How Tokamak Fusion Reactors Actually Work: The Physics of Confining a 150-Million-Degree Plasma

Keywords: how tokamak fusion reactor works, tokamak physics, magnetic confinement fusion, plasma confinement

Reading time: 20 minutes | Word count: 5,800

Introduction: The Binding Energy Problem

Fusion energy—the process that powers the sun—is humanity’s most challenging energy problem. It’s not fusion itself that’s hard: at 150 million degrees Kelvin, atomic nuclei collide so violently that the strong nuclear force overwhelms the Coulomb repulsion keeping them apart. The problem is confinement: how do you hold a substance hotter than the sun’s core inside a container made of ordinary materials?

A tokamak is the world’s most developed magnetic confinement fusion reactor. It solves this problem through an elegant geometric and electromagnetic architecture: by curling a plasma into a torus (doughnut) and surrounding it with precisely tuned magnetic fields, a tokamak can hold 150 million degree plasma away from the walls for seconds—or, in recent achievements, minutes. This post explains the complete physics: why we need fusion, why a torus shape is necessary, how the magnetic fields work together, where the instabilities come from, and how modern superconductors and AI are finally making fusion engineering viable.

Part 1: Nuclear Fusion—Why It Matters and Why It’s Hard

The Binding Energy Curve: Where Fusion Lives

Every atomic nucleus has a binding energy—the energy that holds protons and neutrons together against the electromagnetic force pushing them apart. The binding energy per nucleon (proton or neutron) is highest around iron-56. Lighter nuclei (hydrogen, deuterium, tritium) are less tightly bound; heavier nuclei (uranium, plutonium) are also less tightly bound. This asymmetry is the entire basis of nuclear energy.

When two light nuclei fuse—say, deuterium (²H) and tritium (³H)—the products (helium-4 and a neutron) sit higher on the binding energy curve. The difference in binding energy is released as kinetic energy: about 17.6 megaelectronvolts per deuterium-tritium reaction. That’s why fusion is such a rich energy source: fusing 1 kilogram of deuterium-tritium releases energy equivalent to burning 11,000 kilograms of coal.

The challenge is the Coulomb barrier. Both deuterium and tritium have positive nuclear charges; they repel each other electrostatically. At room temperature, thermal motion gives nuclei a few electron-volts of energy. To overcome the Coulomb barrier, nuclei need to be at least 10 kiloelectronvolts apart—which corresponds to a temperature of about 100 million Kelvin. At the sun’s core, gravitational confinement is enough. On Earth, we must artificially confine the plasma.

Why Tokamaks: The Magnetic Confinement Philosophy

A charged particle in a magnetic field experiences a Lorentz force perpendicular to both its velocity and the field direction. This force causes the particle to spiral around the field line. If the field line is closed (forms a loop), the particle is geometrically trapped—it can’t escape perpendicular to the field without crossing many field lines.

A tokamak creates a closed magnetic geometry—a torus. A plasma confined in a toroidal magnetic field can, in principle, be held indefinitely because particles spiraling around the field lines never reach the boundary; the boundary is infinitely far away along the field lines.

This is conceptually simple but physically intricate. We need to explore three things:

Why a torus (and not a simpler geometry)
How the magnetic field inside is structured (toroidal, poloidal, helical)
Why the plasma still leaks and how to minimize it

Part 2: The Magnetic Mirror Problem—Why Linear Devices Fail

To understand why tokamaks are toroidal, we must first understand why linear devices don’t work.

A linear mirror is a straight tube with strong magnetic fields at both ends. Particles spiraling around the field lines in the center should be reflected by the stronger field at the ends (the “magnetic mirror effect”). In principle, this confines plasma without the geometric complexity of a torus.

Historically, magnetic mirrors were built and tested extensively (Pastukhov at Kurchatov Institute, US programs at LLNL and PPPL). They work remarkably well at containing individual particles—but they fail catastrophically at confining pressure. The problem is subtle and fundamental: while individual particles bounce between the mirrors, the collective plasma develops unavoidable drifts perpendicular to the field.

