How Tokamak Fusion Reactors Actually Work (2026)
Understanding how tokamak fusion reactors work starts with an awkward fact: the same reaction that powers the Sun is, on Earth, fantastically hard to bottle. Stars fuse hydrogen almost lazily, held together by their own crushing gravity. We have no gravity to spare, so we cheat with magnets. A tokamak is a doughnut-shaped magnetic cage that holds a gas hotter than the core of the Sun — roughly 100 to 150 million degrees Celsius — just long enough for atomic nuclei to slam together and release energy. The trick is keeping that plasma away from the walls, stable, and hot, all at once. None of those three is easy. Doing them together, at a profit, is the grand challenge of fusion. Here we build the picture from the ground up.
What this covers: the D-T reaction, why magnetic confinement is needed, the field geometry inside a tokamak, the triple product and ignition, why net energy is so hard, and where things stand in 2026.
Fusion in one paragraph: why stars do it easily and we don’t
Fusion releases energy by joining light nuclei into heavier ones. The easiest reaction to ignite on Earth fuses two hydrogen isotopes: deuterium and tritium. They fuse into a helium nucleus plus a fast neutron, releasing about 17.6 million electron-volts of energy per reaction. That is roughly four times the energy per nucleon of uranium fission, from fuel you can extract from seawater and lithium.
Figure 1: The deuterium-tritium reaction. Two hydrogen isotopes fuse into a helium nucleus and a high-energy neutron, releasing 17.6 MeV.
So why is it hard? Both nuclei carry positive charge, and like charges repel. That repulsion is the Coulomb barrier, and it rises steeply as nuclei approach. To get close enough for the strong nuclear force to take over and bind them, the nuclei must collide at enormous speed. Speed, in a gas, means temperature. The deuterium-tritium reaction needs an ion temperature on the order of 100 million degrees Celsius before fusion happens fast enough to matter. The U.S. Department of Energy and the ITER Organization both describe this temperature threshold as the central obstacle: at it, matter is a fully ionized plasma, the fourth state of matter, no longer a neutral gas.
Stars solve this with sheer mass. The Sun’s core is “only” about 15 million degrees, far below our 100-million-degree target. It gets away with that because gravity squeezes its core to immense density and confines it for billions of years, so even slow, rare fusion events add up. We have neither that density nor that patience. The IAEA frames the engineering problem plainly: on Earth we must reach higher temperatures than the Sun, then hold the plasma together with something other than gravity. That something is a magnetic field.
It is worth pausing on why deuterium-tritium, of all possible reactions, is the one we chase first. Many light nuclei can fuse — deuterium with deuterium, or even ordinary hydrogen, as in the Sun. But D-T has the highest fusion rate at the lowest temperature of any candidate reaction. Its reaction cross-section peaks at a temperature we can plausibly reach. Other fuels, like deuterium-deuterium or proton-boron, need far hotter conditions and are off the table for a first generation of reactors. D-T is the easy reaction, and it is still brutally hard. That tells you how steep the climb is.
The fuel logistics are favorable, though. Deuterium is stable and abundant: about one in every 6,400 hydrogen atoms in seawater is deuterium, enough to power humanity for millions of years if we could harvest it. Tritium is the awkward partner — it is radioactive, decays with a half-life of about 12 years, and barely exists in nature. That single fact, as we will see, shapes the entire design of a fusion plant, because a reactor must manufacture its own tritium as it runs.
It is also worth being clear about how fusion differs from the nuclear power we already have. Today’s reactors run on fission, splitting heavy atoms like uranium. Fission and fusion sit on opposite sides of the binding-energy curve, the chart that ranks how tightly each nucleus is bound. Both release energy by moving toward iron, the most stable nucleus, but from opposite directions: fission by breaking heavy nuclei apart, fusion by joining light ones. The practical consequences are large. A fusion plasma holds only a tiny amount of fuel at any moment, so there is no critical mass and no chain reaction to run away. Stop feeding fuel, or let the plasma cool a fraction, and fusion simply stops. That inherent off-switch is one of the most appealing safety features of the whole approach, and it is a direct consequence of how hard the reaction is to sustain in the first place.
