Google, Nvidia, and Anthropic want Emerald AI to find space on the grid for more data centers

Tim De Chant at TechCrunch reports grid-software unicorn Emerald AI forming the AI Energy Management Alliance with Google and Nvidia to use the startup’s tech to secure grid space for more data centers. Anthropic joins the founding set, alongside utilities including AES, Constellation, National Grid, and NRG Energy.

AEMA’s pitch: make demand response—temporarily cutting or shifting electricity use—central to how data centers get built and connected. Pausing noncritical tasks and moving some compute, the alliance says, could allow an additional 100 gigawatts of data centers onto the grid.

Emerald AI’s software links utilities directly to data centers so they can respond fast—more like batteries than factories dialing down a shift. The company recently raised a $150 million Series A led by Energize Capital and DCVC.

We’re for anything that gets useful compute online without pretending the grid is infinite.

Data-center interconnection is one of the binding constraints on AI right now—sites, substations, transmission queues, and peak capacity. Demand response is old utility practice; applying it to AI loads that can ramp quickly is overdue. If Emerald’s path lets operators shed noncritical jobs or shift work to campuses with headroom instead of firing diesel backup, that’s a cleaner answer than the status quo. A Goldman Sachs study cited in the piece estimated that limiting max grid usage to 90% for a few hours could free 76 gigawatts—AEMA’s 100 GW claim sits in the same ambition band.

Google already builds its own tools; Enel X taps UPS gear for peaks. Emerald’s edge, as framed here, is the direct utility-to-data-center pipe and the coalition wrapping hyperscalers, a chip vendor, a frontier lab, and major utilities into one forum. That’s how you turn a clever software idea into an interconnection norm.

Eyes open on the limits. Emerald AI chief scientist Ayse Coskun told TechCrunch the tech can blunt the need for new generation—it won’t eliminate it. Demand response is not a substitute for transmission buildout or firm power. And “pause noncritical tasks” only works if training and inference schedulers actually honor utility signals when the money tradeoff hurts. I’m watching whether AEMA produces measurable megawatts connected—or just another alliance logo page.

Useful AI needs power. The harm if coalitions only optimize for more campuses: communities still eat the peaks, rates, and land use while the 100 GW number does press-tour duty. The upside if it works: more capacity for models that help science and industry, with less diesel and less pure wait-for-a-new-plant paralysis.

Traditional demand response paid large users—factories and the like—to cut load or switch to backup generation when the grid was stressed. Data centers already use those programs, often via diesel. Emerald’s alternative coordinates utility requests with workload pauses and cross-site load shifts.