Data center operators are discovering a new way to solve the energy crunch: instead of just consuming more power, they're learning to flex their demand like a utility company would. Emerald AI, a startup backed by Nvidia, is raising approximately $100 million to scale its grid-orchestration software, which allows data centers to shift their computing workloads in real time based on what the power grid needs.

This funding round comes just three months after the company closed a $25 million seed extension, bringing its total raised to $67.5 million. The new capital will support deployment of Emerald Conductor, the company's flagship product that acts as a mediator between the electrical grid and data center operations.

Why Does This Matter for Data Center Power Consumption?

The problem is straightforward: AI training and inference require enormous amounts of electricity, and data centers typically consume power on a fixed schedule. But power grids don't operate on fixed schedules. They experience peaks and valleys throughout the day, and when demand spikes unexpectedly, grid operators scramble to keep the lights on. Emerald's software flips this dynamic by allowing data centers to become flexible consumers, ramping down non-critical workloads when the grid is stressed and ramping back up when power is abundant.

The company has already proven the concept works. It completed demonstration projects in Phoenix and Chicago, and partnered with the UK's National Grid on a third trial that simulated over 200 real-time grid events to test the Conductor's ability to dynamically adjust power consumption. A fourth demonstration launched in April with Silicon Valley Power in Santa Clara, California.

Emerald announced its first commercial deployment in October, with the orchestration software set to run at Nvidia's 96-megawatt Aurora data center under construction in Manassas, Virginia.

How Does Grid-Aware Data Center Management Work?

• Real-Time Workload Shifting: The Emerald Conductor monitors grid conditions and automatically adjusts which AI workloads run at any given moment, prioritizing critical tasks during peak demand periods and deferring less urgent compute during grid stress.
• Maintaining AI Performance: The system ensures that dynamic power adjustments don't degrade the quality or speed of AI model training and inference, keeping compute performance within acceptable ranges even as electricity consumption fluctuates.
• Grid Stability Support: By becoming a flexible load, data centers can help stabilize the grid during periods of high demand or renewable energy variability, reducing the need for expensive peaker plants or emergency power sources.

The broader context here is that AI's energy hunger is forcing a reckoning across the industry. Nvidia's vision, outlined at its recent GTC Taipei conference, frames the competition between AI factories not just in terms of raw computing power, but in terms of "tokens per watt," meaning how much useful AI output you can generate from each unit of electricity consumed.

To that end, Nvidia has been investing aggressively in the energy layer of what it calls its "five-layer cake" technology ecosystem. Beyond backing Emerald AI, Nvidia's venture capital arm has invested in TerraPower (a nuclear energy company), Commonwealth Fusion Systems, and other firms focused on optimizing data center power consumption and grid management.

What Does This Mean for the Future of AI Infrastructure?

Emerald AI's funding and commercial deployment signal a shift in how the industry thinks about data center design. Rather than treating power as an unlimited resource to be secured through long-term contracts or on-site generation, operators are now treating it as a dynamic asset to be managed intelligently. This approach could help data centers coexist with residential and commercial power users without destabilizing the grid.

The company is backed not only by Nvidia but also by Kate Brandt, Google's chief sustainability officer, suggesting that major cloud providers see grid orchestration as a critical piece of sustainable AI infrastructure.

As AI workloads continue to grow, the ability to run them flexibly and responsively to grid conditions may become as important as the chips themselves. Emerald's $100 million raise reflects investor confidence that this software layer will be essential infrastructure for the next generation of AI data centers.