David Kirtley: Nuclear Fusion, Plasma Physics, and the Future of Energy | Lex Fridman Podcast #485
Summary
Helion’s commercial bet is not merely to demonstrate fusion but to make high-efficiency electricity: its pulsed field-reversed configuration is designed so charged fusion products push against a magnetic field and recharge capacitors directly. Kirtley contrasts theoretical direct-conversion efficiencies of 80–85% with the 30–35% conversion of a steam cycle, while citing demonstrations that recover more than 95% of the input magnetic energy. His governing line is: “We’re not in this to make fusion. We’re in this to make electricity.”
A 2023 Microsoft agreement turns Helion’s physics thesis into a hard delivery target: “first electrons” from a grid-connected fusion plant in 2028. Microsoft would buy the plant’s electricity through the grid, and Helion says manufacturing, siting, interconnection, environmental, and regulatory work is already underway. Kirtley does not minimize execution risk—“It’s gonna be hard”—but argues there is “no physics reason this can’t be done,” leaving engineering, efficiency, and manufacturing as the decisive unknowns.
The claimed physics leverage comes from pulsed magnetic fields, because Kirtley says fusion output scales approximately as magnetic field to the 3.75–3.77 power. Pulsed magnets have demonstrated more than 100 tesla versus roughly the 20s for steady magnets; Helion trades some confinement time for that much higher field. Its self-organized plasma lasts roughly 100 microseconds to a few milliseconds, with stability engineered through the S* over E parameter rather than assumed away: “In an FRC, you make the plasma, which makes the magnets, and it traps itself.”
Helion’s preferred deuterium–helium-3 fuel improves electrical recovery but raises the technical bar from roughly 100 million degrees to an optimum of 200–300 million degrees. The reaction produces a charged proton rather than the neutron central to deuterium–tritium systems, allowing its energy to push back on the magnetic field. The trade-offs are lower density at a given field, potentially larger hardware, and scarce helium-3 that must be made or sourced elsewhere; Kirtley argues the roughly threefold conversion-efficiency advantage can restore comparable plant size on an electricity-output basis.
Safety, regulation, and proliferation could become material deployment advantages rather than merely public-relations claims. A Helion system contains about one second of fuel, so stopping fuel stops fusion; even its deliberately extreme “meteor strike” analysis did not require evacuating the surrounding population. Fusion still produces X-rays, neutrons, and activated materials requiring shielding, but US treatment under Part 30—alongside particle accelerators and medical equipment—differs from fission’s Part 50 framework, while Kirtley says proliferation experts urged rapid fusion deployment to reduce worldwide demand for uranium enrichment.
Helion’s execution evidence is seven generations of hardware and a manufacturing-led culture, not a single decade-long flagship experiment. Trenta reached 100 million degrees in 2020 and, “as far as we know,” performed the only bulk deuterium–helium-3 fusion; the company had about 50 people then and now exceeds 500, with roughly half the workforce technicians. Its operating philosophy includes modular magnets, vertically integrated power-supply lines, used eBay vacuum pumps delivered in two weeks instead of nine months, and accepting a 5%-accurate diagnostic available in one month over a 3%-accurate system requiring three years.
AI data centers are both the first demand wedge and a possible architectural match for pulsed fusion’s high-voltage DC output. Helion is exploring whether it can bypass AC conversion and feed GPUs more directly, while Kirtley thinks forecasts of electricity-demand growth rising from 2% annually to 4–6% may be “wildly underestimating” AI. The longer-term industrial target is a factory shipping 50-megawatt generators as often as one per day; Helion estimates each could fit in a 27,000-square-foot building, roughly an acre, versus about 2,000 acres of solar near Seattle.
Deep dive
1. Fusion and fission cash the same mass defect from opposite directions
Kirtley’s widest framing is that “fusion is what powers the universe.” Stars fuse lightweight hydrogen isotopes into heavier nuclei; plants capture stellar energy, fossil fuels inherit it, and much of civilization therefore already runs indirectly on fusion—just not yet through human-built electricity generators.
Fission starts at the periodic table’s heavy end. Adding a neutron to an unstable uranium-235 or plutonium-239 nucleus breaks it into a spectrum of fragments, releases more neutrons, and leaves products whose combined mass is slightly below the original nucleus; that missing mass appears as energy through E=MC².
