Emerald AI, Google, and NVIDIA launched the AI Energy Management Alliance (AEMA) on September 16, 2026, a first-of-its-kind coalition aimed at turning AI data centers into flexible, grid-responsive resources rather than rigid power consumers. The group seeks to accelerate interconnection of new AI infrastructure while strengthening grid reliability and protecting energy affordability.
Core goal: unlock capacity on the existing grid
U.S. power infrastructure was largely designed for relatively flat, predictable electricity demand. AI data centers, by contrast, can modulate their power draw. AEMA argues that if these facilities can temporarily reduce consumption during the grid's highest-stress hours — by shifting non-critical compute workloads, discharging batteries, using on-site generation, or responding to contingency signals — the existing grid could support significantly more capacity.
Emerald AI CEO Varun Sivaram has estimated that flexibility during the worst hours of the year could unlock as much as 100 GW of additional data-center capacity on the current U.S. grid. Google has already demonstrated more than 1 GW of demand-response capability across its cloud facilities. Multiple studies suggest that limiting net grid withdrawal for fewer than 100 hours per year can free dozens of gigawatts without requiring multi-year transmission upgrades.
Alliance principles and approach
AEMA is developing technical and operational standards so that flexible data centers become a "controllable resource rather than an inflexible load." Key principles include defining clear ride-through, curtailment, and contingency-response obligations before a facility connects. The alliance aims to give system operators greater visibility and control while reducing uncertainty for developers.
In return for committing to flexibility, the group wants utilities and grid operators to grant faster interconnections and, in some cases, share infrastructure costs.
Members and early projects
Founding members are Emerald AI, Google, and NVIDIA. Launch partners include Anthropic, major utilities and energy companies (AES, Constellation, National Grid, NRG Energy, RWE), and technology firms across the power and AI value chains.
NVIDIA and Emerald AI are already collaborating on what they describe as the world's first power-flexible AI factory — a 100 MW data center in Manassas, Virginia, expected online before the end of 2026. Emerald AI's Conductor software platform coordinates utility signals with data-center loads so facilities can respond quickly, similar to how batteries operate. NVIDIA offers tools such as DSX Flex that optimize grid usage, on-site generation, and storage for facilities using its GPUs.
Why it matters
Explosive demand for AI compute is colliding with constrained transmission and generation capacity, leading to multi-year interconnection queues in many regions. By making data centers grid-flexible, AEMA seeks to turn a major source of new load into a tool that can actually enhance system utilization and reliability. If successful, the approach could speed AI-infrastructure deployment, reduce pressure for costly grid upgrades, and help keep electricity affordable for other customers.
The alliance will focus on technical standards, utility collaboration on interconnection solutions, and policy advocacy that recognizes the value of grid-responsive demand.