Supermicro GPUs Slash AI Power Use 75% in 20ms Demand-Response Test
Supermicro GPU clusters responded to CAISO signals in under 20 milliseconds in a joint test with CPower and Bentaus, cutting AI electricity use by up to 75%. It sets up Supermicro’s NVIDIA B200 GPUs for demand-response participation as AI power needs grow tenfold to over 50GW by 2030.
1. Joint Test Execution
CPower, Bentaus and Supermicro conducted an end-to-end demonstration dispatching real-time CAISO wholesale electricity market signals through the Bentaus orchestration platform to a Supermicro-managed GPU cluster. The test validated seamless integration between energy market signals and server-level compute orchestration.
2. Performance Metrics
The Supermicro server cluster equipped with NVIDIA B200 GPUs responded to a full dispatch signal cycle in under 20 milliseconds and dynamically curtailed AI workloads, achieving up to 75% electricity reduction while maintaining service-level agreements under full IT load.
3. Market Implications
This foundational architecture enables Supermicro hardware to participate in PJM, ERCOT and SPP demand-response programs, providing grid flexibility as U.S. AI compute capacity scales from 5GW to over 50GW by 2030 and positioning data centers as active market participants.