Niv-AI exits stealth to wring more power performance out of GPUs

Niv-AI has emerged from stealth with 12 million dollars in seed funding to address power inefficiencies in artificial intelligence data centers. The Tel Aviv-based startup develops sensors and management tools aimed at measuring GPU power use at millisecond scales and reducing waste during surges that occur when processors switch between computation and communication tasks. The company was founded last year by chief executive Tomer Timor and chief technology officer Edward Kizis. It is backed by Glilot Capital, Grove Ventures, Arc VC, Encoded VC, Leap Forward, and Aurora. The startup declined to share its valuation. Data center operators currently face challenges managing power draw from the grid during these frequent surges, which can lead to throttling GPU usage by as much as 30 percent or reliance on temporary energy storage. Both approaches reduce returns on investments in chips, according to the company. Nvidia chief executive Jensen Huang said during a keynote at the company's GTC conference that power is squandered in AI factories and that every unused watt is revenue lost. Niv-AI is deploying rack-level sensors to detect power usage on GPUs it owns and with design partners. The goal is to map power profiles of deep learning tasks and develop mitigation techniques. The company plans to build an artificial intelligence model trained on collected data to predict and synchronize loads across data centers. Niv-AI expects to have an operational system in a handful of United States data centers within six to eight months. Timor said the effort aims to help data centers utilize more GPUs while creating more responsible power profiles between facilities and the grid.