One Second of Fuel: David Kirtley on Fusion Power by 2028

After the intro, Lex Fridman asks David Kirtley to start with the basics: what is fusion, and what is fission? Kirtley is a nuclear engineer and CEO of Helion Energy, which is trying to put a commercial fusion plant on the grid. Fusion, he says, is what powers stars. It also powers plants, which become oil, which became the fossil fuels of the last hundred years. Humanity has been living on fusion the whole time. It just has not yet done the reaction on Earth to make electricity.

The reaction itself is simple in outline and fierce in detail. You take the lightest nuclei, hydrogen and its isotopes, and push them together until they form something heavier. The new nucleus weighs a little less than the parts. That mass defect is energy, E=mc², and the force that finally wins at nuclear distances is the strong force. Fission is the opposite move: uranium and plutonium are so heavy they are barely held together, and a single extra neutron can split them. Add up the fragments and they, too, weigh a little less than the original nucleus.

There is a famous curve in nuclear physics with iron at the hinge. Elements lighter than iron release energy when they fuse; heavier ones release energy when they split. Stars climb that curve. Ours is still burning hydrogen into helium; later stars burn helium into carbon and keep going until iron, after which fusion costs energy and the core collapses. Kirtley likes the implication: every coal plant is a delayed fusion plant, with a hundred million years of geology in the way. Einstein’s equation even hides in chemistry, he adds. Burn hydrogen and oxygen into water and the mass changes. The shift per molecule is almost unmeasurable, but it is still E=mc².

Fission fuel is scarce in a specific way. Uranium and plutonium were made in supernovae and the early universe; we dig them up, and most plutonium is bred from uranium. Fusion fuel is the opposite. Hydrogen is the most common atom in the cosmos, and Helion’s working isotope is deuterium: a proton plus a neutron. It is in seawater, in the glass Kirtley is drinking, in Coca-Cola, in human bodies. At today’s electricity use, he estimates seawater deuterium would last somewhere between 100 million and a billion years. Even at much higher demand there would still be millions of years left.

Getting those nuclei to fuse is the hard part. They are positively charged, so electromagnetism shoves them apart until they are close enough for the strong force to grab. Temperature is just kinetic energy, which is why terrestrial fusion wants on the order of 100 million degrees. The sun cheats with gravitational confinement: the mass of the star pulls the fuel into the core and holds it there. Earth cannot copy that. Fission, by contrast, happens at room temperature. "Fusion is actually really hard to do. You have to overcome those electromagnetic forces to have a single fusion reaction happen."

Kirtley then objects to a word the whole field uses. The U.S. Nuclear Regulatory Commission defines a reactor as "an apparatus other than an atomic weapon, designed or used to sustain nuclear fission in a self-supporting chain reaction." Fusion fails both tests: it is not fission, and it is not self-supporting. Take your hands off a fission core and it can keep going; stop feeding a fusion machine and it dies. Nobody calls a gas turbine a natural-gas reactor. He wants the same honesty for fusion: it is a generator. You put fuel in, you take electricity out, and when the fuel stops, so does the plant. What humanity actually wants, he adds, is not a set of nuclear reactions. It is electricity.

Fusion’s safety pitch is built into the physics. The products are charged particles, already a form of electricity, and the reaction is so hard that it turns itself off. Fission is a chain reaction in a water-cooled core, balanced so that each split makes just enough neutrons to continue, not enough to run away. Fridman notes that modern fission plants are empirically much safer than public fear implies. Kirtley, speaking as a nuclear engineer, agrees the engineering is largely solved. Cores are designed to expand and cool if they heat up. The trouble, he says, is the humans around the plant: how long you run a unit past its design life, and the fact that you had to enrich uranium to fuel it in the first place.

He puts Three Mile Island in its own bin. The station is being restarted because the United States needs clean baseload, which is itself a comment on the technology. Chernobyl and Fukushima he reads as human failure more than design failure. At Fukushima, several reactors on the same site rode out the tsunami and were later shut for political reasons; the oldest unit, perhaps run too long, was where the trouble was. Weapons are the other fission problem that fusion does not share. Put enough uranium or plutonium in a small volume and you get a supercritical bomb. Fusion plants cannot be used that way.