One more piece of intuition before we leave the reaction itself. The 17.6 MeV released per D-T fusion does not come out evenly. About four-fifths of it rides off with the neutron, and only about one-fifth stays with the helium nucleus. That split matters enormously for reactor design. The helium, being charged, stays trapped by the magnetic field and keeps the plasma hot — it is the self-heating that makes a burning plasma possible. The neutron, being neutral, escapes immediately and carries the bulk of the energy out to the walls, where it must be captured as heat to make electricity. So one reaction has to do two jobs at once: keep itself going with the helium, and pay the bills with the neutron. Designing a machine that does both well is a balancing act that runs through everything that follows.
The confinement problem and the tokamak idea
Here is the core dilemma. A plasma at 100 million degrees will vaporize any solid container instantly. No material survives contact. Yet the plasma is also far too thin to be held by walls in the ordinary sense — it is millions of times less dense than air. We do not need a strong container. We need a container made of nothing, that touches the plasma nowhere. Magnetic fields can do exactly that.
Plasma as the fourth state, and why magnets grip it
First, what is a plasma? Solids become liquids, liquids become gases, and gases, if you keep heating them, become plasma — the fourth state of matter. In a plasma, electrons have been stripped from their atoms. What is left is a soup of free electrons and bare positive nuclei — charged particles. Plasma is actually the most common state of ordinary matter in the universe; stars and interstellar gas are plasma. It is only rare in our everyday surroundings, which is part of why it feels exotic.
And charged particles feel magnetic fields. A moving charge in a magnetic field does not travel straight; it spirals tightly around the field line, like a bead threaded on an invisible wire. The stronger the field, the tighter the spiral. This is the whole basis of magnetic confinement fusion: if you can shape the field lines so they never lead into a wall, the particles riding them stay trapped. The field grips the charges, but the charges in turn distort the field — the two are locked together, which is exactly why plasma physics is so rich and so hard to predict.
The simplest closed shape for a field line is a circle. Bend the field into a loop and the particles chase each other around it endlessly, never reaching an edge — because, along the field, there is no edge. This is why fusion machines are shaped like rings rather than boxes. The plasma lives inside a vacuum chamber bent into a torus, a doughnut.
It helps to picture the scale of the spiral. In a strong field, an ion’s spiral is only millimeters wide, while the plasma itself might be meters across. So each particle is, in effect, glued to a field line, free to slide along it but barely able to wander across it. Confinement is never perfect — collisions nudge particles from one line to the next, and that slow sideways leak sets a limit on how long heat stays in. But the leak is slow enough to be useful, and making it slower is much of the art of tokamak design.
Toroidal and poloidal fields: the twist that makes it work
A single field running the long way around the doughnut — the toroidal field — sounds like enough. It is not. A pure toroidal field has a fatal flaw. The field is stronger on the inside of the doughnut hole and weaker on the outside, simply because the field lines are bunched tighter where the ring is tighter. That gradient causes the positive and negative particles to drift in opposite vertical directions. Charges separate, an electric field builds, and that electric field combined with the magnetic field flings the whole plasma outward into the wall. A simple torus leaks itself empty in microseconds.
Figure 2: A tokamak combines a toroidal field with a poloidal field. Together they twist field lines into helices that wind around the plasma, canceling the drift.
The fix is to add a second field that wraps the short way around the doughnut’s cross-section — the poloidal field. Add the two together and each field line becomes a helix, spiraling around the ring as it goes. A particle following a helical line spends part of its orbit on the top and part on the bottom, so the upward and downward drifts cancel out over a lap. The plasma holds together. This twisting is the defining idea of the tokamak, a word from a Russian acronym for “toroidal chamber with magnetic coils.”
