LinkedIn Will Not Expand Data Centers in the Next Year
Microsoft
NVIDIA
LinkedIn has decided not to expand its AI data centers in the current fiscal year, keeping GPU investment and compute/storage footprint flat. The company achieved efficiency gains by using existing GPUs twice as effectively, saving about $24 million. This contrasts with the aggressive spending by other tech giants.
LinkedIn, owned by Microsoft, has decided to keep its data center investment flat for the fiscal year ending June 2025, avoiding the aggressive expansion seen at OpenAI, Meta, and Google. The company achieved efficiency gains by using existing GPUs twice as effectively over the past six months, saving approximately $24 million in the last 12 months. LinkedIn's CTOs Erran Berger and Raghu Hiremagalur stated that the company optimized GPU utilization to over 95%, used techniques like distillation to train smaller models, and reworked software to run tasks on CPUs instead of GPUs. The company operates its own data centers in Oregon, Texas, and Virginia, which gives it control over infrastructure. Despite flat spending, LinkedIn plans to purchase new servers to replace aging hardware. The efficiency drive is part of a broader trend called 'tokenomics' where enterprises analyze the cost of generative AI tools. However, Gartner analyst Chirag Dekate warns that too many financial constraints could compromise AI ambitions.
- Сокращения
- GPU = Graphics Processing Unit — графический процессор
- CPU = Central Processing Unit — центральный процессор
Source: Wired AI —
original