The so-called hydrogen bomb does not change the point. An H-bomb still has uranium in it. A fission primary makes radiation that lights a small fusion stage, which then boosts more fission. "Most of the energy, in fact 90% of the energy in an H-bomb, is all still from the uranium reactions themselves." Take away the fissile material and the fusion does not happen. Researchers have tried an all-fusion bomb for decades and failed; the physics, as understood today, does not support it. When Helion went to proliferation experts, Kirtley expected the word "nuclear" to scare them. They said the opposite: please build fusion as fast as possible, or the world will keep enriching uranium for baseload and spreading centrifuges.

The geopolitical follow-on is almost blunt. Deuterium is in seawater everywhere. No one can monopolize it, turn it off, or cut a pipeline. If Fridman interviews world leaders, Kirtley wants one question asked: what would you do with low-cost, clean, tens-to-hundreds-of-megawatts fusion that does not carry uranium and plutonium? He thinks that decoupling would ease a lot of energy tension. Fusion power, on this telling, is not only a climate technology. It is a way to stop building the fuel cycle that weapons require.

Fridman comes back to the simple question: are fusion power plants safe? Kirtley says yes, because the physics has no runaway. For the NRC they ran a deliberately absurd case: the plant is operating, a meteor strikes, the whole thing is vaporized. The result is that nearby people do not need to evacuate. The reason is fuel inventory. In a Helion generator, and in most fusion systems, there is about one second of fuel in the machine at any moment. Stop feeding it and fusion stops. Smash it and that deuterium just goes back into the environment. A coal pile can burn for a long time; a fission core can hold years of fuel. Fusion has a tank of heavy hydrogen that cannot fuse by itself.

The reaction still makes ionizing radiation: X-rays, neutrons, charged particles. The charged particles stay in the magnetic system; the X-rays are absorbed in the hardware, like a much fiercer dental office. Neutrons are the part you design around. Helion shields with borated polyethylene and concrete so operators and equipment are protected while the machine is on. Turn it off and the fusion radiation stops. That operating-shield problem is why U.S. fusion is not regulated like a fission reactor.

Reactors live under NRC Part 50, written around special nuclear material, uranium and plutonium. Fusion sits under Part 30, the same statute that covers hospitals, particle accelerators, and other irradiators. That split was written into law as the ADVANCE Act, which Kirtley calls the first clear U.S. rule for how to regulate fusion. He has a story from the paperwork. He took a shielding permit to the Washington State Department of Health, using the particle-accelerator form. The first question was where the patients go. There are no patients; no one is inside; the goal is electricity. As far as they know, Helion received the first U.S. fusion license of that kind in 2020, and has licensed several machines since.

Every fusion scheme is trying to do the same three things: heat light isotopes past 100 million degrees, pack enough of them into a volume (density), and hold them long enough (tau) that they collide and beat the losses. How you confine them is where the roads split. Public money in the United States has mostly funded two mainlines. Inertial fusion, usually with lasers, crushes a target for nanoseconds; colleagues at the National Ignition Facility recently set records doing that at scale. Magnetic fusion tries to hold charged particles on field lines as long as possible. Tokamaks and stellarators both live in that family. The stellarator is Kirtley’s graduate-school favorite because it has a clean mathematical solution. Building one is another matter. Humanity only recently fielded Wendelstein 7-X.

A magnetic field is just current in a coil. Helion’s systems run on the order of 100 million amps, not the 200-amp breaker in a house. Charged particles then gyro-orbit those field lines, the same way the aurora is trapped in Earth’s magnetosphere. The orbits are not nanoscale; they are measured in inches. The physics, Kirtley stresses, is old: nineteenth-century electromagnetism and early-twentieth-century atomic physics. Putting it all into a power plant is the hard part. A first-year grad student can write down a stellarator coil. Making the coil is the career.

Helion’s name for its middle path is magneto-inertial fusion: a bit of crushing, a bit of holding. The lineage starts with 1950s theta pinches, linear solenoids that let particles stream out the ends. One fix was the magnetic mirror, stronger fields at the ends so particles bounce; the hottest ones still leaked. The next fix was not to hold at all, but to squeeze by ramping the field. In 1958, at what became Livermore, federally funded teams switched millions of amps in microseconds with no transistors, no CPUs, and none of the switches Kirtley now takes for granted. They reached 50 million degrees, the best fusion of that moment, then hit a hardware wall and followed the newly invented laser instead.

Decades later, people running those pinches in a particular way saw something that should not have happened. Instead of squirting out the open ends like toothpaste, the plasma pushed back and stayed put. The accident is now called a field-reversed configuration, or FRC. You start with plasma riding a solenoid field, then reverse the current faster than the hot particles can move, on the order of a millionth of a second. The particles cannot flip with the field, so the plasma reconnects internally. Outside you still have a coil. Inside, current now runs in the plasma itself.