Fusion combines light nuclei whose final mass is likewise below the sum of the inputs. Kirtley notes that mass-energy conversion even exists when hydrogen and oxygen chemically become water, though the mass change is extraordinarily small; nuclear reactions expose it at useful scale. Iron is his dividing point: lighter elements release energy by fusion, while heavier ones can release it by splitting.
2. Deuterium makes fuel abundance a structural advantage
Humanity largely obtains fission fuel by digging up uranium, then producing much of its plutonium from uranium. Kirtley describes those heavy nuclei as inheritances of the primordial universe and stellar catastrophes, unlike something a power system can obtain almost anywhere.
Fusion’s foundational material is hydrogen, the universe’s most common element. Helion principally uses deuterium: ordinary hydrogen’s proton plus one neutron, giving atomic mass two. It occurs throughout terrestrial water—including drinking water, bodies, and Coca-Cola—as D₂O or, more commonly, mixed HDO molecules.
Kirtley estimates seawater contains enough deuterium to supply humanity’s current electricity consumption for somewhere between 100 million and one billion years. Even at far higher consumption, he says the runway remains millions of years; Lex’s conclusion is that exhausting it would imply a civilization already capable of reaching other water-bearing worlds.
3. A 100-million-degree barrier makes fusion difficult—and self-terminating
Fusion nuclei carry positive charge, so the electromagnetic force repels them. Heating the fuel to around 100 million degrees supplies kinetic energy—velocity high enough for occasional nuclei to approach within nuclear dimensions, where the attractive strong force overtakes electromagnetic repulsion and permits fusion.
Temperature here is not conventional warmth but particle motion. Once two nuclei moving at extreme velocity fuse, the heavier product has slightly less mass and the difference emerges as energy; obtaining enough such collisions requires simultaneously adequate temperature, density, and confinement time.
Kirtley’s blunt distinction is that “fusion is hard and fission is easy.” Fission can proceed at room temperature and sustain a chain reaction; the Sun achieves fusion through gravitational confinement. An Earth-bound system lacks stellar mass, so engineers must create confinement with electromagnetic forces—and continuously maintain the exact conditions fusion needs.
4. Calling it a generator keeps the product and safety model honest
Helion avoids the word reactor because the NRC definition centers on apparatus sustaining nuclear fission in a self-supporting chain reaction. Fusion is neither fission nor self-sustaining: operators must keep supplying fuel and imposing the required fields, or the process ends.
“Generator” captures both the control model and the intended product. Kirtley compares it to natural gas equipment: fuel goes in, electricity comes out, and stopping the fuel shuts the machine down. Humanity does not ultimately want reactions or even heat; “what I really want is electricity.”
A conventional fission plant instead balances a neutron chain reaction, heat removal, and steam generation. Water generally absorbs heat and neutrons before driving a turbine, while operators must keep the reaction continuing without accelerating. That creates interconnected thermal, fluid, cooling, and control systems around a core containing substantial potential energy.
5. Modern fission may be engineered safe while its human system remains fragile
Lex’s pushback is worth keeping: public fear of fission looks disproportionate to empirical safety data. Kirtley agrees that modern plants can be engineered with passive responses—when temperature rises, components expand and the system naturally cools—plus multiple protection layers.
His distinction is between power-plant engineering and the humans surrounding it. A plant may be safe as designed, yet uranium had to be enriched, operators may extend equipment beyond its intended decades of service, and institutions must continually respect the assumptions embedded in the design.
Asked about Chernobyl, Fukushima, and Three Mile Island, Kirtley groups the first two around human failure more than unsolved plant physics and places Three Mile Island in a different category, citing plans to restart it because clean baseload power is scarce.
At Fukushima, his chosen example is that several reactors on the site survived the tsunami and were later shut for political reasons, while problems occurred at the oldest unit, perhaps after too long in service. It supports his narrower claim: plants operated as designed can be safe, but institutional decisions remain part of the system.
6. Fusion attacks both proliferation and fuel monopoly
Kirtley states categorically that fusion power plants cannot be used to make nuclear weapons. A hydrogen bomb still requires a fission primary and uranium reactions; he says roughly 90% of an H-bomb’s energy remains fission-derived, while efforts to create an all-fusion weapon have failed and are unsupported by today’s physics and engineering.