There is a sibling approach worth a nod, because it shows the twist can be done differently. A stellarator achieves the same helical field entirely with elaborately shaped external coils, no plasma current required. That makes it naturally steady-state and disruption-free, at the price of fiendishly complex magnets that took decades of computing power to design. The tokamak made the opposite trade: simpler coils, but a reliance on plasma current that brings pulsing and disruptions. For most of fusion history the tokamak’s simpler magnets won out, which is why it remains the most developed design. The cross-section also need not be a perfect circle. Modern tokamaks squash the plasma into a tall “D” shape, which packs in more plasma and improves stability, and that shaping is part of why the divertor lives at the bottom.
Where the poloidal field comes from: the plasma current
Here is the elegant part. In a tokamak, the poloidal field is generated largely by the plasma itself. The machine drives a huge electric current — millions of amperes — flowing around the ring through the plasma. A current produces a magnetic field looping around it, and that loop is precisely the poloidal field needed. So the plasma is simultaneously the fuel, the conductor, and a source of its own confining field.
Figure 3: A simplified tokamak cross-section. External toroidal coils, a central solenoid, and poloidal coils shape the field while the divertor handles exhaust.
That current is usually kicked off by a central solenoid, a giant electromagnet down the doughnut’s central column. Ramping its current induces a voltage around the ring — the plasma acts like the secondary winding of a transformer — and that voltage drives the plasma current. This transformer trick has a catch we will return to: you cannot ramp a solenoid forever, so this method is inherently pulsed, not steady-state. Real machines wrap all this in several layers. There is a vacuum vessel, the toroidal field coils, poloidal shaping coils, and a divertor at the bottom to channel out heat and helium “ash.”
The coils themselves are worth a closer look, because they are where much of a tokamak’s cost and difficulty live. To bend a plasma carrying millions of amperes, the toroidal field coils must produce fields of several tesla — tens of thousands of times Earth’s magnetic field. Doing that with ordinary copper would waste staggering amounts of power in resistive heating, more than the reactor could ever produce. So modern large tokamaks use superconducting magnets, cooled to a few degrees above absolute zero, which carry current with no resistance at all. The result is a strange machine: a chamber holding plasma at 150 million degrees sits centimeters away from magnets colder than deep space. Managing that temperature gradient is one of the quiet triumphs of fusion engineering.
The divertor deserves a mention too, because it is where the plasma is, deliberately, allowed to touch a surface. Field lines at the plasma’s edge are steered so the outermost layer of particles spirals down into the divertor at the bottom of the vessel. There, specially cooled tiles absorb the heat flux and pump away the helium ash produced by fusion. The heat load on those tiles can rival the surface of the Sun, concentrated onto a narrow strip, which is why divertor design is an active research area in its own right.
Size, it turns out, is itself a confinement strategy. A bigger plasma loses heat more slowly, for the same reason a large pot of water cools more slowly than a teacup. The interior is further from the cold edge, so heat takes longer to leak out. This is why the traditional path to better confinement was simply to build bigger machines, and why ITER is enormous. The newer high-field approach offers an alternative lever: a stronger magnetic field grips the plasma more tightly, improving confinement without the sheer bulk. Roughly speaking, confinement performance scales steeply with field strength, so even a modest jump in tesla can substitute for a large jump in machine size. That trade-off — bigger versus stronger — is the central design argument running through fusion today.
Heating, the triple product, and ignition
Before a plasma can be heated, it has to be created and kept fed. A tokamak starts with a near-perfect vacuum, into which a small puff of hydrogen gas is admitted. A jolt of voltage strips the electrons from the atoms, turning the gas into plasma, and the magnetic fields take hold. Once burning, the plasma must be refueled continuously, because fusion consumes fuel and the helium ash must be flushed out. Modern machines do this by firing frozen pellets of deuterium and tritium deep into the plasma at high speed, like a hailstorm of fuel, so that fresh fuel reaches the hot core rather than just the cooler edge.
Driving a current through the plasma also heats it, because the plasma has electrical resistance. This is ohmic heating, the same effect that warms a toaster wire. It is free and gets you started, but it fades: the hotter a plasma gets, the less resistive it becomes, so ohmic heating stalls out well below fusion temperatures. You need more.
