Plans for AI data centers already call for more power than the largest nuclear power plants in America can provide, according to 7investing founder Simon Erickson.
Most of the recent nuclear enthusiasm has focused on reviving old plants. Erickson and executive producer Heather Horton spent their Wednesday livestream on the newer bet instead: small modular reactors, sized to sit behind a single data center's meter rather than feed the wider grid.
"We now have plans for AI data centers that are going to require more power than the largest nuclear power plants in America today can provide."
Erickson has run 7investing's stock research service for nearly seven years; Horton previously worked in the renewable energy industry before joining as the show's executive producer.
I listened to the full episode so you can skip it. 27 minutes of audio, 11 minutes of reading.
Here are the 8 numbers that matter.
🎙️ Hosts: Simon Erickson, founder and chief executive of 7investing, and Heather Horton, 7investing's executive producer and a former renewable-energy industry professional
📰 Published: 10 September 2026 on The 7investing Podcast's YouTube livestream
🔴 YouTube | ⏱️ 27 min | ✅ Time saved: 19 min
Key Takeaways
AI data centers already need more power than America's largest nuclear plants can supply
That's the reason SMRs, sized to fit a single project, are getting serious attention now
Demonstration-phase SMRs cost $80-130 per megawatt hour, against $50-70 for wind, solar and hydro
The low end of the SMR range is still pricier than the high end of renewables
A second generation of SMRs is expected to reach $40-90 per megawatt hour within five or six years
That's the point Erickson says nuclear starts getting cost-competitive with renewables
Nuclear ranks among the safest power sources by deaths per unit of energy, just behind solar
NuScale's reactor was the first SMR design approved by the US Nuclear Regulatory Commission
Its modules run about 80 megawatts each; a dozen strung together approaches a gigawatt
GE Vernova is building SMRs with Hitachi as an iteration of prior reactor designs, not a clean-sheet approach
Oklo cools its reactors with liquid sodium instead of water and plans to recycle spent uranium fuel
It's also selling some of its spent radioactive isotopes for uses like cancer therapy
The power some single AI data center projects need is starting to approach a mid-sized American city's usage
1. Data Centers Outgrow Grids
Erickson opened by explaining why small modular reactors are suddenly relevant, tracing the shift from decades of commoditized power to today's data center demand.
He framed the historical problem with nuclear as scale mismatched to need. Traditional plants are "very, very large," tend to run as "mega projects" tied to "mega utilities," and can cost "$50 billion or more" to build — a price that needs enormous demand to justify
The premise flipped once AI data centers arrived. "We've completely changed the game again where we've got these AI data centers which are consuming more power than the largest nuclear power plants in America can provide." He repeated the point directly: "We now have plans for AI data centers that are going to require more power than the largest nuclear power plants in America today can provide."
SMRs are pitched as the fix because they're modular rather than monolithic. Reactors can be built at whatever scale a single project needs and tied together for more, which Erickson called "perfect for AI" because data centers themselves vary widely in size
2. Today's Cost Gap
Erickson walked through a chart comparing the levelized cost of energy (LCOE) across power sources, setting expectations before naming any of the three SMR companies.
Mature renewables have gotten cheap. Excluding heavily subsidized Chinese production, he put hydropower at roughly $70 per megawatt hour, solar photovoltaic at $55, and onshore wind at about $51 — all within reach of natural gas, the traditional benchmark for new power
Demonstration-phase SMRs are still more expensive than the priciest renewable on the list. NuScale, GE Vernova and Oklo have each claimed levelized costs of $80-130 per megawatt hour for their initial reactors, and Erickson said the real number is likely to land above that range, "closer to $150 plus," since none of the three has actually produced power yet
He was explicit that the technology isn't cost-competitive today. By his own estimate, SMRs won't compete on price with hydropower, solar or onshore wind anytime soon
3. The Path to Cost Parity
Having laid out the bad news, Erickson pivoted to why he still expects the economics to improve.
Unlike mature technologies, SMRs have real room to get cheaper. Coal and gas plants are too mature to gain much more from scale; nuclear SMRs, by contrast, can cut both operating and capital costs as manufacturers build more units and refine the process
He put a number on the next phase. "Second phase of iteration is pretty much expected to be between $40 and $90 per megawatt hour later on," once demonstration projects give way to commercial reactors
That's the point he says nuclear starts competing directly. Getting to $40-50 per megawatt hour "within five, six years" would put SMRs in range of today's cheapest renewables
4. Nuclear's Safety Record
Horton raised a safety comparison chart, prompting Erickson to address the industry's reputational baggage directly.
