The grid's AI problem

Data centers want immediate power on an expensive grid built to handle peak demand. Utilities and communities become "foot soldiers" of the AI buildout, bearing the infrastructure costs.

A new alliance launches

Emerald AI, Google, and NVIDIA founded the AI Energy Management Alliance with 18 members including Anthropic, National Grid, AES, and NRG. The idea: flexible data centers get faster power connections.

Flexibility saves money

Flexible data centers spare peak demand moments that trigger expensive grid upgrades. The Brattle Group estimates each 10% gain in grid utilization lowers rates by about 3.4%.

“each 10% gain in utilization lowers rates by about 3.4%”

The connection bottleneck

New U.S. data centers wait a decade or more for grid connections because utilities must reserve capacity for peak demand. The grid is only 50% utilized on average, leaving massive unused capacity.

Multiple paths to flexibility

Batteries, on-site generation, and software that pauses less urgent computing can all make data centers flexible. The alliance is technology-neutral and rewards any measurable grid relief.

Proof of concept exists

Google runs a nationwide demand-response portfolio of roughly one gigawatt. NVIDIA, Digital Realty, and Emerald AI will launch the world's first power-flexible AI factory in Virginia at nearly 100 megawatts later this year.

Regulators are moving

The Federal Energy Regulatory Commission directed grid operators to accommodate large customers willing to limit power draw for faster connections. Texas and California have already launched flexible-load programs.

The ask is simple

Governors and regulators should offer faster, larger grid connections to data centers that commit to flexibility. This unlocks more AI innovation, lower energy bills, and a more reliable grid.