Two main methods top up the heat. Neutral beam injection fires beams of fast neutral atoms into the plasma; they ionize on entry and dump their energy through collisions, like injecting hot gas. Why neutral, when the plasma is held by magnetic fields? Because charged particles would be deflected by those very fields before reaching the core. Neutral atoms slip through, then ionize deep inside. Radio-frequency heating pumps in electromagnetic waves tuned to the frequencies at which ions or electrons naturally gyrate, so the plasma absorbs the energy resonantly — a microwave oven for nuclei. Combined, these push the plasma past 100 million degrees, far beyond what ohmic heating alone could reach.
It is worth picturing how a single experiment, called a shot, actually unfolds. The machine pumps down to vacuum, the magnets ramp up, and a puff of gas is ionized to form the plasma. The current ramps to its flat-top value, heating systems switch on, and for a few seconds to a few minutes the plasma sits at full performance while hundreds of diagnostics record its every twitch. Then everything ramps down in a controlled way, the magnets cool, and the team pores over the data before the next shot. A large tokamak might manage only a handful of these shots per day. Each one is a carefully staged event, not a continuously running furnace, and turning that staged event into uninterrupted operation is one of the leaps still to be made.
But temperature alone is not enough. To get useful fusion you need three things at once: a high enough density of fuel ions, a high enough temperature, and a long enough confinement time before the heat leaks away. Multiply those three together and you get the triple product — density times temperature times energy confinement time. This idea traces back to the physicist John Lawson, whose 1955 analysis gave the Lawson criterion for a self-sustaining reaction.
Figure 4: The energy balance. Heating power and fusion self-heating must exceed losses. Crossing the triple-product threshold enables ignition.
The triple product matters because it captures a race between heating and leaking. Fusion reactions deposit energy back into the plasma — the helium nucleus produced in D-T fusion stays trapped by the field and dumps its energy locally, helping keep the plasma hot. If the plasma is dense enough, hot enough, and well-confined enough, this self-heating can outrun the losses. When the fusion self-heating alone sustains the temperature with no external heating, the plasma has reached ignition.
A related and more practical yardstick is Q, the ratio of fusion power produced to heating power injected. Q greater than 1 means the plasma produces more fusion energy than the heat fed into it — scientific breakeven. This is not the same as a power plant turning a profit. A real reactor must also run its magnets, pumps, and cooling, and convert heat to electricity at maybe 30 to 40 percent efficiency. Engineering breakeven — net electricity out the door — demands a far higher Q, perhaps 10 or more for a commercial plant. The gap between these two breakevens is where much of the hard work still lives.
There is a nice intuition that ties Q back to ignition. A Q of 5 means the helium self-heating is starting to dominate the energy budget. Ignition is the limit where Q becomes effectively infinite: the plasma is fully self-heating and you could switch off the external power entirely. Yet most realistic plant designs do not actually aim for full ignition. They target a high but finite Q and keep a modest amount of external heating on hand, because a plasma that needs no outside input is also a plasma that is very hard to control. A little injected power is a useful steering wheel.
A subtle point trips up many newcomers. The famous “ten times hotter than the Sun” figure is real, but temperature is only one of the three terms in the triple product, and arguably the one we have already mastered. Tokamaks have reached and exceeded the required ion temperatures for years. The persistent difficulty is the product — holding that temperature at sufficient density for long enough at the same time. Push density too high and instabilities appear. Stretch confinement time and small leaks accumulate. The triple product is hard precisely because its three factors fight one another, and a reactor has to win all three battles in the same plasma, in the same instant.
One of the most important discoveries in this whole field was almost accidental. In 1982, researchers operating a tokamak in Germany found that above a certain heating power, the plasma spontaneously jumped into a new regime with far better confinement. They called it the high-confinement mode, or H-mode. A thin barrier forms near the plasma edge that sharply reduces the leak of heat and particles, roughly doubling the energy confinement time for free. Almost every reactor design today assumes H-mode operation. The catch is that the edge barrier is itself prone to bursts called edge-localized modes, which periodically dump energy onto the wall and must be tamed. It is a recurring theme in fusion: every improvement brings its own new instability to manage.
