Emerald AI, Google and Nvidia launched the AI Energy Management Alliance on Sept. 16, 2026, to promote data centers that can adjust electricity use when power grids are under stress. Anthropic, utilities and power producers are also participating, putting major AI developers and energy companies behind a model that treats computing demand as flexible rather than constant.
The alliance’s commercial promise is faster access to scarce grid capacity. New data centers often face long interconnection queues because utilities must plan for their maximum demand. If facilities can verifiably reduce load during peak periods, grid operators may be able to connect more projects without building every upgrade first.
Demand response moves into AI infrastructure
Nvidia said the coalition will develop data centers that manage electricity consumption in response to grid conditions. The approach can include pausing noncritical computing jobs, shifting work to another location or using storage and on-site generation when power is constrained.
Utilities have used demand-response programs with factories and other large customers for decades. The new effort applies that operating model to AI infrastructure, where workloads can vary sharply and some tasks are more time-sensitive than others.
Alongside Google, Nvidia and Emerald AI, participating organizations include Anthropic, AES, Constellation, National Grid, NRG Energy and RWE. The group spans the companies building AI systems, the operators supplying electricity and the software vendors coordinating demand.
The 100-gigawatt estimate needs proof
The alliance says power flexibility could unlock as much as 100 gigawatts of additional data-center capacity on existing grids. That figure is an alliance estimate, not a guaranteed capacity increase. It depends on how much computing can be moved without disrupting customers, how utilities value the response and whether data centers perform reliably during grid events.
TechCrunch reported that Emerald AI’s software connects utilities directly with data centers so facilities can respond to requests quickly. Emerald AI Chief Scientist Ayse Coskun also cautioned that flexibility could reduce the need for new generation but would not eliminate it.
Power becomes part of the computing contract
For data-center developers, flexible-load agreements could turn operating discipline into a site-selection advantage. Projects able to demonstrate controllable demand may compete more effectively for grid connections than facilities built around an uninterrupted maximum draw.
For AI customers, the tradeoff will appear in service levels. Providers must decide which training, inference and maintenance tasks can move, how delays are priced and what happens when both the grid and a customer need capacity at the same time.
The alliance does not resolve the AI industry’s power shortage. It does, however, formalize a shift in infrastructure strategy: the next unit of computing capacity may come not only from building more generation, but from making data centers respond more intelligently to the grid they already use.