Once it exists, Kirtley says, an FRC is almost ordinary transformer physics. Lenz’s law says a changing current in a primary induces an opposite current in a nearby conductor. Here the secondary is not copper. It is a 100-million-degree gas. Current in a loop is an electromagnet, so the plasma makes a magnetic field and then sits in that field. "In an FRC, you make the plasma, which makes the magnets, and it traps itself." Nature already does this. A solar flare is an arc of current that tears off the sun and drifts as a plasmoid, holding together longer than a simple cloud of gas should. Researchers have watched those structures for a century. Helion’s job is to make one on command, every shot.

The 1958 teams could not reverse tens of millions of amps that fast. Semiconductor switching can. A CPU transistor already flips in a nanosecond; Helion needs that speed at power-electronics scale. Some days, Kirtley jokes, the company feels more like an electrical-engineering firm than a fusion lab. Making the FRC and keeping it stable are different problems, and they spend years on both. A solar flare looks beautiful until it flies away and vanishes. Fusion needs the structure to stay, on purpose, while you compress it.

Plasma beta is the ratio of particle pressure to magnetic pressure. Magnetic pressure is B²/2μ₀; particle pressure is nKT, density times temperature. In an FRC, beta is close to one by definition: push on the donut and the plasma pushes back equally. That is a gift, because the two fusion ingredients you care about, n and T, are then written directly in the magnetic field. It is also why textbooks say high-beta plasmas are unstable. A tokamak plasma is boxed in by coils. An FRC has no mechanical axis you can plant at 100 million degrees, so the whole donut can turn over. That mode is called tilt.

The analogy is a spinning top. Spin it slowly and it falls; spin it hard and angular momentum keeps it up. Drive the FRC’s ions fast enough and it stays. Geometry helps more. A spinning coin still falls; a roll of duct tape, thicker and longer, keeps going. Helion’s parameter is S*/E: S* for the kinetic, top-like stability, E for elongation. Long solenoids naturally make long plasmas, which is a rare gift from the hardware. The catch is that S*/E is also a temperature. You must satisfy it the whole shot, including while the plasma is still cold. Other groups, Kirtley says, lost FRCs because they could not heat them fast enough. The thing tilted before it got hot.

Basic theory says these objects should live a few microseconds. Helion and others now hold them for thousands of microseconds, thousands of times the naive lifetime. At 100 million degrees, "hot" stops meaning a room thermometer. You have left solid, liquid, and ordinary gas, passed through rarefied collisionless plasma, stripped electrons around 10,000 degrees, and arrived at a world that is mostly velocity. Particles move on the order of a million miles per hour, or meters per microsecond. You never let them touch a wall; they would blast material off it. You bounce them on magnetic fields until they hit each other instead.

Once you ask for fusion, Kirtley says, you have volunteered to live in microseconds. A shot is a flash: inject fuel, reverse, compress, recover energy, done, before the eye really responds. He is still amazed the 1950s pioneers controlled any of this without computers. Megahertz processors, which he used early in his career, can only tick once per microsecond. Gigahertz means a thousand operations in that same window. Humans are out of the loop. Operators pre-program a sequence, then programmable logic runs it.

The software stack is a museum and a factory at once. Some fusion codes are still Fortran. More and more of the work is Python, with some Java, and the fastest timing is assembly on FPGAs. Triggering cannot go over copper; the wires are too slow. Fiber optics carry photons and can answer in nanoseconds. The currents are huge. Even large transistors handle about 30,000 amps, so a 100-mega-amp pulse needs tens of thousands of switches firing together. Each switch has fiber in and fiber out. Rogowski coils, powered by the current they measure, flash an optical "I conducted" back to a central crate. If one device misses, the sequence has to stop.

Simulation sits in layers because computers still cannot do everything. A fluid MHD code, the fusion cousin of CFD, is driven by a SPICE model of capacitors, coils, cables, and switches; that is the daily design tool for which capacitor and how fat a cable. Deeper particle-in-cell codes treat ions as particles and need orders of magnitude more compute. GPU buildout for AI data centers has, conveniently, sped those runs up. They are how the team now explains the old empirical S*/E rules instead of only fitting them. A run still takes a day or two, so operators sweep parameters, then analysts compare. Kirtley wants reinforcement learning to put a prediction on the screen during the shot, not two days later.

