The unexpected, accelerated growth in AI data center demand sparked a gold rush for grid connections among data center operators. But they quickly ran into a major roadblock: There isn’t enough power to meet all connection requests, leading to lengthy queues, with some waitlists stretching up to 10 years or more. In addition, large load requests—typically 20 MW and above—require comprehensive grid impact studies to determine whether the transmission and distribution system can reliably support the new demand. The legacy connection process wasn’t designed for the massive increase in requests, and completing the grid impact designs required for these requests takes time.
This AI explosion turned data center development on its head. Operators used to build capacity and then try to fill it; now they scramble to meet client demand. They are facing strong headwinds when trying to access power. In the U.S. and Europe, some demand queues require as much capacity as the peak demand of the entire system. In places like the United Kingdom and Texas, queues exceed peak demand. The International Energy Agency predicts global electricity consumption for data centers could double between now to 2030.
Interconnect overreach
Until recently, data center operators were exacerbating the issue of power restrictions. Looking to increase their odds of approval, some got in the habit of applying for grid interconnects from several utilities when they needed only one. Others sought grid capacity for speculative projects that were far from certain to move forward. This overreach created a lot of noise as grid operators attempted to forecast demand, chose which projects to approve, and built the required capacity.
The utilities caught on to the overreach and changed the way they review and approve connection requests. For one thing, they implemented a fee schedule that requires an upfront deposit and subsequent fees as project milestones are reached. This reduced queues by up to half, enabling a more accurate demand picture and better planning.
Utilities have also changed how they perform impact analyses for data center projects. In the U.S., the Federal Energy Regulatory Commission’s FERC Order 2023 made it possible for transmission operators to evaluate multiple data center proposals simultaneously rather than one by one. All projects submitted within a certain time window are pooled and reviewed as a single group to determine the cumulative impact on the grid.
Energy strategies that bypass grid connection delays
Although utilities have streamlined interconnection processes, securing grid power remains one of the biggest obstacles to building AI data centers in many regions. Rather than waiting years for new transmission capacity, operators are increasingly diversifying how they power their facilities by investing in generation on-site or adjacent sites, energy storage, renewable energy, and microgrids. Another option is an energy park—a megawatt-scale power generation site with multiple gas turbines powering data centers directly.
Major hyperscalers are already putting these strategies into practice. For example, Stargate’s 1.2-gigawatt mega data center project in Abilene, TX, draws most of its power from its own natural gas plant. Meta is taking a similar approach with its Hyperion AI data center in Richland Parish, Louisiana. The company is planning to build 10 natural gas power plants to generate 7.5 gigawatts of electricity.
In other cases, hyperscalers are making deals to reactivate dormant nuclear power plants – which is the case with Microsoft and Three Mile Island – and to secure power from coal-fired plants that were scheduled for decommissioning but will now operate longer than planned.
Beyond large-scale generation projects, operators are adopting a range of technologies to improve flexibility and reduce dependence on utility interconnections:
- Battery Energy Storage Systems (BESS) – On-site or nearby energy storage allows operators to overcome interconnect delays and tap into renewable energy sources for backup power.
- Wind and Solar Farms – In areas where these renewable sources are abundant, data centers can harness them to run facilities at least some of the time. In addition, excess power that normally would be curtailed can be stored in BESS.
- Microgrids and Energy Parks – By generating their own power from renewables and other sources, microgrids allow data centers to run independently of the grid.
Other approaches involving innovations such as hydrogen fuel cells and nuclear-powered small modular reactors (SMR) are being tested. Data center operators are cooperating with a consortium called DCFlex to explore collaboration with grid operators to minimize grid stress through more flexible data center loads. Membership in the group, organized by the Electrical Power Research Institute (EPRI), includes NVIDIA, Schneider Electric, utilities, and other stakeholders.
Planning beyond the grid connection
AI demand isn’t waiting for the grid to catch up. The organizations that think beyond the traditional grid connection with solutions such as combining utility power with on-site generation, energy storage, and intelligent energy management will be in the strongest position to deploy AI infrastructure faster. Learn how Schneider Electric helps data center operators navigate grid constraints with integrated power, energy management, and resilient infrastructure solutions.
Add a comment