Niv-AI Emerges to Enhance GPU Power and Performance

Electricity is a fundamental resource for artificial intelligence, but emerging computational methods are surpassing the capacity of data center operators to effectively manage their grid interactions, compelling them to reduce operations by up to 30%.

"A tremendous amount of energy is wasted in these AI facilities," stated Nvidia CEO Jensen Huang during his keynote at the company's annual GTC conference. "Every watt not utilized represents lost revenue," the company emphasized in its annual presentation.

Today, the startup Niv-AI has launched with $12 million in seed funding to address this issue. The company aims to precisely measure GPU power consumption using novel sensors and create tools for more efficient energy management.

Founded last year in Tel Aviv by CEO Tomer Timor and CTO Edward Kizis, the startup is supported by Glilot Capital, Grove Ventures, Arc VC, Encoded VC, Leap Forward, and Aurora Capital Partners. The company chose not to disclose its valuation.

As leading AI labs coordinate thousands of GPUs to train and run sophisticated models, rapid, millisecond-scale power spikes occur when processors alternate between computational tasks and inter-GPU communication.

These surges complicate data centers' ability to regulate their power draw from the electrical grid. To prevent shortages, data centers either invest in temporary energy storage to buffer these spikes or curtail GPU usage. Both approaches diminish the return on investment for costly hardware.

"We simply cannot keep constructing data centers using current methods," remarked Lior Handlesman, a partner at Grove Ventures and a member of Niv's board.

The initial phase of Niv's strategy involves gaining detailed insight. The company is currently installing rack-level sensors that monitor GPU power usage at millisecond intervals on its own hardware and with design partners. The objective is to analyze the distinct power patterns of various deep learning workloads and develop methods to help data centers better utilize their existing capacity.

The engineering team plans to build an AI model using the collected data, aiming to train it to forecast and coordinate power loads throughout the data center—effectively creating a "copilot" for facility engineers.

Niv anticipates deploying a functional system in several U.S. data centers within the next six to eight months. This solution is particularly appealing as major cloud providers encounter challenges related to land use and supply chain disruptions when building new facilities. The founders envision their final product as a crucial "intelligence layer" bridging data centers and the power grid.

"The grid is genuinely concerned about data centers drawing excessive power at peak times," Timor explained to TechCrunch. "The issue we're tackling has two interconnected aspects. First, we aim to help data centers utilize more of their GPUs, maximizing the power for which they are already contracted. Second, we can foster more responsible power consumption profiles between data centers and the grid."