How Nuclear Will Unlock Energy Abundance with Valar Atomics Founder Isaiah Taylor

2026-07-02 - 62 min - source - Read full transcript

Sarah Guo (host)Isaiah Taylor

Key insights

Nuclear stalled in the US after Three Mile Island not because the technology failed, but because a PR-mismanaged incident with zero deaths and zero public radiation dose killed public and political appetite, right as the US was also losing its large-scale civil-infrastructure building capability.

Taylor argues the industry needed to shift from a civil-infrastructure construction model (bridges, dams, highway systems) to an advanced-manufacturing model once building capability atrophied, but nobody executed that shift through actual hardware iteration - most 'nuclear startups' since have stayed paper-and-simulation companies.

nuclear-energy

The nuclear industry's central bottleneck for decades was a regulatory chicken-and-egg: startups need empirical operating data to satisfy the NRC, but the NRC's commercial-deployment framework assumes you already have a mature, proven system.

The industry's historical workaround was producing 'paper reactors' - precise modeling and simulation with no physical testing - because that was the only way to look credible to a commercial-deployment regulator without ever turning a plant on.

regulatory-pathways

A little-known Department of Energy authority, originally the testing-focused half of the old Atomic Energy Commission, lets the government authorize live reactor testing entirely outside the NRC's commercial-deployment process.

The DOE was briefly named the Energy Research and Development Agency (ERDA) after Congress split the Atomic Energy Commission into a commercial-deployment arm (the NRC) and a testing arm; that testing authority sat unused in legislation for roughly 40 years until Executive Order 14301 invoked it to bring three advanced reactors, including Valar's, to criticality on American soil by July 4th.

regulatory-pathways

Valar splits nuclear companies into two camps - those who treat nuclear as a hardware-execution problem and those who treat it as a design problem - and argues the design-optimizing camp is chasing the wrong goal.

Companies pursuing maximally sophisticated, high-performance reactor designs end up dependent on rare materials and supply chains that don't exist yet. Valar deliberately picked the 'Toyota Camry' over the 'Lamborghini': a simple, cheap, safe reactor that can be mass-produced beats a marginally more efficient one that can't scale, because cost falls with volume.

hardware-iteration

Advanced reactor safety should be designed around reducing the consequence of failure, not just reducing its odds, and Valar's safety case is built on assuming total system failure.

Traditional nuclear safety engineering pours effort into making a meltdown vanishingly unlikely, but 'odds are stochastic' - something unpredictable eventually happens. Valar's safety basis instead assumes every plant system has already failed and shows, through passive physics (core geometry, TRISO fuel, materials) rather than active cooling or operator action, that workers and the public still receive zero radiation dose.

reactor-safety-design

Nuclear component pricing is often inflated by an atrophied supply industry rather than by genuine engineering difficulty, and a small in-house team can undercut it dramatically.

An outside vendor quoted Valar $5 million and 2.5 years for a reactor protection system (RPS) module. After failed negotiation, a 5-person internal team built a working RPS in 6 weeks for about $400,000. Taylor says the vendor's CTO then tried to discredit Valar to industry contacts as unsafe for daring to build it themselves - which he frames as evidence of how threatened legacy suppliers are by real execution.

hardware-iteration

Valar invented its own custom concrete because no commercial product met its bio-shield requirements, cutting a three-month construction step to about 42 hours.

The concrete needed enough density to block gamma rays, enough strength to self-stack without rebar (rebar would become irradiated and produce nuclear waste) or grout, and the right atomic composition not to activate under irradiation. Two engineers in their early twenties spent about three weeks flying around the US collecting and acid-testing rock samples ('the rock hunt') to formulate it.

hardware-iteration

Valar finances reactor construction with venture equity rather than project or debt finance, arguing US venture capital's core skill - underwriting technology-execution risk - is exactly the risk nuclear now needs priced.

Taylor says the traditional nuclear-startup playbook of assembling paper designs and LOIs to attract debt/project finance has failed repeatedly because financiers are structurally risk-averse. Valar instead builds on equity balance sheet, betting that being years ahead on an operating track record becomes a moat that's effectively impossible for later, debt-seeking competitors to close.

nuclear-energy

Valar's 'gigasite' strategy is to build large-scale power on its own land and timeline rather than first negotiating multi-party site and permitting deals with hyperscaler customers.