Helion expected proliferation specialists to distrust anything labeled nuclear. Instead, their response was: “Please, please go develop fusion power plants absolutely as fast as possible.” Their concern was that electricity demand would otherwise spread uranium enrichment and centrifuges worldwide, increasing the inventory and infrastructure that might later support weapons.
The geopolitical question Kirtley would pose to leaders is what they would do with tens or hundreds of megawatts of low-cost, clean, industrial-scale baseload power without uranium or plutonium—and how that would alter their view of the next 30 years.
Deuterium also breaks the logic of fuel chokepoints. Because seawater is distributed globally, “you can’t have a monopoly on the fuel”; no country can close a pipeline or cut another state off. Rapid deployment could therefore decouple electricity from control by a few resource-owning governments.
7. One second of fuel bounds the fusion worst case
Helion’s regulatory analysis began with an intentionally absurd scenario: assume an operating fusion plant is struck by a meteor and completely vaporized. Kirtley says the modeled consequence still did not require evacuating nearby residents, illustrating how little stored fusion potential exists at any instant.
A Helion machine and most fusion systems contain about one second of fuel. Interrupt the feed and fusion stops; destroy the finely tuned apparatus and the deuterium simply returns to the environment. A storage tank cannot fuse by itself because achieving fusion requires the entire magnetic and electrical system.
The comparison is inventory, not perfection. A coal plant can leave a large pile burning and emitting toxic fumes, while a fission core may hold years of fuel. Fusion’s hazardous process is intense during operation, but its instantaneous fuel inventory is tiny.
Fusion still creates ionizing radiation: charged particles, X-rays, and neutrons. The vessel contains charged products and absorbs X-rays, while concrete and borated polyethylene shield escaping neutrons and limit activation. Kirtley says the NRC framework places fusion under Part 30, used for accelerators and hospitals, rather than fission’s Part 50; he also points to a landmark NRC agreement codified into US law the previous year as the ADVANCE Act. Helion’s first state particle-accelerator license came in 2020, after officials initially asked, “Where do the patients go?”
8. Every fusion architecture trades density, temperature, and time
All approaches pursue the same three requirements: heat particles beyond roughly 100 million degrees, gather sufficient density in a volume, and confine them long enough to collide. Architectures differ in which variable they maximize and what mechanism supplies pressure or confinement.
Laser inertial fusion crushes fuel with extremely powerful pulses lasting nanoseconds. Kirtley credits the National Ignition Facility with recent world records demonstrating that this can produce fusion at scale for a billionth-of-a-second interval.
Magnetic systems pursue longer confinement. Current flowing through large coils creates fields that trap charged particles in gyro-orbits—measured in inches for fusion particles, not at atomic dimensions. Helion’s systems carry hundreds of mega-amps, compared with a household breaker box around 200 or 400 amps.
Tokamaks and stellarators bend the magnetic channel into a torus so particles circulate instead of escaping from open ends. Kirtley loves the stellarator because its mathematical solution is clean, while construction is brutally difficult; he points to Wendelstein 7-X as evidence that humanity can now build the required geometry.
9. Helion revives the theta pinch with switches the 1950s lacked
Early theta-pinch experiments used a straight solenoid, but particles streamed out its open ends. One branch bent the solenoid into a donut, eventually yielding tokamaks and stellarators; another strengthened the end fields into magnetic mirrors so particles bounced between them.
Mirrors heated plasma, but pressure drove the hottest—and most valuable—particles out first. Researchers then rapidly increased the magnetic field to squeeze the plasma rather than holding it indefinitely, raising density and triggering fusion before the fuel could escape like toothpaste from a tube.
By 1958, California researchers in programs that fed into Livermore were switching millions of amps in microseconds and reaching roughly 50 million degrees, outperforming competing fusion systems. The transistor was still a laboratory object, so theta pinches hit a technological ceiling; research shifted toward newly available nanosecond lasers.
Decades later, experiments found that a carefully operated pinch did not squirt plasma from the ends. Rapidly reversing the magnetic field caused the plasma to reconnect internally and form a closed topology despite the open linear vessel—the accidental discovery now called a field-reversed configuration, or FRC.
