Energy forecasters have been caught flat-footed by AI's explosive power appetite, and the numbers tell the story. BloombergNEF just revised its projection for US data center electricity consumption by 2035 upward by 83% in just seven months, now forecasting 194 gigawatts instead of the 106 gigawatts it predicted in December 2025. That would mean data centers alone consume roughly one-fifth of all US electricity by the middle of the next decade, compared with just 5.9% today.

The shock is not that AI uses a lot of power. The real story is that the buildout is outrunning the assumptions used to finance it. When BloombergNEF, the Electric Power Research Institute (EPRI), and S&P Global all move sharply upward in the same window, it signals something deeper: the infrastructure world has been caught unprepared for the speed and scale of hyperscaler expansion.

Why Are Energy Forecasts Missing the Mark So Badly?

The revision cascade reveals a pattern. Hyperscalers announce new facilities, utilities scramble to create interconnection plans, forecasters revise their numbers upward, and investors then act as though the new forecast was always obvious. It wasn't. BloombergNEF's December forecast was already substantial. The July number makes it look cautious.

Other forecasters have been dragged in the same direction. EPRI's 2026 Powering Intelligence report says its revised projections are about 60% higher than its 2024 estimates, with data centers consuming between 9% and 17% of US electricity by 2030. S&P Global lifted its 2030 US data center grid power forecast from 134.4 gigawatts in October 2025 to 183.2 gigawatts just six months later in April 2026.

Four companies sit near the center of this pressure. According to Rabobank's analysis of BloombergNEF's data center capacity database, Meta, Amazon Web Services, Microsoft, and Google controlled about 13 gigawatts of live information technology capacity in North America as of September 2025, representing 42% of the region's total live IT capacity. These same four companies could collectively spend $700 billion this year based on their quarterly earnings calls.

How Are Tech Giants Planning to Power AI Data Centers?

Nuclear energy has become the answer many tech companies want to hear. According to SMR Intel's nuclear data center tracker, 13 announced projects had committed 9.8 gigawatts of nuclear capacity for AI infrastructure as of early July, with Microsoft, Google, Amazon, and Meta all tied to nuclear deals. Microsoft has secured a 20-year agreement with Constellation Energy tied to the planned restart of Three Mile Island Unit 1. Google has a deal with Kairos Power. Amazon has backed X-energy. Meta has been linked to multiple nuclear procurement efforts.

The scale of these commitments is striking. One gigawatt roughly equals the output of a traditional nuclear reactor. Some next-generation AI campuses are being planned at a scale where that comparison stops sounding dramatic and starts sounding operational. A single large campus can become a power system problem in its own right.

Behind-the-meter procurement, old power plants, and nuclear agreements represent the primary strategies developers are using to secure power while grids struggle to keep up. BloombergNEF has tracked 124 gigawatts of announced on-site gas capacity for data centers, although only a small share is under construction. That caveat matters significantly. Announced capacity does not cool servers. Built capacity does.

What Infrastructure Gaps Are Holding Back AI Expansion?

The uncomfortable reality is not BloombergNEF's 194-gigawatt forecast itself. It is what happens if this one is also too low. The US has not built energy infrastructure on this kind of timetable in recent memory, and local resistance is already showing up in county board rooms, state legislatures, and the ratepayer fights that follow. AI may still be sold as software, but the bill underneath it is concrete, steel, turbines, substations, and land.

For founders and venture capitalists, this changes where the interesting companies are likely to appear. When electricity access becomes the bottleneck, certain markets shift from nice-to-have to mission-critical:

• Cooling systems: Data center cooling is no longer a side market but part of the critical path to expansion, as thermal management directly limits how much compute can be deployed in a given space.
• Grid software and demand response: Startups that help a data center shift inference workloads to cheaper hours or prove flexible demand to a utility are no longer selling efficiency tools; they are selling permission to build.
• Distributed generation and workload scheduling: Companies that optimize where and when compute happens can unlock new capacity without waiting for grid upgrades, making them essential to hyperscaler timelines.

The same pressure changes the chip market fundamentally. If power is the constraint, efficiency becomes the premium. Full stop. Nvidia still has the commanding position in AI accelerators, but chips and systems that produce more inference per watt will be judged differently when power contracts, substations, and generation sit inside the cost of compute. AMD, hyperscaler custom silicon, networking gear, cooling loops, and related infrastructure all get pulled into the same calculation.

How Are Policymakers Trying to Protect Ratepayers?

The White House has attempted to address affordability concerns through a voluntary pledge. When President Donald Trump unveiled the Ratepayer Protection Pledge on March 4, 2026, it bound Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI to cover the cost of generation and grid upgrades their data centers require rather than pass those costs to existing customers. The pledge has since expanded dramatically.

A White House list of signatories obtained by the Wall Street Journal shows nearly 200 entities, including electric utilities, data center developers, and others, have now committed to the pledge. The signatories include utility majors NextEra Energy and Duke Energy alongside data center landlords Equinix and Digital Realty. Republican governors Greg Gianforte of Montana, Mark Gordon of Wyoming, and Mike Kehoe of Missouri also signed on.

The commitments themselves are straightforward. Signatories agree to build, bring, or buy the new power their facilities need; pay for delivery-infrastructure upgrades those facilities trigger; negotiate separate rate structures with utilities and states; and hire from local communities. The load-bearing clause is a promise to pay for contracted power "whether they use the electricity or not," the provision meant to stop households from absorbing the cost of capacity a data center reserves and then leaves idle.

However, enforcement remains the gap the expansion does not close. Retail electricity prices are set by state public utility commissions, not the White House, and the effort could prove hard to enforce for that reason. The Brookings Institution has argued that the pledge will not protect anyone until state legislatures and regulators write its commitments into the tariffs that govern what large customers actually pay. The Energy Department says it is working with the White House to implement the pledge, but its levers are federal grid-reliability and interconnection rules, not the retail rates that show up on a monthly bill.

States are not waiting for Washington. Virginia, Ohio, and Oregon have already created separate rate classes that require large-load data centers to carry their own infrastructure costs. A bipartisan bill in Congress would write the pledge's terms into federal law, though authority over retail rates would still rest with the states.

The test for all these efforts remains unchanged. Whether any of this reaches a household's bill will be decided in rate cases before state commissioners bound by none of it, which is why the White House keeps trying to enlist the people who actually set electricity prices.

" }