Taylor argues that going to a customer first means getting bogged down negotiating site and permit details with 'too many parties.' Instead, Valar plans to put a gigawatt of power, land, and fiber on the ground on its own schedule and expects data-center demand to follow, prioritizing customers who want power delivered in a year over those still deciding on a three-to-four-year horizon.

ai-power-demand

Because energy demand is price-elastic, Taylor argues abundant cheap nuclear power creates an effectively infinite market on its own, independent of the current AI compute boom.

'Energy being a commodity, the demand is set by the price' - if Valar can get energy to one cent and then a tenth of a cent, new uses get invented at each price drop. AI's hunger for power is a visible current tailwind and source of public awareness, but the underlying thesis is that cheaper energy has always induced its own new demand throughout history, regardless of what technology is driving that decade's demand.

ai-power-demand

Taylor's 'hyper-techno-industrialism' thesis holds that as AI and robotics substitute for human labor, energy becomes the dominant cost input to nearly everything manufactured.

Any manufactured good reduces to three inputs - people, materials, and energy - but materials and machines are themselves the output of earlier factories that also ran on people, materials, and energy, so tracing the chain back far enough leaves only energy. As AI-directed robots replace the human-labor input, the cost of a physical good converges toward the cost of the energy used to make it, meaning radically cheaper energy (via nuclear) could make most physical goods radically cheaper too.

ai-power-demand

Companies

Valar Atomics - Taylor's company: builds mass-manufactured, hardware-execution-focused nuclear reactors. Recorded on-site at its Utah facility around the first-power milestone of its Ward 250 reactor.
Nvidia - Provided the Blackwell chip that Valar connected directly to its nuclear reactor, powering the first AI chip ever run on nuclear power and hosting Valar's website live from the plant.
SpaceX - Cited twice as the reference case for underestimated exponential progress: people doubted its launch-cadence and satellite numbers the way they now doubt nuclear scale-up, and it is invoked as the model for nuclear's still-missing 'Ford moment.'
Toyota - Used as the target archetype: Valar wants to build the 'Toyota Camry' of reactors - simple, cheap, safe, and mass-producible - rather than a more efficient but complex 'Lamborghini' design.
U.S. Nuclear Regulatory Commission (NRC) - The commercial-deployment regulator; described as unsuited to hardware iteration because it evaluates mature systems, creating a chicken-and-egg problem for startups that need operating data before they can even approach it.
U.S. Department of Energy (DOE) - Originally the testing-focused Energy Research and Development Agency (ERDA) spun out of the Atomic Energy Commission; its authority, invoked via executive order EO 14301, let Valar bring Ward 250 to criticality outside the NRC's commercial pathway.
Fort St. Vrain - Historical helium-cooled, graphite-moderated reactor (garbled in the ASR transcript as 'Fort St. Brain'/'Fortin the brain') cited as the source of known reliability lessons - helium circulator issues and graphite's tendency to absorb moisture - that informed Valar's design.

Techniques and frameworks

Tick rate - Valar's internal metric for time between successive hardware milestones (e.g., first atom split to second atom split). First criticality took 2 years 4 months from incorporation; the second took about 7 months; the stated goal is to compress the interval to minutes.
Consequence-based safety design - Valar's safety philosophy: reduce the severity of a failure's consequences (via passive physics, geometry, and materials) rather than only trying to drive down the odds of failure. Its safety case assumes every plant system has already failed and shows radiation exposure to workers and the public is still zero.
Gigasite strategy - Build large-scale power on Valar's own land and timeline rather than negotiating site/permitting deals with hyperscaler customers first, on the bet that cheap, available power will itself pull in data-center demand.
Verticalization on the critical path - Valar's default is to buy commodity components where possible, but to build in-house (concrete formulation, reactor protection system) whenever an external supplier's cost or timeline would bottleneck the reactor's schedule.

Summary

Sarah Guo interviews Isaiah Taylor, founder and CEO of Valar Atomics, on location at the company's Utah facility, timed to the first-power milestone of its Ward 250 reactor - the first advanced reactor built by a startup, and the fifth new nuclear device to make power in the US since 2000. Taylor's origin story is personal: his great-grandfather was a nuclear physicist on the Manhattan Project, and Taylor grew up assuming nuclear was a solved industry until he realized, disorientingly, that the US essentially stopped building reactors after the 1970s. He traces the stall to Three Mile Island - a meltdown that killed no one and dosed no one with radiation, but was mismanaged as a public-relations event badly enough to kill political appetite for decades, compounding with the country's separate loss of large-scale civil-infrastructure building capability.