The fusion figure of merit is n, T, and tau. High beta already sets nT from B², so the job is to maximize magnetic field. Fusion rate then rises even faster because hotter fuel fuses more. In these machines Kirtley quotes a scaling of B to the 3.75, sometimes 3.77: double the field and power jumps by about an order of magnitude. Pulsed magnets have been shown above 100 tesla; steady magnets, superconducting or copper, sit around 20 tesla, maybe the high 20s. The trade is tau. Lasers live at nanoseconds and extreme pressure; tokamaks and stellarators live at long tau and low density. Helion sits in the middle: very high field, hold for 100 microseconds to a few milliseconds, then exhaust and repeat about one to ten times a second.

A shot starts with field filling the chamber. Fuel is injected and ionized to a "cold" million degrees, hot enough to be plasma. The team forms an FRC and ramps the field from one tesla toward five, ten, twenty and beyond. Fusion is born as new charged particles inside the donut. Because beta is one, that extra pressure pushes the magnetic field outward, which means it pushes current back into the coils and into the capacitors that started the pulse. Steady fusion, in Kirtley’s picture, is a campfire: deuterium-tritium makes a helium nucleus that stays and a neutron that leaves, and you stand next to the fire to boil water at 30 to 35 percent efficiency. Helion is a piston. The expansion does work on the field and comes out as electricity.

Theory papers put that direct conversion around 80 to 85 percent, not because fusion is magically lossless, but because you skip the steam cycle. There is a second number he thinks may be the tail that wags the dog: the magnetic energy you invested can be recovered at over 95 percent, a figure already shown in experiments. You are not throwing away a blowtorch every shot. Deuterium-tritium is then a poor fuel for a high-beta machine, because the neutron does not push on the field. Helion burns deuterium and helium-3 instead. The products are an alpha particle and a proton, both charged, both trapped, both useful. The helium-3 nucleus is called a helion; that is the company name. The prices are real: optimal temperature is 200 to 300 million degrees, density falls as temperature rises so the machine grows, and helium-3 is scarce on Earth, though Jupiter has a lot and the Moon has some. Triple the electrical efficiency, Kirtley says, and the electricity per machine lands back at a similar size. Trenta, online in 2020, was the first system he knows of to do bulk deuterium–helium-3 fusion, and the first FRC to 100 million degrees.

Size matters because size is money. If manufacturing is good, Kirtley says, product cost asymptotes to materials: how much concrete, steel, copper, aluminum. "The cheapest function is the least amount of materials." Physics still sets a floor. Fusion will not fit on the back of a DeLorean. Make the machine too small and it does not make electricity, which wastes the whole cost argument. The way through, in his career, has been the opposite of a single mega-project: the smallest thing that can do the job, built quickly, so you learn. Rapid manufacturing, he found, is how you do the science faster. Helion has built seven systems. The first six were end-to-end prototypes for forming FRCs, compressing them, hitting thermonuclear temperature, then scaling and switching to advanced fuel.

Names started as a joke. The inductive plasmoid accelerator, IPA, also a beer, was the first to show the team could make FRCs and hold them. Machines were too small for a person in the photo, so Starbucks cups sat on top for scale in the Redmond shop: Tall, Grande, Venti, then Trenta in 2020. Early money was SBIRs and other government grants. The design rule was ordinary alloys, ordinary thicknesses, even of tungsten if you must use it. Making a hundred small magnets, each light enough for a person and a standard mill, beat one giant magnet that needs a crane and a rare CNC. When they built Trenta the company had about 50 people. It now has more than 500, and more than 300 of them never watched a machine go from a computer model to a fusion shot.

He is not sure he had said this in public before: they spend a lot of time on eBay. A new turbomolecular pump from the factory can take nine months. The same model from ten years ago can be at the dock next week. They buy three, test them, and maybe one meets spec. You do not go to eBay to save money, though it is cheaper; you go so you can work. The same filter applies to diagnostics. A 3-percent plasma instrument that takes three years and millions of dollars loses to a 5-percent one you can build in a month. Fifty percent would not be good enough; quality and safety are not for bargaining. Helion is an odd fusion company in another way: about 50 percent technicians, not scientists, with conveyor belts making power supplies in-house. On Trenta’s end cap, standard G10 fiberglass, the same stuff as circuit boards, was too small, so they bolted two sheets with nine fasteners rather than wait six to twelve months for a custom mold.

