
Kevin Kong, CEO and founder of Everstar, discusses how artificial intelligence can streamline nuclear licensing and accelerate the deployment of advanced nuclear power projects.
Loading summary
A
Foreign. Welcome to Currents and Norton Roads Fulbright podcast. Today we're recording with Kevin Kong, CEO and founder of Everstar. Everstar is building a platform that changes the way that advanced nuclear power projects are licensed. Kevin, thanks for recording with us today.
B
Pleasure to be here, Todd. I'm excited.
A
All right, so you got a startup company that's going to help create more nuclear generation. Maybe you could start by what? Why did you think you needed this company? What's the problem you're trying to solve?
B
Well, it really starts my personal obsession with energy abundance. How do we 10x energy production for humanity wherever anybody might live? I just stumbled upon this problem as the common denominator of what everybody here on this earth has to deal with. And our history just been humanity just trying to unlock more and more energy for useful things like getting farther and farther into frontier territories, to discovering new lands and moving useful commodities back and forth. It's just the cost function of life. So today we're really under the aegis of the hydrocarbon economy. The economics of moving carbon and carbon derived useful work is very the outer limits of what we can achieve. And I just kept thinking about how great power came at great cost to health and the climate. So what would it be like to 10x that energy production and imagine a very different cost function for society? Just been chasing that for a long time and I realized nuclear fission is the only viable pathway. After exploring very deeply with real lessons learned and real valuable lessons learned through mistakes in solar batteries, fusion, space, solar, I just realized fission been staring at us in the face. And there's 70 years of research proven out deployments that we can build upon. And East Asia has proven that with consistent, repeatable, dedicated deployments, you can actually take American reactors down a learning curve and make it cost effectively. It's just only in America that American designs are not being deployed at scale. And so from there on that much rocket science here, I just needed to talk to a lot of smart people. We're at the cutting edge of deployments operating these reactors today and advanced reactor startups that were trying to put new generation types to the grid and started identifying what those key problems were and realized it's not fuel bottlenecks, although there is one that keeps us from being able to deploy new reactors. It's really all the rigor of safety approvals that are required for both the reactor and the site and the supply chain that's keeping everybody from being able to deploy tomorrow. Get shovel ready tomorrow. And all of those were done with the wrong tools of the last few decades a lot of the tooling still stuck in the 70s. So I'm sure you're not update that.
A
I'm sure you're not saying that the safety concerns are not legitimate because that you wouldn't be raising money from people if you said that. So of course what, what Given the legitimate concerns regarding safety, what is it that you think you can do to streamline the approval process to ensure safety while while also increasing decreasing costs and decreasing time to market?
B
What our tool does today is not automate people, it's helping existing professionals do way more with less. So just to level set for people who are new to the world of nuclear, for a given new reactor licensing application, you might be submitting something on the order of 2 million pages of core documentation and 4 million pages of additional documentation to go along with that just to support your initial claims. In the 2 million pages that was the actual numbers from the NuScale license completed around 2016ish timeframe. It took a full decade and half a billion dollars to just pull those documents together. That is way crazier than even people think that FDA and pharmaceutical regulatory procedures are bottlenecks. This is another beast because you can't build anything until the final proof was given for design licensing basis. So even in pharmaceuticals you can at least produce like prototype versions of the pills to test against animals or things like that. But absolutely no development happens until you get the final approval. And those kinds of paperwork loads are just bonkers compared to other industries. A 40 ton pressure vessel might come with more than 40 tons of paperwork if you print out the documents. So that's the kind of rigor that is expected of a reactor design basis. There's many other applications that the NRC and now today if we're first of a kind, Department of Energy must process to prove the safety of different aspects that go into a nuclear power plant. There's a lot that could be done for the lower hanging fruit with modern tooling. There's a lot that could be meta procedures, automated say like looking up different regulatory guides and referring to those with referential integrity, making sure you pull the right regulations, guidelines, prior precedence forward so that you're assigning the right thing instead of the wrong thing that will get you in more trouble or cause more questions to come downstream. Those are things that AI can do well. The final arbiter of safety for a very long time will continue to be a human professional that's been trained in the industry for a long time.
