Five. That’s how many companies NVIDIA put on stage at Climate Week to show what AI is actually doing for clean energy right now. Not five hundred pilots, not a slide deck of someday. Five: ThinkLabs AI, Atomic Canyon, Redwood Materials, TerraPower, and Commonwealth Fusion Systems.
Small number, and I think that’s the most interesting part. When a company as loud as NVIDIA picks five examples to highlight, you’re looking at the work that’s far enough along to talk about out loud. So let’s walk through what these companies are doing, and more importantly, what kind of AI they’re actually using — because it’s probably not what you’re picturing.
The unglamorous bottleneck nobody talks about
Start with ThinkLabs AI, because their story is the clearest window into how this works.
If you want to connect a new solar farm, wind project, or battery installation to the electrical grid, you can’t just plug it in. The utility has to run an interconnection study — a detailed analysis of what happens to the grid when your new power source shows up. Does it destabilize things? Does a line overload on a hot afternoon? These studies are slow, technical, and there’s a queue.
Southern California Edison used ThinkLabs software and reduced the time those evaluations take.
That’s the whole headline, and it’s deliberately understated compared to what people expect from AI news. No robot. No chatbot. Just a very hard engineering calculation getting faster. And faster interconnection studies mean clean energy projects that already exist on paper get to start producing electricity sooner. The generation capacity was never the hard part. The queue was.
What these systems are actually doing
Here’s the pattern I keep seeing across the five: none of them are asking AI to be creative. They’re asking it to be fast and tireless on problems where the math is understood but the volume is brutal.
The work spans three areas — grid access, nuclear operations, and fusion development. Think about what those have in common:
- Enormous document piles. Nuclear operations run on regulatory paperwork, inspection records, and technical procedures measured in millions of pages. Atomic Canyon works in this space. Finding the right paragraph in that archive used to be a person with a search bar and a lot of patience.
- Simulations that take forever. Fusion development, which is Commonwealth Fusion Systems’ territory, and advanced reactor design, which is TerraPower’s, both depend on modeling physics that’s expensive to compute. Every design tweak means another run.
- Materials and process complexity. Redwood Materials works on recycled EV batteries repurposed for energy storage. Sorting, grading, and matching used battery cells is a pattern-recognition problem at scale.
These are all jobs where a human expert knows exactly what needs doing and simply cannot do enough of it. That’s the sweet spot for AI agents right now, and it’s a lot less cinematic than the marketing suggests.
Why NVIDIA is in this story at all
Quick plain-language detour, because this confuses people reasonably often.
NVIDIA doesn’t build clean energy. It builds the chips and software tools that AI models run on. When a fusion company wants to run a simulation a thousand times instead of ten, it needs computing hardware built for that kind of parallel work. NVIDIA sells the shovels.
There’s a tidy irony here that’s worth sitting with. AI data centers consume a lot of electricity — enough that tech companies are now collaborating specifically on data center energy efficiency. So the same technology driving up power demand is being pointed at the problem of generating more clean power. Whether that nets out positive is a real open question, and I’d rather name it than pretend the story is purely triumphant.
The quiet money signal
One more data point that tells you where this is heading. Erhan Eren runs Enki, a commercial intelligence platform for emerging technologies and infrastructure projects. His backers include Equinor, Techstars, and NVIDIA.
Equinor is an energy company. Techstars is a startup accelerator. NVIDIA is a chip company. When those three end up on the same cap table, it means the energy industry and the AI industry have stopped being separate conversations. Infrastructure people want AI tooling. AI people want energy access. They’re funding the same companies now.
What I’d take away from this
If you’re trying to understand AI agents without a technical background, this NVIDIA showcase is a genuinely useful teaching example — not because of the technology, but because of what got picked.
The winning applications aren’t replacing engineers or physicists. They’re removing the waiting. An interconnection study that took months. A document search that took days. A simulation queue that took weeks. Compress those, and projects that were already designed, funded, and approved simply arrive sooner.
That’s not a dramatic story. It’s a scheduling story. And in clean energy, where the technology mostly exists and the deployment timeline is the actual constraint, scheduling might be the thing that matters most.
🕒 Published: