Google, Nvidia and Anthropic back Emerald AI coalition to unlock 100 GW of grid capacity for data centers

A newly formed coalition featuring NVIDIA, Google and Anthropic, alongside grid-software startup Emerald AI, is aiming to free up 100 gigawatts of electrical capacity on the US power grid to accommodate the fast-growing footprint of AI data centers.

The initiative addresses one of the biggest bottlenecks in the AI buildout: power. Utilities and grid operators have struggled to interconnect large new loads, with queues for grid access stretching years in many markets. The coalition’s premise is that better coordination of data center energy use — particularly flexible scheduling of training and inference workloads — could allow far more computing capacity to be added without equivalent new generation being built.

Emerald AI, which develops technology for managing the power profiles of AI infrastructure, anchors the effort. Backing from Google, Anthropic and NVIDIA — three of the most prominent names in the AI ecosystem — signals that hyperscalers, model developers and chipmakers all see grid access as a shared constraint on scaling.

NVIDIA, which describes itself as a data-center-scale AI infrastructure company, has increasingly been drawn into power and energy conversations as its accelerators have become the dominant hardware in AI clusters. The company’s stock traded at $223.67 on the day of the announcement, down 0.73% from the prior close of $225.31, valuing the company at roughly $5.56 trillion.

The 100 GW target is significant in scale. For comparison, that figure represents a multiple of the capacity typically required by a single large data center campus, and utilities have warned that unconstrained AI demand could outpace grid planning. Coalition members argue that flexible, demand-responsive data centers — ones that can shift or throttle compute loads in response to grid conditions — can relieve congestion and speed up interconnection approvals.

For grid operators, the pitch is that data centers could act as a controllable load rather than an inflexible one, reducing the need for expensive peaker plants and grid upgrades. For the AI industry, unlocking stranded grid headroom could shorten timelines for new facilities, which today are often gated more by power availability than by access to chips or capital.

The announcement comes amid intensifying competition among hyperscalers to secure land, transformers, turbines and interconnection agreements for AI capacity, and follows a wave of power-related partnerships across the sector, including nuclear, gas and renewable deals aimed at serving data center demand.

What to watch

• Details on how Emerald AI’s demand-flexibility technology will be deployed and measured by participating utilities
• Whether additional hyperscalers, chipmakers or grid operators join the coalition
• Regulatory and utility responses to flexible-load proposals in grid interconnection queues
• Upcoming earnings reports from NVIDIA and Google parent Alphabet, which may address data center power constraints and capacity guidance

Source: original release