A
So who's actually using this AI tool to is it a sponsor who wants to license technology or is it people within the government or who's doing it?
B
We're catering to both sides of the process, both the submitters and the reviewers. We're in a very fortunate place to be able to do that. We and on the private sector side, the submitters, we're working with fleet operators, so Fortune 500 utilities that own fleets of nuclear power plants to advanced reactor vendors that are trying to put new generation on the grid. So some specific names that we work with today are Southern Co. Operators of multiple units of nuclear power plants, and NuScale, a publicly traded SMR stock. They are one of our larger customers and we also work with supply chain companies. So some sister portfolio companies that run manufacturing operations that service the nuclear industry, they also have pretty tough quality assurance work that slows things down. And then yes, on the reviewer side, we also work with Department of Energy, Idaho National Labs. Those folks will be the ones that have to handle a lot of the first and first of a kind applications. There's a vendor list of 11 approved vendors that have to build first of a kind reactors on INL land. And some of them are trying to reach criticality by 4th of July this year, 2026. And we're fortunate to be working with some of them to accelerate the reviews. They also have a hard time processing a lot of these documents. There's a team of about 10 or so people in America that have to handle all of the forthcoming submissions that are submitted through doe. And we're excited to put our hand tools in their hands to be able to accelerate that process. And we're now currently working with some partners to get into the NRC as well.
A
Isn't the point or one of the points, the small nuclear reactor, our small modular nuclear reactor design, is that the, a large part of the design is fabricated off site and is consistent. So once you have approval for that design, you don't have to go through such an extensive review process going forward. And so isn't that how the market's been trying to address this concern that you identified as well? I mean, not, not that you don't have the problem the first time through, but will this problem also be alleviated by these SMR developers or is it something that is going to be with us as long as we're trying to develop nuclear projects?
B
There are things that will get easier, there are things that will stay hard. That's the cop out answer, but that is the reality. There are these size and scale of SMRs and MMRs, micromodular reactors, you could think of that as usually truck bed reactors, sub 100 megawatts. They test to the limits what current regulations should look for. All of the existing regulations so far and in large part are hypertuned to large gigawatt scale pressurized water reactor regulations or sorry, not, not just pressurized light water reactor that's both pressurized and boiling. Our regulations are tough to deal with because of that. It's not really flexible to scale down to what SMRS should be governed by and what MMR should be governed by. There's a lot of rules pertaining to the site. What you must do to protect the environment and the workplace, the mitigations you must put in place. That's just in the federal. There's a lot of state level, municipal and even sometimes tribal regulations that you must contend with. And that permit matrix has to be checked off all the way through to get a qualified site for that particular reactor you want. A lot of the regulations are based on assuming that you have a very large reactor and a lot of potential fallout and safety measures to be put in place that assume you are working under, you know, light water reactors constraints. But when you're on different kinds of grades of fuel with different geometries that just have much better passive safety, maybe there's not as much cooling water access needs that will necessitate reform. So that's where the industry has been moving, but it's not clear cut just yet. People are hoping for it and working towards that, but it's not guaranteed. A lot of the administration's efforts and executive orders have been moving us in that direction, so we hope to see that. But going back to the cop out answer, SMRs alone will not save the planet. They are going to produce energy at much higher rates, wholesale rates per kilowatt hour than gigawatt class reactors. You just can't beat the physics of reactor economics. You get more power the larger you build it spread across linearly growing costs that scale with surface area. So you just can't beat that. Unless you know, the hope here is that the scale, economies of scale, building in factories overcomes that. But there's a lot of ifs to be proven out for that to happen. And even if it does per kilowatt hour, the smaller the reactor, the higher the cost basis will be. So you know, if you're building MMRs, you're replacing diesel generators. If you're building SMRs, likely you'll have to target a lot of premium buyers. First of Electricity or heat, say refiners, industrial processes, or, you know, the customer du jour, the hyperscalers that can pay for premium access to dedicated power.
