LinkedIn замораживает вычислительные мощности: год без расширения дата-центров

LinkedIn will not expand its data centers over the next year and will keep its computing power at the current level — despite the artificial intelligence boom. Instead of buying new hardware, the company is requiring engineers to make more efficient use of every GPU it already has. This is reported by Wired.

What exactly LinkedIn decided

LinkedIn has frozen its computing power at its current level: over the next 12 months, the company does not plan to expand its data centers or grow its fleet of accelerators. Infrastructure growth is being paused at the very moment when almost the entire industry is accelerating it.

The decision was made not because of a drop in demand, but as a deliberate strategy. According to Wired, instead of capital spending on new servers, LinkedIn is betting on efficiency — pushing teams to squeeze the maximum out of the GPUs already installed.

• Computing power is being frozen for roughly a year — with no data center expansion
• The GPU fleet stays at its current level, with no major new purchases planned
• The reason is not a slowdown, but a deliberate bet on efficiency despite the AI boom
• LinkedIn is owned by Microsoft — one of the largest investors in AI infrastructure
• The data source is a Wired report

Why the company isn't adding more hardware

LinkedIn is choosing efficiency over capital spending: instead of buying new accelerators for its growing AI features, the company is optimizing the use of the ones already sitting in its data centers. The logic is simple — every GPU must deliver maximum value before buying the next one is justified. The industry calls this mode "doing more with less," and for AI teams it means competing for a limited, not expanding, pool of compute.

This approach runs against the broader trend. While the biggest tech companies and cloud providers are pouring record sums into data centers and AI infrastructure, LinkedIn is deliberately holding its compute spending steady. For a platform that is embedding generative AI into its feed, search, and recommendations, this is a non-trivial choice: demand for compute is rising, but capacity is not.

The contrast is sharpened by the parent company. LinkedIn has been owned by Microsoft since 2016, and Microsoft, by contrast, has announced roughly $80 billion in capital investment in AI data centers for fiscal year 2025. Against this backdrop, the subsidiary platform's decision to keep its own capacity unchanged looks like a deliberate counter-move: not every node in the ecosystem is obligated to join the hardware race.

"LinkedIn is holding the line on compute spending and challenging engineers to make every GPU work at full capacity," is how

Wired sums up the essence of the strategy.

What changes for engineers

The main burden falls on engineering teams: since there will be no new GPUs, all growth in AI features will have to come from optimization. In practice, that means denser utilization of accelerators, fine-tuning models to fit the available hardware, and strict prioritization of tasks that actually deliver value.

This isn't about secondary features: LinkedIn is embedding generative AI into recruiter assistants, post-writing suggestions, and feed personalization for more than 1 billion registered users. Serving AI demand of that scale on a fixed GPU fleet is exactly the task engineers have been set.

In effect, LinkedIn is turning a constraint into an engineering challenge. Instead of the usual "let's buy more capacity," teams now have to find where compute is being wasted and rewrite pipelines so the same result can be achieved with fewer GPU-hours.

What this means

LinkedIn's strategy is a signal that even in the middle of the AI race, major players are willing to bet on efficiency, not just scale. If the "fixed capacity plus optimization" model works, it will become a powerful argument against the endless race of capital spending on data centers — and an example for other companies whose budgets don't allow for unlimited GPU purchases.