Why net energy is so hard: trade-offs and gotchas
Confining the plasma is a constant fight against itself. Plasmas are riddled with instabilities — the hot, current-carrying fluid wants to kink, wobble, and tear. A current flowing in a loop wants to writhe like a garden hose at full pressure. Pressure gradients want to bulge outward. Magnetic field lines want to reconnect and form islands that short out confinement. Each of these has a name and a theory, and each sets a limit on how far you can push the plasma before it bites back.
The worst case is a disruption, a sudden collapse where the plasma loses confinement in milliseconds, dumping its energy and current into the vessel. The released energy can melt or crack components, and the collapsing current can induce enormous mechanical forces in the structure. Worse, a disruption can spawn a beam of runaway electrons accelerated to near light speed, capable of drilling into the wall. Predicting disruptions early enough to soften them — increasingly with machine-learning models watching the plasma in real time — is one of the most active and consequential areas of fusion control research.
Then there are the neutrons. The D-T reaction’s energy mostly leaves as fast neutrons, which carry no charge and so ignore the magnetic field entirely. They fly straight into the walls. Over time these neutrons damage and weaken structural materials and make them radioactive through neutron activation — a real but manageable waste challenge, far shorter-lived than fission waste. Those same neutrons are also needed for tritium breeding. Tritium does not occur naturally in useful amounts. So reactors must wrap the plasma in a lithium “blanket” where escaping neutrons convert lithium into fresh tritium fuel. Closing that fuel cycle is unproven at scale.
The materials problem deserves more than a passing mention, because it may be the hardest of the lot. A fusion neutron carries around 14 million electron-volts, far more energetic than a typical fission neutron. When it strikes the reactor’s first wall, it knocks atoms out of place, builds up helium and hydrogen gas inside the metal, and gradually makes the structure brittle and swollen. Over a plant’s lifetime each atom in the wall might be displaced dozens of times. No existing material has been tested to the full neutron dose a commercial reactor would deliver, partly because we do not yet have an intense enough fusion-neutron source to test them in. Developing and qualifying these materials is a slow, expensive program that runs in parallel with the plasma physics, and it is a big reason that a working plasma is not the same as a working plant.
Figure 5: The main hurdles between a hot plasma and a working power plant — instabilities, disruptions, neutron damage, tritium breeding, and steady-state operation.
Finally, there is steady-state. The transformer that drives the plasma current works in pulses, but a power plant should run continuously. Sustaining the current without a transformer — through external current drive or self-generated “bootstrap” current — is essential and still being refined.
Stack these challenges and you see why net energy has stayed just over the horizon for decades. Each problem alone might be tractable. The difficulty is that they are coupled. A solution that improves confinement might worsen disruptions. A material that resists neutron damage might be hard to fabricate into a breeding blanket. A plasma shape that suppresses one instability can invite another. Fusion is less a single puzzle than a tightly knotted bundle of them, and progress comes from loosening the whole knot a little at a time rather than cutting any single strand.
Where we are in 2026
Fusion in 2026 is a field of real, reported progress and stubborn remaining gaps. ITER, the giant international tokamak under construction in France, remains the flagship effort to demonstrate a large burning plasma, though its assembly schedule has been revised and first plasma pushed out from earlier targets. On the science side, tokamaks elsewhere have reported steadily longer high-performance pulses and record values of the triple product over recent years.
Private fusion has also drawn enormous investment, with several companies reporting milestones toward compact, high-field tokamaks built on modern high-temperature superconducting magnets. These magnets let you reach strong fields in smaller machines, which the physics rewards.
To keep these claims in proportion, it helps to know what has genuinely been demonstrated over the years, framed as reported results. Tokamaks have produced multi-megawatt bursts of fusion power. They have reached the required ion temperatures. They have run plasmas for many minutes in advanced devices. They have set successive records for the triple product and for total fusion energy in a single pulse. Each of these is real and important. What none of them has done is the combination that defines a power plant: high gain, long duration, full fuel cycle, and net electricity, all in one machine, repeatedly and reliably. Progress is best read as a slow convergence of separately-proven pieces, not a single finish line already crossed.