10. An FRC makes its own magnet and traps itself
Field reversal must occur before the hot particles can respond, on the order of one millionth of a second. Modern semiconductor switches make that possible: the original magnetic field flips, but the plasma current cannot rearrange instantly, so it reconnects into an internally closed structure.
Kirtley explains the mechanism through Lenz’s law and a transformer. Current in an outer conductor induces equal-and-opposite current in an inner conductor; replace the inner metal with charged plasma, and massive current flows through the plasma itself.
A current loop creates a magnetic field, so the plasma becomes its own electromagnet and confines itself on the field it generated. His clean comparison: tokamaks and stellarators make magnets that trap plasma, while “in an FRC, you make the plasma, which makes the magnets, and it traps itself.”
Nature produces related plasmoids in solar flares: current-carrying arcs detach from the Sun and remain coherent temporarily. The analogy also exposes the problem—natural plasmoids eventually disperse—so Helion’s difficult work is making formation repeatable and the self-organized structure controllably stable.
11. High beta buys pressure while inviting the plasma to overturn
Plasma beta compares magnetic pressure with particle pressure. In an FRC, beta is near one, making the relationship B²/(2μ₀) = n k T especially useful: a specified magnetic field directly constrains the combination of fuel density and temperature needed for fusion.
High beta means that every external magnetic push meets a strong internal response, which later enables electricity recovery. It also makes the complete plasma topology unstable: like the rotor of a motor without a mechanical shaft, the donut can simply tilt over. No physical support can survive inside a 100-million-degree plasma.
Helion designs around a parameter called S* over E, combining a hybrid kinetic stability measure with elongation. Kirtley’s analogy is a spinning top: more angular inertia keeps it upright; a long roll of duct tape is still more stable than a thin coin because both speed and geometry resist overturning.
S* over E must remain satisfied throughout formation and compression, and it also tracks temperature because hotter particles move faster. FRCs often failed by tilting before heating sufficiently; Helion’s long geometry and fast electronics aim to cross that unstable region. Where elementary theory predicts only a few microseconds, experiments have maintained FRCs for thousands of microseconds.
12. At 100 million degrees, control is measured in microseconds
Matter first moves from solid to liquid to gas, then becomes rarefied as particles collide less often. Around 10,000 degrees, electrons separate from nuclei and create charged plasma; at 100 million degrees, temperature is most intuitively understood as individual particle velocity.
Those particles travel on the order of one million miles per hour—roughly 100 kilometers per second. Collision with a wall would damage it and blast material away, so magnetic fields must keep the fuel from touching physical components while still making nuclei collide with one another.
At meters per microsecond, the machine must measure and influence events on the same scale. A fusion shot is a flash: Helion ionizes fuel to a “nice, hot one million degrees,” forms the FRC, compresses it, performs fusion, recovers energy, and ends before the human eye can really respond.
Megahertz processors could barely react once per microsecond; gigahertz computing allows roughly 1,000 operations in that interval. Helion therefore pre-programs shots through programmable logic and FPGAs, using Fortran, Python, Java, and assembly-level instructions, with fiber optics carrying triggers in nanoseconds because electrical wires are too slow.
13. Tens of thousands of switches turn a shot into a managed event
A single large semiconductor switch handles about 30,000 amps, while the complete system may move 100 million amps. Helion therefore coordinates tens of thousands of parallel switches “all operating in harmony,” making electrical architecture and manufacturing as central as plasma physics.
Operators begin with a numerically predicted sequence for fuel injection, field reversal, and compression. Once triggered, computers execute the shot while continuously checking whether each switch fired correctly; deviations can stop the sequence before an asymmetric or mistimed system damages hardware.
Every switch receives and returns optical status signals. Rogowski coils powered by the current being measured report conduction optically to a central system, giving Helion nanosecond-scale evidence that the programmed electrical event actually occurred.
Cameras provide a more intuitive view. Ordinary cameras integrate the event into one purple or fuchsia flash from the roughly million-degree startup plasma, while high-speed cameras and wavelength filters distinguish emissions from hydrogen, helium, or helium-3. The hottest thermonuclear phase radiates in X-rays and is invisible to human eyes.
