303 Gigawatts Agreement, AI Power Grid Verification Underway

The surge in power requests from AI data centers in the United States has initiated a process to filter the actual amounts that will be connected and operational on the power grid. While the scale of applications has increased, they must pass financial agreements, transmission network bottlenecks, and cost burden criteria to be recognized as actual loads.

Wood Mackenzie stated in a podcast description released on June 16 that the moment utilities require financial agreements, two-thirds of data center power requests tend to disappear. However, the statement from Zero Hedge that 'more than two-thirds of 1,066 GW remain unrealized' cannot be viewed as a figure defined in the same way as Wood Mackenzie's public description.

The key point confirmed is the gap between 'requests' and 'agreements.' Wood Mackenzie noted in a summary of its July report that the reliable large load agreements for U.S. utilities increased to 303 GW in the first quarter of 2026. In separate published data, it tracked proposed data center capacity in the U.S. at 134 GW, up from 50 GW a year ago.

Compared to the scale of the U.S. power grid, the burden is not small. According to the U.S. Energy Information Administration (EIA), the capacity of utility-scale power facilities in the U.S. is 1,280,799 MW by the end of 2025. The 1,066 GW corresponds to about 83.2% of this capacity in simple arithmetic.

This figure illustrates how much difference it makes to assume that all requests become actual loads compared to the current scale of the power grid. Data centers can be built relatively quickly, but expanding power plants and transmission networks typically requires a longer time.

The biggest point of contention is Texas. The Electric Reliability Council of Texas (ERCOT) approved the 'Batch Zero' procedure for connecting large power users on June 18 and reported tracking large load requests exceeding 438,000 MW, of which about 89% are data center requests.

We previously covered the proportion of data centers in Texas's large load requests. At that time, the issue was whether the requested power could be recognized as long-term demand and who would bear the necessary facility costs.

The Texas state government has also put the cost issue front and center. Texas Governor Greg Abbott directed the Public Utility Commission (PUC) and ERCOT on June 10 to ensure that the costs of data center expansion are not passed on to existing residential ratepayers. On July 24, a follow-up response indicated that new data centers must meet stronger standards and be able to quickly reduce power when directed by ERCOT.

PJM is also addressing the same issue. The PJM board announced measures on January 16 to reliably integrate data centers and other large loads into the power grid. This includes a conditional connection system where large loads secure their own generation or reduce power use when needed by the grid.

Reuters reported that a large amount of data center power suddenly disconnected from the grid in Northern Virginia, causing voltage disturbances. This case shows that the demand for data center power can lead to actual grid stability issues beyond simple supply and demand forecasts.

Alternatives being discussed include self-generation, batteries, gas generation, and flexible connection methods. Wood Mackenzie explained in a press release on May 27 that while expanding the U.S. transmission grid takes 5 to 10 years, data centers can be built more quickly. The same report stated that there are over 90 GW of co-located generation in the U.S. interconnection queue, but many projects are unlikely to materialize.

The interconnection queue is a list of power plants or large power users waiting for power grid connection reviews. Co-located generation involves placing generation facilities near data centers. This structure attaches generation sources close to demand rather than building a new power grid, but it requires capital, land, and fuel procurement conditions to be met.

There is also a contrary trend. BloombergNEF raised its forecast for U.S. data center installation capacity to 118 GW in a public document released on July 29, stating that the U.S. data center pipeline announced since December 2025 has increased by an additional 101 GW. While demand forecasts continue to rise, actual construction and completion must pass through power grid, cost, local opposition, and regulatory standards.

The direct connection to the domestic market was not confirmed in this public document. The focus of this issue is not on specific stocks or token prices but on how the U.S. power grid will filter large load applications based on certain criteria.

The debate over power requests from U.S. data centers is occurring at the intersection of expanding AI infrastructure and the physical limits of the power grid. The conclusion revealed by public documents is that the scale of applications has increased, and grid operators have begun to demand faster and more expensive verification processes.