Google, Nvidia and Anthropic back Emerald AI grid alliance to unlock data center capacity

A new coalition for grid-constrained AI

Emerald AI, a grid software unicorn, has formed the AI Energy Management Alliance (AEMA) together with Google and Nvidia, according to TechCrunch. Anthropic has joined the founding trio, along with utilities including AES, Constellation, National Grid and NRG Energy. The group's aim is to make demand response — temporarily reducing a facility's electricity draw — a standard part of how data centers are planned and operated.

AEMA argues that suspending work that is not time-sensitive and moving compute to sites where the grid has spare room could open capacity for another 100 gigawatts of data centers on existing infrastructure.

Demand response, an old idea repurposed

Demand response is a practice utilities have used for decades. Grids are built to handle peak loads, which means that for most of the year demand sits well below what the network can actually carry. Utilities have traditionally paid large consumers, such as factories, to cut usage during peaks, usually by halting production or switching to backup power, and often at generous rates.

Data centers already take part in such programs, though TechCrunch notes they typically do so by firing up backup generators. That diesel-burning workaround is what Emerald AI wants to replace. Its software sits between utilities and data centers, coordinating grid requests so a facility can pause noncritical workloads or shift jobs to other data centers in regions with available headroom.

A competitive corner of energy tech

Emerald AI is not alone in the space. Google has been developing its own tooling for flexible data center power use, and Enel X lets facilities lean on their uninterruptible power supplies to shave peaks in demand. A Goldman Sachs study published last year estimated that capping a data center's draw at 90 percent of its maximum for a few hours at a time could release 76 gigawatts of capacity.

Emerald AI's edge, per TechCrunch, is that its software links utilities directly to data centers. That direct line should let facilities respond to grid signals within seconds, behaving more like batteries than the slow-cycling industrial loads demand response programs were originally designed around.

The company also has fresh capital behind it: Emerald AI recently raised $150 million in a Series A led by Energize Capital and DCVC, funding that could support a wider rollout. Beyond demand response itself, AEMA says it will help tech companies and utilities identify new sites for data centers, a search that has proven difficult for both sides.

Why it matters

AI workloads are both power-hungry and unusually flexible. Training runs and batch inference can often be paused or relocated for a few hours without users noticing, which makes data centers a natural fit for demand response — arguably a better fit than the factories the model was invented for. If flexible consumption becomes a first-class design principle for new facilities, grids could absorb substantially more AI compute without waiting years for new power plants and transmission lines to come online.

The approach is not a cure-all, though. Ayse Coskun, Emerald AI's chief scientist, told TechCrunch that the company's technology can reduce the industry's appetite for new generating capacity without removing it entirely. Demand response redistributes load in time and space; it does not create energy. How much of the promised 100 gigawatts actually materializes will depend on utilities, regulators and hyperscalers changing long-established habits, and on whether shifting workloads across regions proves as painless in practice as it sounds on paper.