Case Study: Argonne National Laboratory achieves energy-efficient, low-latency agentic AI inference with SambaNova SambaStack

Argonne National Laboratory powers 1,000 scientists with SambaNova and 200B tokens per month

Argonne National Laboratory, a U.S. Department of Energy research center, faced the challenge of delivering high-performance, low-latency AI inference for its agentic coding and research workflows at a massive scale. They needed an energy-efficient and cost-effective solution to serve roughly 1,000 scientists consuming 200 billion tokens per month, a figure expected to double, without relying on their 40 MW supercomputer for every task. They partnered with vendor SambaNova Systems to implement a solution.

SambaNova Systems implemented its SambaStack SN40 platform on-premises at Argonne. The solution provided energy-efficient inference, drawing an average of 20 kW or less, and delivered the required low-latency performance. The measurable results include successfully serving the scientists' agentic workflows and enabling the consumption of 200 billion tokens per month, with the infrastructure in place to support the expected doubling of that usage.

Argonne National Laboratory