14. Simulation links SPICE circuits to particle-scale stability
Helion’s broad magnetohydrodynamic model treats ions and electrons as fluids, effectively computational fluid dynamics plus electromagnetic equations. It starts with a SPICE representation of capacitors, switches, cables, and coils, then uses that circuit behavior to drive the plasma model.
Engineers use those simulations at literal component level: which capacitor and switch to install, how many cables are required, and how thick each cable should be. More detailed models then optimize magnet shape, field topology, and the precise sequence for triggering adjacent coils.
Hybrid particle-in-cell codes treat ions as particles and expose behavior that fluid approximations miss, including S* over E stability. They require orders of magnitude more computation, but GPU advances driven by AI data centers have made these calculations practical only in the last few years and helped explain stability rules once known mainly through experiments.
The remaining bottleneck is latency: advanced runs can take one or two days. Operators therefore perform simulations, run parameter sweeps, collect machine data, and compare afterward. Kirtley is beginning to explore AI and reinforcement learning that could surface predictions closer to real time, but presents that as work underway, not a solved control loop.
15. B³·⁷⁷ is the pulsed system’s scaling thesis
Fusion performance depends on density n, temperature T, and confinement time tau. Because an FRC’s high-beta relationship ties nT to B², Kirtley says the clearest design instruction is to maximize magnetic field; increasing it raises density and temperature together, while higher temperature further accelerates fusion reaction rates.
He gives the resulting scaling as magnetic field to approximately the 3.75 power, and in many demonstrations 3.77. That exponent makes pulsed magnets attractive: researchers have exceeded 100 tesla in pulsed fields, while steady systems have reached roughly 20 to the high-20-tesla range.
The price is confinement time. Laser inertial fusion operates for nanoseconds; tokamaks and stellarators seek much longer duration at lower density. Magneto-inertial fusion lives between them, with high field and FRC lifetimes from about 100 microseconds to a few milliseconds—long enough, Helion argues, to produce useful reactions.
During a shot, the field may rise from one to five, ten, twenty, and still higher levels, compressing and heating the plasma. Newly created charged products increase n k T and push outward; that pressure strengthens the external field, drives current back through the coils, and recharges the capacitors that initiated the pulse.
16. Direct conversion changes the metric from heat to electricity
Conventional deuterium–tritium fusion follows Kirtley’s “match and campfire” analogy. Tritium is rare and unstable, but at roughly 100 million degrees it fuses readily with deuterium, producing a charged helium particle that remains in the plasma and a neutron that escapes with energy.
Tokamaks and stellarators use that neutron energy as heat, boil water, run a steam turbine, and convert roughly 30–35% of the extracted thermal energy into electricity. The quoted efficiency is specifically heat-to-electricity conversion, not a full accounting of the energy used to start and sustain the plasma.
An FRC resembles a piston rather than a campfire. Charged fusion products expand against the confining field; their pressure induces electrical current, allowing the machine to recover energy electromagnetically rather than routing all of it through heat exchangers, turbines, and cooling towers.
Kirtley cites theoretical direct-conversion efficiencies of 80–85%, plus demonstrations recovering more than 95% of input magnetic energy before adding the new fusion yield. High beta may therefore be “the tail that wags the dog”: the decisive advantage is not only making fusion, but recovering both the startup electricity and fusion-product energy efficiently.
17. Helium-3 raises the physics bar to improve the electrical product
Helion favors deuterium–helium-3 because the reaction produces an alpha particle and a proton instead of the neutron central to deuterium–tritium fusion. Kirtley emphasizes that the proton is charged, remains trapped in the magnetic field, pushes back on it, and can contribute directly to electrical recovery. The helium-3 nucleus is also called a helion, giving the company its name.
Helium-3 is stable but scarce on Earth because its low mass lets it escape the atmosphere. Kirtley says it must be manufactured or obtained beyond Earth, mentioning lunar proposals and “massive amounts” on Jupiter without presenting off-world sourcing as Helion’s near-term operating plan.
The physics price is temperature: deuterium–helium-3 reacts at 100 million degrees but performs optimally around 200–300 million. At fixed B² = nT, raising temperature lowers density, leaving fewer particles to react and potentially requiring a larger machine for comparable reaction rates.