Why Pressure Is the Bottleneck

The key insight is understanding drift velocity. In a non-uniform magnetic field (strong at the mirrors, weak in the center), a particle with guiding center at the center experiences a net force. The drift velocity is proportional to the pressure gradient and inversely proportional to the field strength:

$$v_{\text{drift}} \sim \frac{\nabla p}{qBn}$$

For a 10 keV plasma at density 10^20 m^-3 in an 8 Tesla field, this drift is on the order of kilometers per second—much faster than the Alfvén wave speed that could dynamically stabilize the plasma.

Particles with certain velocity distributions—specifically, those with a pitch angle near the trapping boundary—escape over the mirror’s “loss cone.” More critically, the collective plasma acts as a pressure-driven fluid. The pressure gradient drives outward drift faster than magnetic reflection can hold particles back. Historically, fusion researchers achieved plasma parameters in mirrors, but they always hit a confinement threshold beyond which the plasma rapidly cooled.

The Beta Limit: A Fundamental Problem in Linear Geometry

This failure is quantified by the beta limit—the ratio of plasma pressure to magnetic pressure:

$$\beta = \frac{P_{\text{plasma}}}{B^2/(2\mu_0)}$$

Linear mirrors have a theoretical beta limit of roughly 5–10%. This is not an engineering problem; it’s a consequence of magnetic geometry. Even a perfect mirror with infinite field strength at the ends cannot hold plasma at higher beta.

Tokamaks, by contrast, can achieve β ~ 5% in standard designs, and modern advanced scenarios push toward 10% or higher. How is this possible if the physics is the same?

The tokamak’s geometric advantage is elegant: the plasma pressure is directed tangentially (around the major circumference of the torus), so it doesn’t directly push radially outward. Only the small toroidal drift (∇B drift, related to the gradient in toroidal field) interacts with the confining field, and it’s much smaller than in a linear device. The torus also has continuous rotational symmetry, which prevents systematic particle loss over a single “loss cone.”

This geometric insight—closing the loop into a torus—is why tokamaks dominate fusion research. The pressure-confinement problem is moved from a fundamental physics limitation (unavoidable in linear geometry) to an engineering challenge: keeping the plasma centered using poloidal field coils and feedback control. This is hard, but it’s not impossible.

Part 3: The Tokamak Geometry—Toroidal, Poloidal, and Helical Fields

A tokamak is defined by its magnetic field structure. Three field components work together:

1. The Toroidal Field (B_T)

The toroidal field is the primary confinement field. It runs around the major circumference of the torus, like circles of latitude wrapping around the Earth. In a standard tokamak, the toroidal field is generated by massive external coils arranged symmetrically around the torus.

This field is strong: typically 2–6 Tesla in present devices, and 11.75 Tesla in ITER. The stronger the field, the tighter the plasma is confined and the smaller the device can be for a given fusion output.

Why toroidal? The toroidal field alone cannot confine plasma against pressure. It prevents radial (outward) drift, but plasma still drifts vertically. You need a second field component.

2. The Poloidal Field (B_P)

The poloidal field runs like circles of longitude, perpendicular to the toroidal direction. It is induced by a large electric current flowing through the plasma itself. A central solenoid at the core of the torus acts like the primary winding of a transformer; the plasma is the secondary. When current flows through the solenoid, it induces an electric field in the plasma, driving a plasma current of millions of amperes.

The poloidal field is much weaker than the toroidal field—typically 10–50 times smaller—but it is essential. It provides the force that pinches the plasma inward, keeping it centered away from the walls. Without the poloidal field, the plasma would drift outward and hit the vessel.

3. The Helical Field: Toroidal + Poloidal

When you superimpose these two fields, they create a helical or spiral magnetic field. Plasma confined by this helical field has particles spiraling around field lines in a corkscrew trajectory. The pitch of the helix (the ratio of toroidal to poloidal field strength) determines how many times a field line wraps around the minor circumference before completing one loop around the major circumference.

This ratio is called the safety factor (q). Formally:

$$q(r) = \frac{B_\phi(r)}{B_\theta(r)} \cdot \frac{r}{R}$$

where B_φ is the toroidal field, B_θ is the poloidal field, r is the minor radius, and R is the major radius.

