LinkedIn has decided not to expand its AI data centers this fiscal year, maintaining a steady investment in GPUs and a flat compute and storage footprint. The company’s strategy aims to balance increased AI workloads with cost control, a bold move in an industry dominated by aggressive infrastructure spending. LinkedIn’s chief technology officer for engineering, Erran Berger, emphasized the challenge of managing compute demands without increasing budgets. "That’s a pretty bold statement to make in today's world," Berger said.
To achieve this, LinkedIn has focused on optimizing GPU usage and improving efficiency across its AI pipeline. The company’s infrastructure team developed tools to measure and allocate compute resources more effectively, achieving utilization rates of over 95 percent on training GPUs. LinkedIn also employed techniques like model distillation to train smaller, more cost-effective models without sacrificing quality. "People are finding and discovering jobs that they were not successfully finding before," Berger said, highlighting the benefits of these efficiency gains.
LinkedIn’s approach reflects a broader industry shift toward cost-conscious AI development. While major companies like OpenAI, Meta, and Google continue to invest heavily in data centers, LinkedIn’s strategy highlights the potential for innovation without massive infrastructure expansion. The company’s efficiency efforts have saved an estimated $24 million over the past year, though executives acknowledge the savings are modest against its $18 billion annual revenue. LinkedIn’s focus on optimization and agility is seen as a model for balancing AI ambitions with financial discipline.
Source: wired
