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

Amidst the escalating demands of artificial intelligence, efficient electricity usage has become critical for data centers. With new processing methods outpacing the capabilities of operators to balance their energy consumption with the power grid, many facilities are forced to reduce their GPU performance by as much as 30%. Nvidia's CEO, Jensen Huang, highlighted this crisis during a keynote at the annual GTC conference, stating, "Every unused watt is revenue lost." In response to this pressing issue, Niv-AI, a Tel Aviv-based start-up, has officially launched after securing $12 million in seed funding. Founded by CEO Tomer Timor and CTO Edward Kizis, the company aims to revolutionize power management in AI-driven environments. With backing from notable investors such as Glilot Capital and Grove Ventures, Niv-AI is poised to bring significant advancements to the industry. As data centers increasingly rely on thousands of GPUs to run complex AI models, they encounter frequent surges in power demand. These surges occur as GPUs transition between computation tasks and inter-processor communication, complicating energy management from the grid. To mitigate this, data centers often invest in temporary energy storage or limit GPU usage, which ultimately undermines the return on investment for costly hardware. Lior Handlesman, a partner at Grove Ventures and board member of Niv-AI, remarked, “We just can’t continue building data centers the way we build them now.” Niv-AI's initial focus is on gaining insights into power usage. The company is deploying advanced rack-level sensors capable of detecting GPU power consumption at millisecond intervals, in collaboration with design partners. Their objective is to analyze the distinct power profiles associated with various deep learning tasks and devise strategies to optimize the capacity of data centers. Leveraging the collected data, Niv-AI plans to develop an AI model designed to predict and manage power loads effectively, serving as a “copilot” for data center engineers. In the next six to eight months, Niv-AI anticipates having operational systems in several U.S. data centers. This initiative comes at a crucial time as hyperscale operators face challenges related to land use and supply chain management in constructing new facilities. The founders envision their product as an essential “intelligence layer” that bridges the gap between data centers and the power grid, addressing concerns about excessive energy consumption during peak times. Timor explained the dual challenge they aim to tackle: enabling data centers to utilize more GPUs while fostering more responsible energy profiles in their interactions with the grid.