In the core, q might be 1.0; at the edge, q might be 3–4. A field line with q = 3 wraps around the minor circumference three times before closing on itself. This winding is crucial: it prevents certain plasma instabilities and determines how shear in the velocity profile stabilizes kink modes.

The safety factor also determines the rotational transform (ι = 1/q)—a measure of how quickly the field line rotates as you move from the center to the edge. Positive transform creates shear, which is inherently stabilizing: if a perturbation grows in one region, it’s sheared apart by the changing twist at different radii. This is why tokamaks carefully profile their q, typically maintaining q(0) ~ 1 at the core and q(a) ~ 3–4 at the edge. If q drops below 1 at any radius, the plasma becomes unstable to kink modes and confinement is lost.

Visual: The Nested Flux Surfaces

A perfect tokamak plasma is bounded by nested magnetic flux surfaces—topologically distinct sets of field lines that don’t cross. Each surface is a toroid, smaller and smaller as you move inward. Think of them as Russian dolls made of twisted field lines.

When a charged particle moves, it stays on its flux surface (ideally). The surfaces themselves don’t rotate or deform. This frozen-flux condition is the foundation of tokamak stability.

The diagram above shows a schematic tokamak with the toroidal field, poloidal field, nested flux surfaces, and the resulting helical field line trajectory. The key insight: the poloidal field pinches the plasma inward and couples with the toroidal field to create the helical structure that confines hot plasma away from the walls.

Part 4: Plasma Heating—From Ohmic to ICRH and ECRH

A cold plasma carrying current experiences resistive heating, like a wire heating up from electrical resistance. This is ohmic heating. Early tokamaks relied on this, but it’s fundamentally limited: as the plasma heats, its resistivity drops, so ohmic heating becomes less efficient. No tokamak has achieved the 100+ million Kelvin temperatures needed for fusion ignition using ohmic heating alone.

Modern tokamaks employ three auxiliary heating methods:

1. Neutral Beam Injection (NBI)

A neutral beam injector fires a beam of fast neutral atoms (typically deuterium) into the plasma at energies of 50–150 keV. These atoms are ionized by collisions with the hot plasma and slow down through Coulomb interactions, depositing their energy into the plasma. NBI is very efficient and has high power density. Most present tokamaks use NBI.

The physics: a 100 keV deuteron has a kinetic energy equivalent to 1.2 billion Kelvin. When it thermalizes (loses energy to the plasma), that energy heats the bulk plasma.

2. Ion Cyclotron Resonance Heating (ICRH)

An ICRH system launches radiofrequency (RF) waves at the cyclotron frequency of a specific ion species (e.g., hydrogen-1). When this RF wave resonates with the cyclotron motion of the ions, it accelerates them perpendicular to the field, heating the plasma.

ICRH is power-efficient and can be tuned to heat specific species. It’s used in many tokamaks and is planned for ITER.

3. Electron Cyclotron Resonance Heating (ECRH)

ECRH launches RF waves at much higher frequencies (gyrofrequency of electrons, ~100 GHz). It directly heats electrons, which then transfer energy to ions through collisions. ECRH has excellent spatial localization—you can heat a specific location in the plasma—making it valuable for controlling plasma profiles and stabilizing instabilities.

Modern tokamaks use a combination of all three methods to reach ignition-relevant temperatures.

Part 5: The Lawson Criterion and the Triple Product

Fusion requires not just hot plasma, but a balance of density, temperature, and confinement time. The Lawson criterion quantifies this balance. In its modern form, it’s expressed as the triple product:

$$n \cdot \tau \cdot T \geq \text{constant}$$

where:
– n is the plasma density (number of particles per unit volume)
– τ (tau) is the energy confinement time (how long heat stays in the plasma before leaking out)
– T is the plasma temperature (in Joules or keV)

For deuterium-tritium (D-T) fusion with a burning plasma at ~10 keV, the triple product must exceed roughly 3 × 10²¹ m⁻³ s·keV (exact values depend on reactivity assumptions and whether alpha heating is included).