In 2023 Helion signed with Microsoft to build a plant that puts fusion electricity on the grid for a data center. Microsoft buys that power; there is no private wire into a hall of GPUs. The deadline is ambitious: first electrons in 2028. Microsoft had watched Grande, Venti, and Trenta hit milestones, scale by orders of magnitude, and move to advanced fuels, so the contract was not a slide deck. Siting, interconnects, environmental work, and regulation started two years earlier. Kirtley thinks about the date every day. People still call it a pipe dream. His answer is that there is no physics reason it cannot be done. The remaining questions are how fast you can build and how efficient you can make it. "If we're not discovering new hard problems, we probably didn't push hard enough."

Skeptics once said plasmas could not be merged, then that FRCs could not be compressed because they are unstable. He still hears the second objection, and still answers with S*/E and twenty years of work. When a machine first comes online it flashes pink. Fusion light itself is X-ray, invisible. What people see is the "ice-cold" million-degree start, a bright purple-fuchsia through ceramic and quartz vacuum vessels. Everyone, he says, changes their Windows desktop to that glow. High-speed cameras with filters can tell hydrogen from helium-3; a still camera only catches the integrated flash. Polaris, the seventh machine, is the one a few hundred new hires will watch turn from a simulation into that light.

Grid connection is not trivial, but the path is familiar. Recovered energy sits as high-voltage DC on capacitors, then inverters make 60-hertz AC the way a battery plant does. Because the machine is pulsed at one to ten hertz, output can be turned up and down with repetition rate. Smaller plasma systems have already gone from off to 100 hertz and run more than a billion shots. Data centers, though, want DC. The team is looking at stepping capacitor voltage straight to GPUs and skipping two conversion hops. The cooling problem rhymes too: Helion mostly cools semiconductors and power switches, the same class of hardware a data center cools. For AI, Kirtley argues, the cost of computation asymptotes to electricity, and that electricity wants to be dense and on site. Industry forecasts of 4 to 6 percent power growth look low to him. He does not want cognition limited by the grid.

Demonstrating fusion once, Kirtley tells the team, would be a failure. The goal is factories: 50-megawatt generators rolling off a line once a month, then a week, then a day, driven away on a truck. Helion already has competitors, which delights him, because that means there is a fusion industry. A White House Bold Decadal Vision asked how to start replacing some 4,000 gigawatts of fossil capacity, not how to publish one more shot. Data centers help the chicken-and-egg of deployment: you site once and park a hundred modules for a huge local load. That will take more than two conveyor belts.

Fridman asks when humans reach Kardashev Type I, matching the solar power incident on Earth, and then Type II, wrapping a star. Kirtley notes we are a fraction of Type I, and few fuel sources can get there. If Earth really holds 100 million to a billion years of fusion fuel at today’s use, then even 100 times today’s demand still has room. A 50-megawatt Helion plant, as he sketches it, fits in about 27,000 square feet, on the order of an acre. The same 50 megawatts of solar near Seattle would take about 2,000 acres. Other people jump to desalination. He thinks about food. So much former wild land is now five-foot-tall farmland. Why not 500 feet, stacked in a building, and give the surface back?

Space is how he got into fusion. Far from the sun, irradiance falls as distance squared and mass is expensive, so energy-dense fuel matters. A steam cycle is almost useless in vacuum: there is no coolant to dump heat into, only heavy radiators. Direct conversion is then not a nice-to-have. Before Helion spun out, the founding team worked on advanced materials, rocket propulsion, and fusion thrusters. Aerospace teaches you not to waste a watt, to recirculate water and air, to reject every joule of heat on purpose. They brought that ethic to electricity as the product. Fridman is haunted by the Fermi paradox, the fear that ambitious civilizations invent something that destroys them. Kirtley prefers Matrioshka brains: a Dyson-sphere cousin that wraps a star and spends the energy on thought instead of flags. AI plus fusion, he says, might be a start down that inward path. From outside, the star would simply go missing.

Asked for the most beautiful idea in the physics he lives with, Kirtley does not name an equation. "I'm continuously in awe that it works." The temperature window for life, the balance between electromagnetism and the strong force: those are hard to picture as accidents. Look at a leaf, then a cell, then an atom, then the quantum structure inside it, and the parts still fit. Fridman adds that humans somehow find the working point. Kirtley answers: it just works. Last minute sometimes, but it does work.