Who Pays to Power AI? A Nuclear Restart Loan and New State Data Center Rules

In the space of two days this week, Washington agreed to lend up to $1.9 billion to bring an Iowa nuclear plant back to life for Google, and Massachusetts told large data centers they cannot get state permits unless they supply their own clean power and win over their host communities. The two moves pull in different directions, one subsidizing new supply and the other putting conditions on new demand, but they answer the same question: who pays for the electricity behind AI? For enterprise leaders, the answer increasingly shapes where AI capacity gets built, how fast, and at what price.

Why is a mothballed Iowa reactor getting a federal loan?

On September 8, the Department of Energy said it had closed a loan of up to $1.9 billion to NextEra Energy to restart the Duane Arnold Energy Center in Linn County, Iowa. The plant, which stopped operating in 2020, would supply 615 megawatts of baseload power, enough for nearly 500,000 homes, according to DOE. The restart still needs Nuclear Regulatory Commission licensing approvals.

Google is the main reason the plant is coming back. TechCrunch reported that Google announced in October 2025 that it would help revive the plant, that a restart is targeted for 2029, and that Google is reportedly considering up to six data centers near the site. NextEra mothballed the plant after a 2020 storm damaged it, TechCrunch noted.

Deputy Energy Secretary James Danly said in the DOE announcement that the restored output "will drive down electricity costs." TechCrunch pointed out that he did not explain how. DOE said the project would create about 1,500 construction jobs and more than 450 operations jobs.

The loan follows a similar pattern. DOE lent $1 billion to Constellation Energy in 2025 to restart a Three Mile Island reactor that Microsoft has contracted to use, and Meta has bought the clean energy attributes of Constellation's Clinton plant in Illinois, according to TechCrunch.

What does the Massachusetts data center order require?

The same day, Massachusetts Gov. Maura Healey signed Executive Order No. 658. It applies to new or expanded data centers with more than 25 megawatts of peak demand. Key provisions include:

• Community benefits first. State agencies may not issue permits unless the developer submits a community benefits agreement that meets state standards and shows it follows the state's framework on air, water, public health and community engagement.
• Bring your own clean power. Projects must procure enough new clean electricity to match their annual consumption. Those that fall short will make alternative compliance payments into a Ratepayer Protection Fund.
• Grid costs stay with the data center. The order directs regulators to prioritize rate schedules for large loads so other customers do not pay for distribution upgrades, and to use fees or deposits to clear speculative projects from interconnection queues.
• No secrecy deals. State agencies may not use nondisclosure agreements with data center projects.

"Unless a community says yes to a data center, we're saying no," Healey said at a press conference, WBUR reported. The Conservation Law Foundation called the order a useful step but said past decisions also need attention. Republican gubernatorial candidate Mike Minogue argued the state instead needs more energy supply, including gas pipelines and nuclear, WBUR reported.

TechCrunch described Massachusetts as the third state in three months to tighten data center rules. Texas announced in August that new data centers must undergo audits by its utility commission and grid operator, and New York halted construction of new data centers of 50 megawatts or more in July, according to the report.

Is the AI power crunch only an American problem?

No. On September 9, Nvidia announced it is working with eight Australian data center and cloud operators, including AirTrunk, NEXTDC, CDC and IREN, to expand land, power and building capacity for up to 2 gigawatts of AI infrastructure by 2027. IREN is developing an 800-megawatt campus in South Australia, and CDC said it has 800 megawatts under construction on top of more than 550 megawatts it already runs in Australia and New Zealand. Atlassian is among the named users. The release framed power as the core input, with Nvidia's Raj Mirpuri saying AI factories "turn energy into intelligence."

What should enterprise leaders take from this?

Taken together, the week's news shows governments trying two answers at once: underwrite new generation, and make large new loads carry more of their own costs. Both approaches will reach enterprises that never build a data center themselves.

Expect power costs to show up in AI pricing. If developers must fund clean generation, pay into ratepayer funds or cover grid upgrades, those costs flow into colocation rates and, eventually, cloud and AI service prices. Finance teams modeling multi-year AI spending should treat energy as a variable, not a constant.

Location now carries regulatory risk. Rules differ sharply by state, from audits in Texas to a construction halt in New York to community approval in Massachusetts. Companies planning private AI infrastructure or long-term colocation contracts should weigh permitting timelines alongside latency and tax incentives.

Power density is an engineering limit, not just a budget line. HPE Federal AI Chief Technologist Steve Heibein made the point bluntly during a AI Connect webinar: "A thousand watt GPU needs a thousand watts of cooling." Older facilities often cannot support modern AI racks, which is part of why the AI factory model has gained ground and why cooling now shapes data center design.

Idle capacity is the costliest kind. When power and GPUs are scarce, wasted compute hurts more. At The AI Enterprise Conference 2026, Everpure VP of AI Kaycee Lai argued that the AI winners are not the companies that buy the most hardware. "They're not the ones who buy the most GPUs," he said. The advantage goes to teams that bring AI to well-prepared data and keep their infrastructure busy.

Sustainability reporting will get harder to sidestep. The Massachusetts order requires annual disclosures from operating facilities. Enterprises with emissions targets should ask their providers how AI workloads are being powered and whether the clean energy claims rest on new generation or on certificates.

What happens next?

In Iowa, the restart now depends on federal licensing. In Massachusetts, regulators face December 31 deadlines to publish municipal guidance and set up the compliance payment system, and a state energy affordability bill with similar guardrails remains in conference committee, WBUR reported. Whether other states follow the Massachusetts model, or choose the Texas approach of audits rather than conditions, will help decide where the next wave of AI capacity lands.

Frequently asked questions

What is the Duane Arnold restart loan?

The Department of Energy closed a loan of up to $1.9 billion to NextEra Energy on September 8 to restart the Iowa nuclear plant, which stopped operating in 2020. Google backs the revival and, per TechCrunch, a restart is targeted for 2029, pending NRC approvals.

What does Massachusetts Executive Order 658 require of data centers?

Projects above 25 megawatts of peak demand need a community benefits agreement before state permits, must procure new clean electricity to match annual use, and cannot use NDAs with state agencies.

How do data center power rules affect enterprises that do not build data centers?

Costs for clean power, grid upgrades and compliance tend to pass through to colocation and cloud pricing, and permitting differences affect where and how fast new capacity comes online.

Filed under: Data & AI Infrastructure, AI Strategy & ROI, AI Governance & Trust