As tech giants seek nuclear power for AI data centers, the flip side of that coin is emerging — using AI to improve reactor licensing and construction.

Why it matters: Energy demand for data centers is rising fast, and nuclear energy is carbon-free and round-the-clock.

• But getting reactors approved and built is really complex and time-sucking.
• So hyperscalers, startups and federal agencies want AI to speed things up — but carefully.

Driving the news: This week, the latest efforts emerged around both next-wave reactor designs and incumbent models.

• DOE's Idaho National Lab will use a Microsoft tool to create safety and analysis reports that are part of construction and license applications, the parties said yesterday.
• It ingests and analyzes engineering and safety info, and generates documents that the Nuclear Regulatory Commission and DOE require, they said.

"This is a big deal for the nuclear licensing process," Jess Gehin, a top INL scientist, said in a statement.

• "Introducing AI technologies will enhance efficiency and accelerate the deployment of advanced nuclear technologies."

The intrigue: INL and the company are emphasizing that humans aren't taken out of the loop.

• Microsoft's Nelli Babayan tells Reuters: "It's created for human refinement, so a human can go through each of the sections and, specifically as needed, edit any of the sections, whether manually, or maybe with the help of AI — it's really up to the human."

State of play: Separately, Westinghouse and Google said Tuesday they're working together to harness AI to make building Westinghouse reactors an "efficient, repeatable process."

• They're combining Westinghouse AI tools built on proprietary data with Google's cloud capabilities.
• The tools can help deploy the Westinghouse AP1000 — the same reactor at Plant Vogtle in Georgia, the most recent U.S. project — as well as its SMR design, or improve existing plants, they said.

Catch up quick: Applying AI to nuclear is underway on several fronts. It spans efforts to build new reactors to aiding with regulatory compliance at today's plants.

• The NRC is exploring a suite of ways to stitch AI into the fabric of its work.
• Startups in the space include Atomic Canyon, which recently raised a $7 million round led by prominent climate VC firm Energy Impact Partners.
• Its platform is already being deployed at PG&E's Diablo Canyon plant in California.

The bottom line: Look for more tie-ups ahead like the Westinghouse-Google and INL-Microsoft collaborations.

🛑 The Transportation Department said it's terminating about $4 billion in unspent funding for California's long-delayed high-speed rail project, alleging state officials haven't shown they can meet obligations.

• What's next: Legal battles, maybe. California Gov. Gavin Newsom accused DOT of illegally pulling funding and said all options are on the table.

⚛️ DOE is launching a pilot program to help create more domestic nuclear fuel production lines.

• The big picture: It "leverages the DOE authorization process" for lines used for R&D and to "provide a fast-tracked approach to commercial licensing," the agency said.

⚔️ DOE head Chris Wright tells Bloomberg that the U.S. may leave the International Energy Agency.

• Why it matters: It raises the stakes of GOP criticisms that IEA takes an unrealistic view of energy transition, though IEA has rebutted the claims.
• State of play: "We will do one of two things: we will reform the way the IEA operates or we will withdraw," he tells Bloomberg, adding his preference is to remain.

🌤️ Via Politico, an internal memo shows that "solar and wind energy projects must now get Interior Secretary Doug Burgum's personal sign-off to receive permits across the hundreds of millions of federal acres."

• What we're watching: Whether the new policy further slows efforts to deploy wind and solar on federal lands as Trump 2.0 officials target those technologies from multiple angles.

🗞️ Microsoft has a new offtake deal with the removal startup Vaulted Deep for up to 4.9 million tons over 12 years through 2038.

• Why it matters: That's a really big contract for the young CO2 removal industry, and assumes Vaulted Deep can scale a lot — its removals to date total nearly 18,000 tons, it said.
• How it works: Vaulted collects organic materials — like farm wastes and biosolids — that would otherwise release CO2 and methane when burned, landfilled or spread on land. They convert this into a slurry that's injected deep and permanently underground.

📦 Amazon's CO2 emissions rose 6% in 2024 after two years of declines, but the tech and e-commerce giant also claimed progress in several areas, like a 16.4% drop in plastic packaging. Full sustainability report...blog post

🤝 Palo Alto Networks has bought 10,000 tons of removal credits from direct air capture firm 1PointFive, a subsidiary of oil giant Occidental Petroleum.

• Why it matters: It's among the furthest along of the DAC players. The credits will come from 1PointFive's Stratos plant in Texas that's coming online this year.

General Motors is partnering with Redwood Materials, a battery recycler and energy company, to supply U.S.-built batteries for backup storage systems.

Why it matters: The companies are responding to soaring demand for battery storage systems that can offset power outages and reinforce the grid during peak demand.

Between the lines: It's also a hedge against slower-than-expected EV sales.

• While GM's EV lineup continues to expand, microgrid storage is another market for its made-in-America batteries.

Driving the news: The deal expands an existing partnership with Redwood Materials to supply end-of-life EV batteries for recycling or energy storage solutions.

• Already, second-life batteries from GM's EVs are being repurposed to help power the largest microgrid in North America. The Redwood installation in Sparks, Nevada, supports the AI infrastructure company Crusoe.

What's next: The agreement ensures more feedstocks for Redwood Materials' newest business, Redwood Energy, which seeks to deploy new and used EV packs into energy-storage systems.

Full story

Nvidia unveiled a new AI weather forecasting model it calls capable of making very fast forecasts and better estimating risks of extreme events.

Why it matters: Tech companies are moving deeper into AI-based models at a time when climate change is worsening some risks and harms.

Driving the news: Nvidia said its new FourCastNet 3 can generate a 15-day global forecast in just 64 seconds.

• Beyond speed, the firm published research stating it provides more accurate forecasts of high-impact events like extreme heat waves.

The bottom line: AI holds the promise of providing more accuracy than traditional, physics-based practices.

• "AI changes the game in so many ways — how we estimate the state of the atmosphere, how we launch forecasts from that state, how we sidestep physics to make better forecasts, how we build climate simulations we couldn't dream of before," Mike Pritchard, Nvidia's head of climate simulation, said in a video alongside the launch.

Storms sweeping through the U.S. this summer have dumped intense rain on cities across the country, left towns flood-ravaged and forced water rescues.

The big picture: Scientists who spoke to Axios say the deadly floods in Texas that killed more than 130 people underscore the risk that climate change can worsen extreme rainfall events.

• Climate change "is supercharging the water cycle," sparking heavier precipitation extremes and related flood risks, according to Climate Central, a climate research group.
• "With all these events, what they have in common is that in a warmer world, our atmosphere can hold more moisture," said meteorologist Shel Winkley, the weather and climate engagement specialist with the group.

Full story

That's Morningstar Equity Research's new estimate of U.S. data center power demand in 2030, roughly triple 2024 levels.

The intrigue: That's a big jump, but Morningstar stresses that it's below some of the most bullish projections floating around that exceed 100GW.

• "We believe such forecasts overlook the practical limitations associated with building large-scale infrastructure and also underestimate the long-term rising energy efficiency of AI chips," Morningstar said in a note.