Stealth startup Niv-AI has raised $12 million to address a critical bottleneck in artificial intelligence infrastructure: inefficient power usage in data centers. As highlighted by Jensen Huang of Nvidia, AI operations waste significant energy, with GPU utilization often reduced by up to 30% due to unstable power demands, translating into lost revenue and underused capacity.

The issue stems from rapid, unpredictable power spikes generated by large-scale AI workloads, making grid stability increasingly difficult to maintain.

To manage these fluctuations, data centers currently rely on costly energy buffering systems or throttle GPU performance, both of which undermine efficiency and profitability.

Niv-AI aims to solve this by deploying millisecond-level, rack-based sensors to track GPU power consumption in real time, enabling precise mapping of energy usage across different AI tasks.

Backed by multiple venture firms and led by CEO Tomer Timor, the company is building tools to unlock unused compute capacity without compromising grid stability.

Looking ahead, Niv-AI plans to develop an AI-driven “copilot” that predicts and synchronizes power demand across entire data centers, effectively acting as an intelligence layer between infrastructure and the electrical grid.

This approach promises to maximize GPU utilization while creating more stable and efficient energy profiles, a crucial advancement as hyperscalers face growing constraints in scaling data center capacity.

The company expects to begin deployments in select U.S. facilities within the next six to eight months, positioning itself at the intersection of AI growth and energy optimization.