A
How do you convince regulators and sponsors that your system is not hallucinating? You know, that what, what they, what's getting spit out here, you know, is not. You just, you get an answer, but you're still gonna have to spend all the same amount of time to confirm whatever it is that you came up with. So that because the stakes are so high and people's obligations to sign off on the information that's being shared is so high that you, you know, it's, to me, it's almost like, you know, autonomous driving, but something another level up, like autonomous flying or something, you know, for an airplane.
B
Yep. Nuclear information is one of those things you just don't want to be hallucinating on, for sure. Of all the things to be hallucinating on, aircraft safety, self driving, it's on that class of like, what is safety critical and how much rigor needs to be applied. So we started the company with that full knowledge that it will be very high stakes. I mean, we have many customer testimonials of people before Everstar and adopting our internal tool, Gordian, they've had cases where an error of misplacing a Semicolon costed them $30,000 for that one turnaround to work with consultants. So we knew the stakes were high coming into the industry, but that comes with lack of competitors, serious competitors, and the opportunity that comes along with it. So we took that very seriously. We apply similar concepts of defense and depth that the nuclear industry is very familiar with. You're not hoping for one bulletproof solution, but you're applying many layers of protection so that it'll get caught at some point down the road. So many advanced approaches that are pretty standard in high stakes AI today. So different levels of benchmarking and evals, QA testing, sampling internally with humans and making sure humans are in the loop to many layers of AI as judges and adversarial testing so that you can improve the rigor of the AI. So there's many layers we do internally and there's some layers we do with external folks too. We just never sell this as an automation tool. We sell this as productivity. You save quite a bit of time. The final arbiter of the decision data curation is you, the end user. Many layers of approvals needed to submit anything. So we sign off on those kinds of liabilities upfront. Not to say we're not Going to shoot for gold. We have proven that our system is now producing hundreds of pages of output at any given time in one in a single run. And the the accuracy of our AI system is at or above human level. In some cases, not all, but in many cases they're coming out as good as human prior outputs and sometimes catching issues with previous submissions. This is already happening with real life enterprise rollouts. So we're constantly dissatisfied with the outputs and pushing the boundaries. But that has led us to already getting to a very impressive level that's helpful to end users.
A
So if your software is successful, what does the nuclear industry look like in say the next. Normally I don't like to go out more than a couple years, but given the pace at which nuclear moves, I think that's probably too short a timeline. So let's say the next five to 10 years. What changes in the industry if, if more people adopt your software and how does that impact how, how the whether we get to 10x energy supply?
B
The administration has already put out aspirational timelines. New reactor qualifications and sites should be approved in 18 months. Renewals should be done in 12 months. That's very aggressive compared to where it is today. The folks that have proprietary data and documentation on parcels of land they want to site on, we should be able to aggregate much more data on weather patterns, seismic activity, soil quality, water access data. If all of those come together and we have pre built modules of calculations and agents that understand how to qualify sites, each site, each new one should be several days long to qualify that site instead of three years. Today could take as long as three years. That is a disruptive amount of transformation to be done in a single industry. Every day or every week delay on a multi billion dollar nuclear plant site, you could be saving every week shaved and generating electricity faster. You're saving about $60 million for a large one. So you know that's extremely large amounts of value that we'd be saving our customers and the nation by accelerating these build outs. The humans will still need to apply a lot of rigor to make sure the numbers that are fed into the AI are correct, that decision trees that happen within the AI system thinking are correct, that the correct precedents, the correct analogs, the correct regulation guidelines are applied. They should be spending a lot of time there. But I foresee that in good order, in due time we will be approximating those aspirational timelines.
A
So it's still Overall development timeline plus
B
or minus 10 years for completely new reactor designs. Yes, but our hot take contention is that we have plenty of safe vetted reactor designs that can just deploy tomorrow if you bring us the land, money and qualified labor. Westinghouse AP1000s are ready to go. They've proven extremely safe in America and extremely safe and reliable in China where they've been deployed. So those should be sub four years in approving new sites and sites it should be done in a year or so end to end with the help of AI and then full construction for commission to electrons on the grid. We want it to be eventually five, six years.
A
All right, well we definitely in today's market could use I don't know if it's 10x but we could definitely use a lot more supply to hit the market. So best of luck with your efforts here and we'll thank you watch and and see how the nuclear industry progresses here. We've had several guests on recently talking about adoption so it's definitely an area where people are putting renewed focus.
B
Yep, we're very excited about that and the role that AI has to play not just as the demand pull for nuclear plants, but as the enabler of the industry to meet the demands and timelines at the same time. So we're at the cutting edge of applying best tools to end users and a cross sectional view of the landscape and players that have to put existing reactors to up U.S. efficiency, new reactors that are trying to put novel technologies on the grid. We're helping all the above get to market in radically different timelines. So very excited to share that story with you today.
A
All right, thanks for joining us.
B
Thanks Todd.
A
You can find us online at www.projectfinance.law or send us an email at currentsordonrosefullbright.com Please rate referral, review and subscribe on Apple Podcasts, Spotify or your preferred podcast app. Our show today was produced by Emily Rogers. Stay ahead of the Currents.
Date: May 1, 2026
Host: Todd Alexander (A), Norton Rose Fulbright
Guest: Kevin Kong (B), CEO & Founder, Everstar
This episode dives into how AI is being leveraged to transform and accelerate approval and licensing processes for advanced nuclear projects. Todd Alexander talks with Kevin Kong, CEO of Everstar, about building a platform that enables more efficient, safer, and faster licensing of nuclear reactors, the hurdles companies face under current regulatory systems, and the potential impact of AI on the future of nuclear energy.
“A 40 ton pressure vessel might come with more than 40 tons of paperwork if you print out the documents.”
— Kevin Kong (04:46)
Augmentation, Not Automation:
How It Works:
“SMRs alone will not save the planet… if you’re building SMRs, likely you’ll have to target a lot of premium buyers. If you’re building MMRs, you’re replacing diesel generators.”
— Kevin Kong (11:43)
“Nuclear information is one of those things you just don’t want to be hallucinating on, for sure.”
— Kevin Kong (13:01)
Aspirational Timelines:
Changing Economics:
“We have plenty of safe vetted reactor designs that can just deploy tomorrow if you bring us the land, money and qualified labor.”
— Kevin Kong (18:18)
“Just been chasing that for a long time and I realized nuclear fission is the only viable pathway… fission been staring at us in the face.”
— Kevin Kong (01:25)
“A 40 ton pressure vessel might come with more than 40 tons of paperwork if you print out the documents.”
— Kevin Kong (04:46)
“SMRs alone will not save the planet… you get more power the larger you build it spread across linearly growing costs that scale with surface area. You just can’t beat that.”
— Kevin Kong (11:43)
“Nuclear information is one of those things you just don’t want to be hallucinating on, for sure.”
— Kevin Kong (13:01)
“Final arbiter of the decision data curation is you, the end user. Many layers of approvals needed to submit anything.”
— Kevin Kong (14:18)
“Every week shaved and generating electricity faster, you’re saving about $60 million for a large one.”
— Kevin Kong (17:14)
“We have plenty of safe vetted reactor designs that can just deploy tomorrow if you bring us the land, money and qualified labor.”
— Kevin Kong (18:18)
The conversation remains pragmatic, balancing optimism about AI’s potential with a clear understanding of the regulatory and technical challenges. Kevin Kong emphasizes that AI’s role is to enable qualified humans to work more efficiently, not to replace them, especially in a domain where safety and accuracy are paramount. The future painted is one where nuclear projects could move at unprecedented speed, provided new tools and collaboration are embraced across industry and government.