Google Nvidia and Anthropic back Emerald AI grid coalition

Google, Nvidia, and Anthropic have joined a new coalition built around a grid-software startup called Emerald AI, aiming to make “demand response” a standard part of data-centre expansion. According to TechCrunch, the group—called the AI Energy Management Alliance—argues that data centres could pause noncritical tasks and shift some compute loads to free up enough capacity for roughly 100 gigawatts of additional data centres to connect to the electrical grid.

Demand response is not new: utilities have long paid large users to cut consumption during peak periods. What is new is the target and the marketing. Data centres have typically participated by running backup diesel generators, effectively moving the load off-grid rather than reducing it. Emerald AI is pitching software that connects utilities directly to data centres so they can ramp down quickly, pause workloads, or move compute to other sites with spare capacity, making the facilities behave more like grid resources. TechCrunch notes that utilities including AES, Constellation, National Grid, and NRG Energy are involved, and cites a Goldman Sachs study suggesting that shaving peak usage for a few hours could unlock large amounts of capacity.

The coalition’s formation reflects a constraint that is now shaping the AI industry as much as chips do: interconnection queues, local transformer limits, and the politics of siting new generation. For hyperscalers and AI labs, demand response offers a way to claim flexibility without waiting for new power plants or new transmission lines. For utilities, it offers a controllable lever—if the software works as advertised—at a time when they are being asked to connect large loads faster than infrastructure can be built. Emerald AI’s chief scientist, Ayse Coskun, told TechCrunch the approach can reduce the need for new generation but will not eliminate it.

But the economics cut both ways. A data centre that can be interrupted becomes a negotiable load, which can speed approvals and lower connection costs—while shifting operational risk onto customers whose workloads are deemed “noncritical.” The coalition is also trying to normalise the idea that compute can be moved around the map to chase available electrons, a logic that favours the largest operators with multiple sites and the software to orchestrate them. Smaller developers, and communities hosting the infrastructure, still face the same physical footprint: substations, cooling, backup systems, and the water and land-use conflicts that come with them.

Emerald AI has raised a large funding round, TechCrunch reports, and the alliance is already selling a future where grid constraints are handled by coordination rather than construction. The test will come the first time a utility calls for a cut and the “noncritical” tasks turn out to belong to someone who expected the lights to stay on.