The regulatory core of the episode is a chicken-and-egg problem: startups need empirical operating data to satisfy the NRC's commercial-deployment framework, but generating that data requires already running a plant, which the NRC's framework isn't built to permit for unproven designs. The industry's historical workaround was building "paper reactors" through modeling and simulation rather than physical testing. Taylor's key discovery was a dormant DOE authority: the DOE began as the Energy Research and Development Agency (ERDA), the testing-focused half of a split Atomic Energy Commission, and that testing authority sat unused in legislation for roughly 40 years. Executive Order 14301 invoked it to authorize three advanced reactors, including Valar's, to go critical on American soil by July 4th - the legal basis for Ward 250 operating outside the NRC's usual commercial pathway.

Valar's stated philosophy is that nuclear is a hardware-execution problem, not a design problem: rather than chasing the most sophisticated, high-performance reactor (which drags in rare materials and supply chains that don't exist), the company deliberately builds the "Toyota Camry" of reactors - simple, cheap, safe, and mass-producible - tracking its own progress with an internal metric it calls "tick rate," the time between successive hardware milestones. Its first criticality took two years four months from incorporation; its second took about seven months; the explicit goal is to compress that interval to minutes. On safety, Valar's approach is to design for consequence rather than odds: instead of just trying to make a meltdown vanishingly unlikely, its safety case assumes every system in the plant has already failed and shows, through passive physics rather than active cooling or operator intervention, that workers and the public still receive zero radiation dose - a claim the company was set to demonstrate live by scramming the reactor and cutting all power to its safety systems.

A recurring theme is how much of "nuclear is expensive" is actually an atrophied-supply-chain problem rather than a physics problem. Taylor describes being quoted $5 million and 2.5 years for a reactor protection system module, then building a working version in-house in six weeks for about $400,000 - after which the vendor's CTO reportedly tried to discredit Valar to industry contacts. A similar story covers Valar inventing its own bio-shield concrete formulation from scratch (dense enough to block gamma rays, strong enough to self-stack without rebar or grout, and inert enough not to become irradiated waste), cutting a normally three-month construction step to about 42 hours. Financially, Valar deliberately builds on venture equity rather than project or debt finance, betting that US venture capital's core skill at underwriting technology-execution risk is the right tool for nuclear now, and that an early operating track record becomes a moat competitors waiting on cheap debt can't close.

The episode closes on demand and vision. Taylor connected an Nvidia Blackwell chip directly to the reactor - the first AI chip ever run on nuclear power - and hosted Valar's website live from the plant as a demonstration. His deeper argument is that energy demand is price-elastic: AI compute is a visible current tailwind, but abundant, cheap nuclear power would create new demand at every price drop regardless of what's driving that decade's need for power, which is why Valar's "gigasite" strategy is to build power on its own land and timeline rather than wait on slow, multi-party customer negotiations. He ends on a broader thesis he calls "hyper-techno-industrialism": every manufactured good ultimately reduces to people, materials, and energy, but materials and machines are themselves the output of earlier energy-consuming factories - so as AI and robotics substitute for human labor, the cost of nearly everything converges toward the cost of the energy used to make it, meaning radically cheaper energy could make most physical goods radically cheaper too.

Notable Quotes

Note: the source transcript (podscripts ASR) carries no speaker diarization. Isaiah Taylor does the large majority of the talking in this on-location interview; quotes below are attributed to him based on content and context and should be treated as best-effort, not verified against audio.

"Nuclear has never had its Ford moment or its Tesla moment... We have nuclear reactors that are more manufactured than constructed, and they're extremely safe." - Isaiah Taylor

"If you can figure out how to make energy cheaper, you will have demand... I view Valor as having a fundamentally infinite market." - Isaiah Taylor

"There are totally fake costs from an industry that is just totally anemic. It doesn't know how to build anything anymore." - Isaiah Taylor

"We just run toward gunfire on the most complicated things every time." - Isaiah Taylor

"When we figure out AI and robotics, that allows us to do semi-autonomous manufacturing, energy will become the cost of all things - the cost of buying a thing will become the cost of energy used to make it." - Isaiah Taylor