Horton said the data surprised her. She called nuclear "actually one of the safest" power sources, ranking just behind solar, and noted it also produces fewer greenhouse gas emissions than solar, wind or hydro by her source
Erickson attributed nuclear's poor reputation to a handful of severe, well-known accidents — Three Mile Island, Chernobyl and Fukushima — even though the broader safety data holds up well once redundant safety systems are added, which smaller reactor designs make easier to build in
5. NuScale's TVA Bet
Erickson gave the most detailed rundown on NuScale, the company he said is starting furthest from legacy nuclear design.
NuScale already has a live international project. It's working with Romania's state power utility on a utility-scale SMR deployment, alongside smaller municipal projects in Utah and elsewhere in the western United States
Its US regulatory position is a real edge. Erickson said its power module was the first SMR design to receive approval from the US Nuclear Regulatory Commission, and its modules can be strung together, running roughly 80 megawatts each, so about a dozen reach gigawatt scale
The bigger opportunity he flagged is a potential deal with the Tennessee Valley Authority. He called it a "massive hockey stick potential upside" that "would immediately reduce the cost curve in a very positive way for everybody" if TVA converts part of its system to SMRs — he sized the TVA system itself at about six gigawatts overall
6. GE Vernova's Iteration
Turning to the second company, Erickson drew a sharp contrast with NuScale's from-scratch approach.
GE Vernova is GE's energy business, spun off into its own company. It's building SMRs in partnership with Hitachi
Erickson reads it as evolution rather than a clean-sheet design. He said GE Vernova's approach looks like "a little bit more of an iteration of traditional technology that Hitachi had before" rather than something as novel as its smaller competitors, though he still called the company "certainly another player to keep an eye on"
7. Oklo's Sodium Bet
Oklo drew the longest, most technical discussion, including a back-and-forth with Horton on why it cools reactors differently from the rest of the industry.
The company's name comes from a natural phenomenon. Erickson described the Oklo natural reactor zone in Africa, where uranium deposits sustained their own fission reactions roughly two billion years ago, with groundwater acting as a natural moderator
Oklo's business model goes beyond building reactors. It plans to recycle spent uranium fuel back into new rods rather than sending it to long-term storage, and separately plans to sell other spent radioactive isotopes for uses including cancer therapy and academic research
Its reactors use liquid sodium instead of water as the cooling source, a design Erickson credited to MIT engineers. Horton asked directly what the benefit is; Erickson said it functions as the cooling and heat-transfer medium for the reactor, calling the overall approach a ground-up rethink rather than a tweak to existing designs
The fuel itself is different too. Horton noted Oklo uses a high-assay, low-enriched form of uranium — enriched further than a conventional reactor's fuel — some of it purchased from existing suppliers rather than mined directly
Financially, the company is still pre-revenue in any meaningful sense. Erickson said it currently has "basically zero revenue," relying on licensing deals and demonstration work with other companies while it builds out the technology
8. How Big Is "Big" Now
Closing the discussion, the hosts tried to size just how much power a single AI project can require.
Horton looked up a comparator and found Chicago uses about 87 gigawatt-hours of electricity a day, which Erickson clarified was a daily figure rather than annual
He said some individual Tennessee data center projects already exceed San Diego's power consumption, and that the sector is closing in on the scale of a major US city faster than he expected
He named other buildouts on the same scale. Bloom Energy's fuel-cell business signed a 2.8-gigawatt deal with Oracle, and OpenAI, Anthropic, Grok and SpaceX are all building data centers at a comparable pace
Bonus Insights
Erickson said 7investing's own subscriber call the next morning would cover its five most recent official stock recommendations, several of them in the energy sector, along with updates across its broader coverage list
Erickson and Horton's bottom line is that small modular reactors are the right direction for meeting AI's power demand, but the numbers today still favor established renewables — and NuScale's regulatory lead, GE Vernova's incumbency and Oklo's fuel-recycling bet are the three different paths racing to close that gap over the next five to six years.
Products, Companies & Tools Mentioned
NuScale Power (SMR — the first design approved by the US NRC; working with Romania's utility and pursuing a possible deal with the Tennessee Valley Authority)
GE Vernova (GE's spun-off energy business, building SMRs with Hitachi; Erickson calls it more iteration than reinvention)
Oklo (Sodium-cooled SMR developer planning to recycle spent uranium fuel and sell other isotopes for uses like cancer therapy)
Bloom Energy (Solid-oxide fuel cell company with a 2.8-gigawatt power deal with Oracle, cited as a comparable scale of AI-driven power buildout)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: