Streamlined how the recommender system selects posts for newsfeeds by reusing earlier ranking signals and balancing loads between processors.

In total, LinkedIn estimates these efforts have saved roughly $24 million over twelve months—the equivalent of keeping 1,100 GPUs running continuously for a year. Erran Berger, the company's chief technology officer for engineering, acknowledged the sum is modest relative to LinkedIn's $18 billion in annual revenue, but emphasized that the resulting agility and engineering discipline matter more than the raw savings.

A Counterweight to the Spending Boom#

LinkedIn's restraint stands in contrast to industry-wide fervor. Competitors are scrounging up all the money they can find and to build and operate sprawling AI centers. Labor and parts shortages have delayed projects, and some platforms have had to .