LinkedIn plans to keep AI data center spending flat this year

In short: LinkedIn says it will hold its spending on AI computing steady this fiscal year by getting more work done with the same hardware.

What's going on

Many large tech companies are spending heavily to build or rent more data centers for AI. Data centers are large buildings full of computers, like warehouses packed with servers. Those servers often use GPUs, which are special chips that help run AI quickly.

LinkedIn told Wired it plans to keep its GPU investment steady, and keep its overall computing and storage capacity roughly flat for the fiscal year that started last month and ends next June. Executives said the company has improved how it uses its existing GPUs, including keeping them busy more of the time. LinkedIn says it can now run training work at over 95 percent GPU use, meaning the chips sit idle less often.

The company also says it has made AI models cheaper to run by using techniques like “distillation” (teaching a smaller model to copy what a bigger model learned, like making a short, useful study guide from a long textbook). LinkedIn estimates its efficiency work saved about $24 million over the past 12 months, which it compares to about 1,100 GPUs running nonstop for a year. It also shifted some tasks from GPUs to CPUs, which are more common chips that are often cheaper and use less electricity.

What to watch

LinkedIn leaders say this plan could be tested if AI features need more computing than expected, or if hardware prices keep rising, including memory chips. The company is not ruling out future data center growth, but it says it will plan quarter by quarter instead of guessing far ahead.

Source: Wired