Kirtley’s counter is to measure the machine by electricity, not raw fusion energy. If direct conversion captures roughly 80-some percent instead of 30-some percent and recovers input energy, the higher-temperature machine can end up approximately the same size for equivalent electrical output. That remains the architecture’s integrated physics-and-economics wager.
18. Manufacturing speed is part of Helion’s scientific method
Kirtley argues that low-cost power converges toward material cost: concrete, steel, copper, and aluminum impose a floor no manufacturing process can beat. Smaller machines therefore matter economically, though fusion has a physical minimum—“a fusion generator on the back of your DeLorean” is not supported by known physics.
Helion has built seven generations, with the first six serving as end-to-end scaling prototypes. Names moved from beer—the inductive plasmoid accelerator, or IPA—to Starbucks sizes: Tall, Grande, Venti, and Trenta. Trenta came online in 2020, reached 100 million degrees, and produced what Helion believes was the only bulk deuterium–helium-3 fusion.
The company treats rapid, complete builds as a faster route to science than one multinational megaproject. Early systems used government and SBIR grants, while each iteration increased field, temperature, and energy. “The faster you can build a thing,” Kirtley says, the faster the team can discover whether it works and construct the next one.
Speed shapes every component choice: standard aluminum and copper, available tungsten thicknesses, used turbomolecular vacuum pumps bought on eBay and tested in-house to arrive in about two weeks instead of nine months, 100 human-liftable magnets instead of one crane-moved custom magnet, and bolted pieces of commodity G10 fiberglass rather than waiting six to 12 months for a bespoke mold. The constraint forces both manufacturability and new engineering.
19. Microsoft’s 2028 deadline ties the machine to a real grid
Helion signed its Microsoft agreement in 2023 for a grid-connected fusion plant targeting first electrons in 2028. Microsoft would purchase its power through the grid rather than receiving a dedicated wire, an important distinction from the deeper direct-DC integration Helion is exploring.
Microsoft had watched Helion progress through Grande, Venti, and Trenta, hit milestones, scale fusion by orders of magnitude, and access advanced fuel. Helion says it began manufacturing preparation, siting, interconnection, environmental work, and regulatory planning years ahead of delivery.
Kirtley’s honest risk statement is that unknown hard problems will appear: “If we’re not discovering new hard problems, we probably didn’t push hard enough.” He rejects claims that FRC compression is physically impossible, pointing to S* over E and prior machines, but refuses to understate the remaining engineering and manufacturing difficulty.
Recovered energy sits on capacitors as high-voltage DC, which established inverters can turn into 60-hertz grid power. Pulses can run roughly one to ten times per second and be dialed with demand; a smaller plasma system reached 100 hertz and accumulated more than one billion operations during steady all-day running.
20. Commercial success means factories—and uses beyond today’s grid
Data centers offer a concentrated first market because computing cost ultimately “asymptotes to the cost of the electricity.” Helion is studying direct DC delivery to GPUs, potentially avoiding AC conversion losses, while sharing semiconductor and cooling infrastructure. Kirtley considers forecasts of annual power growth rising from 2% to 4–6% because of data centers “probably wildly underestimating” the trajectory.
Asked when the world gets 1,000 fusion plants, Kirtley gives no year. Instead, he says one demonstration would be failure and points to roughly 4,000 gigawatts of installed fossil capacity: the objective is a Gigafactory shipping truckable 50-megawatt generators monthly, weekly, then daily, with a data-center campus absorbing perhaps 100 identical units.
Energy density broadens the use case. Helion estimates 50 megawatts can fit in a 27,000-square-foot building, around an acre, versus roughly 2,000 acres of solar near Seattle. Kirtley imagines abundant power enabling desalination, 500-foot vertical farms that return surface acreage to nature, and microwave-beamed propulsion that launches craft without carrying conventional fuel.
Beyond Earth, direct conversion matters because steam systems need heavy radiators to reject heat, while sunlight falls with distance squared. Kirtley closes with a more optimistic Fermi-paradox answer: advanced civilizations may build Matrioshka brains, using stellar-scale power to expand cognition rather than territory. AI and fusion could move humanity that way; amid the deadlines, his deepest reaction remains, “I’m continuously in awe that it works.”