Emerald AI Offers Grid-Friendly Solution to AI's Surging Energy Demands

The energy crisis lurking beneath the artificial intelligence boom came into sharp focus this week when President Trump and his cabinet gathered at Carnegie Mellon University in Pittsburgh to discuss a problem that has begun to dominate conversations in boardrooms and government offices alike: data centers are hungry, and the grid may not be able to feed them.

A Department of Energy report released last December laid out the scale of the challenge. Electricity consumption by data centers across the United States could nearly triple within three years, potentially consuming as much as 12 percent of the nation's total electricity output by 2028. That's not a marginal increase. That's a fundamental shift in how power flows through the American grid. At the summit, Senator Dave McCormick of Pennsylvania, who organized the event, framed the race to develop AI as essential to economic survival and national security. "This is a competition we must win," he said. Jonathan Gray, the CEO of Blackstone Group, was more blunt: energy has become the limiting factor. Without solving the energy problem, he argued, the entire AI enterprise stalls.

Tech companies are already scrambling. They're pouring billions into securing grid connections, building new power plants, and in some cases locking down their own energy supplies. But the challenge extends far beyond simple supply. Utility companies and grid managers face a vexing problem: they must plan for peak demand—those moments when electricity use spikes across an entire region—and build generation capacity to match those peaks. Often, the most expensive and most polluting sources in a regional power fleet are "peaker plants," facilities that fire up only during moments of maximum strain. They're dirty, they're costly, and they exist largely because demand is unpredictable. Meanwhile, the AI boom is also pushing major tech companies off track for their net-zero climate commitments, as greenhouse gas emissions from data centers climb.

Into this fraught landscape steps Emerald AI, a startup with a deceptively simple idea: what if data centers could be flexible? What if they could adjust their power consumption on demand without compromising the AI systems running inside them? The company's founder and CEO, Varun Sivaram, explained the concept to Newsweek: "Our goal is to make these data centers flexible in their power consumption." Emerald's platform allows grid managers to remotely shift power demand for data centers without affecting AI performance. The technology applies a principle known as demand-side management—the idea that if you can anticipate and control electricity demand, you can avoid building expensive peak-capacity plants and reduce blackout risk.

The company recently tested the approach in Phoenix, a city with a high concentration of data centers and an acute need for air conditioning during summer heat. On a day when temperatures reached 96 degrees and residents were running their air conditioning at full blast, Emerald's platform reduced power consumption by AI chip clusters by 25 percent during the peak demand window. The company held that reduced consumption level for three hours while the broader power demand peaked and then subsided. The AI systems continued to function normally. The grid got relief.

Sivaram sees the potential to invert the entire narrative around AI and energy. Instead of data centers being viewed as threats to grid stability and community welfare—raising electricity rates, risking blackouts, forcing reliance on dirty diesel generators—they could become what he calls "grid allies." Communities worried about rising power bills and grid crashes could instead see data centers as flexible loads that help stabilize the system during moments of stress. The company's $24.5 million funding round, announced this month and backed by Nvidia, former climate envoy John Kerry, and venture capitalist John Doerr, suggests that major players in technology and energy believe the concept has legs. Whether it can scale to meet the tsunami of demand heading toward the grid remains the open question.