It is also fair to ask why this has taken so long. Part of the answer is funding: for decades, fusion research was funded well below the levels its own roadmaps said were needed. Part is sheer difficulty, the coupled-problem knot described earlier. And part is that each generation of machine teaches lessons that only the next, larger or higher-field machine can act on. The encouraging shift in recent years is that private capital and better magnets have compressed that iteration cycle, which is why the field feels more alive in 2026 than it did a decade ago — even though no one can responsibly promise a date.
It is also worth naming what genuinely changed in the last decade. High-temperature superconducting tape matured from a laboratory curiosity into a manufacturable product, and that single materials advance is what makes today’s smaller, higher-field tokamak designs even conceivable. The bet behind many private efforts is that a stronger field shrinks the machine and the cost, turning fusion from a national-laboratory megaproject into something closer to industrial hardware. Whether that bet pays off is exactly what the next few years of reported results will test.
A note of honest caution: a widely cited 2022 net-energy result came from inertial confinement at a laser facility, not a tokamak, and counted energy reaching the fuel rather than total plant input. No tokamak has yet produced net electricity. Treat every “fusion is almost here” headline as a reported milestone, not a delivered power plant. The direction is genuinely encouraging; the timeline remains uncertain — and anyone quoting a confident “five years away” is selling optimism, not physics.
So what would real success look like, and how would you know it from a press release? The honest checkpoints are concrete. First, a tokamak that sustains a burning plasma where self-heating dominates, not just a brief flash. Then a machine that runs for long durations, minutes and beyond, in steady-state rather than short pulses. Then a demonstration of the full tritium fuel cycle, breeding as much tritium as it burns. Then materials proven to survive years of neutron bombardment. And finally, a plant that delivers more electricity to the grid than it draws, accounting for every magnet, pump, and cooling loop. Each of those is a distinct, hard milestone, and they have to be cleared roughly in order. When you read a fusion headline, the useful question is simply: which of these did they actually do, and which are still ahead? That single habit will keep you grounded amid the hype.
FAQ
How hot is the plasma inside a tokamak?
The deuterium-tritium fuel must reach roughly 100 to 150 million degrees Celsius — about ten times hotter than the core of the Sun. At that temperature matter exists as a fully ionized plasma. The heat is held off the walls entirely by magnetic fields, since no solid material could survive direct contact.
What fuel do tokamaks use?
Most reactor designs target deuterium and tritium, two heavy isotopes of hydrogen. Deuterium is plentiful in seawater. Tritium is rare and radioactive, so reactors plan to breed it inside the machine by capturing fusion neutrons in a lithium blanket surrounding the plasma.
What is the difference between Q greater than 1 and a power plant?
Q greater than 1 means the plasma yields more fusion energy than the heating power injected into it — scientific breakeven. A commercial plant must also power its magnets, pumps, and cooling, and convert heat to electricity inefficiently, so it needs a much higher Q to deliver net electricity to the grid.
Why a doughnut shape?
A torus lets magnetic field lines close on themselves so confined particles circulate endlessly without hitting an end. Combining a toroidal field with a poloidal field twists those lines into helices, which cancels the particle drift that would otherwise fling the plasma into the wall.
Is fusion power dangerous or radioactive?
Fusion cannot melt down: it needs continuous, precise conditions, and any fault simply stops the reaction. Fusion neutrons do make reactor structures radioactive through activation, but that waste is far shorter-lived than fission waste, and there is no long-lived spent fuel.
When will fusion power my home?
Honestly, no one knows. Real milestones are being reported, but no tokamak has yet produced net electricity, and the engineering between a hot plasma and a reliable plant is substantial. Be skeptical of any specific near-term commercial date.
Further reading
- How flow batteries actually work (2026)
- How optical lattice atomic clocks work
- Optical lattice clocks: strontium vs cesium (2026)
- External: ITER — What is a tokamak?
- External: U.S. DOE — DOE Explains: Tokamaks
Riju writes about the physics and engineering behind connected systems, digital twins, and the technologies shaping tomorrow. Learn more.
