Niv-AI Optimizes GPU Energy with $12 Million

Niv-AI: An Innovative Response to GPU Energy Challenges

In a context where electricity is crucial for artificial intelligence, data centers struggle to manage their energy consumption effectively. These facilities are sometimes forced to reduce their consumption by up to 30% to adapt to the limits of the electrical grid.

At Nvidia's annual GTC conference, CEO Jensen Huang emphasized the importance of minimizing energy waste in data centers, stating that every unused watt represents a loss of revenue.

Promising Funding for Niv-AI

The Tel Aviv-based startup Niv-AI, founded last year, has recently emerged with an initial funding of $12 million. Founded by Tomer Timor and Edward Kizis, the company is backed by several investors, including Glilot Capital, Grove Ventures, Arc VC, Encoded VC, Leap Forward, and Aurora Capital Partners. While the startup's valuation remains confidential, its goal is clear: to optimize GPU energy usage through new sensors and management tools.

Technical Solutions for Optimized Energy Management

Modern data centers utilize thousands of GPUs to train and execute advanced models, generating spikes in energy demand at the millisecond level when processors switch from computation tasks to communication with other GPUs. These fluctuations complicate energy management, sometimes forcing centers to invest in temporary energy storage or reduce GPU usage, which impacts return on investment.

Lior Handlesman, a partner at Grove Ventures, stated that the current construction of data centers is not sustainable in the long term. Niv-AI is working to address this issue by deploying rack-level sensors capable of measuring GPU energy usage with millisecond precision. The goal is to free up a larger portion of the existing capacity of data centers.

Towards a Predictive and Intelligent System

Niv-AI's roadmap begins with a deep understanding of the energy consumption profiles of deep learning tasks. By collecting this data, the company plans to develop an AI model capable of predicting and synchronizing electrical loads, acting as a "co-pilot" for data center engineers.

Niv-AI hopes to have an operational system in several U.S. data centers within six to eight months. This initiative is particularly relevant as hyperscalers face challenges related to land use and supply chains.

Tomer Timor, CEO of Niv-AI, explained that their solution aims to help data centers maximize GPU usage while establishing more responsible energy consumption profiles concerning the electrical grid. The founders envision their ultimate product as a layer of "intelligence" missing between data centers and the electrical grid. The grid is indeed concerned that data centers may consume too much energy at any given time, which could lead to disruptions.