The beauty of the triple product is that it’s relatively temperature-insensitive. Empirically, the product scales as T^(-1/3), meaning you can achieve the same fusion rate at lower density and longer confinement time or at higher density and shorter time. This flexibility allows different reactor designs to compete: a small, dense, short-pulse reactor (like SPARC or Commonwealth’s designs) can achieve ignition with the same triple product as a large, diffuse, long-pulse reactor (like ITER).

Confinement Time: The Bottleneck

The challenge is that confinement time scales poorly with reactor size and parameters. Historically, early tokamaks followed Bohm scaling:

$$\tau_E^{\text{Bohm}} \sim \frac{a^2}{\nu_e} \sim a^2 \cdot \frac{T^{3/2}}{n}$$

This was pessimistic—it predicted confinement time would decrease with increasing density, opposite to what’s needed. Later experiments showed Alcator scaling (named after MIT’s Alcator tokamak):

$$\tau_E \sim n \cdot a^2 \cdot B$$

Modern tokamaks achieve even better empirical scaling (ITER-89 scaling, IPB98 scaling), which depends on plasma shape, q, and heating power. The current standard is roughly:

$$\tau_E \sim I_p^{0.85} \cdot B_T^{0.2} \cdot P^{-0.73}$$

where I_p is plasma current, B_T is toroidal field, and P is heating power. The key point: confinement time scales roughly as a², and only weakly with field strength. A 10× increase in minor radius gives ~100× improvement in confinement time, but a 10× increase in magnetic field gives only ~1.6× improvement.

This scaling constraint has profound consequences. ITER is large (R = 6.2 m, a = 2.0 m) specifically to achieve the confinement time needed for the triple product. Smaller, high-field designs like SPARC (R = 1.85 m, a = 0.57 m) compensate for worse confinement scaling with 20 Tesla fields, achieving the same triple product in 1/10th the volume.

Part 6: Plasma Instabilities—The Barriers to Confinement

A confined plasma is fundamentally unstable. Even small perturbations can grow and destroy confinement. Understanding and mitigating these instabilities is core tokamak physics. Plasma instabilities fall into several categories: MHD instabilities (affecting bulk plasma), kinetic instabilities (affecting particle distributions), and turbulent instabilities (driving anomalous transport).

1. The Kink Instability (n=1 External Mode)

Imagine the plasma column shifts slightly outward at one point. The poloidal field at that spot is now stronger on the plasma side and weaker on the opposite side. This imbalance creates a net force pushing the plasma outward further, amplifying the perturbation. This is the kink instability.

The kink mode has a simple physical picture: the plasma “kinks” like an unstable rope. Mathematically, it’s an n=1 MHD mode (m=1 helical perturbation) where the mode grows exponentially with a growth rate:

$$\gamma_{\text{kink}} \sim \omega_A^2 \cdot (q – 1)$$

where ω_A is the Alfvén frequency. The safety factor q controls whether the mode grows or decays. If q > 1 everywhere, the term (q−1) is positive, which actually appears in the stability criterion in a subtly different form—the Kruskal-Shafranov criterion requires q > 1 for stability. If q < 1 anywhere, instability develops exponentially. This is why tokamak operators maintain q > 1 in the core, and why the core safety factor is typically designed to be near or slightly above 1.

2. Ballooning Instability (Pressure-Driven)

The ballooning instability is more subtle. The plasma pressure gradient and the curvature of the magnetic field compete. In regions of “bad curvature” (where the field lines curve away from the plasma center, like the outer equator of a torus), the pressure gradient can drive an outward bulge of plasma, like a balloon inflating. These perturbations grow on high-beta (high-pressure) equilibria.

Physically, the ballooning condition compares the destabilizing effect of pressure gradient (which wants to expand) against the stabilizing effect of magnetic shear. The instability growth rate is:

$$\gamma_{\text{balloon}} \sim \omega_A \cdot \alpha \cdot \sqrt{s}$$

where α is the normalized pressure gradient (proportional to β) and s is the magnetic shear. A sufficiently steep pressure gradient or weak shear triggers the instability.

Ballooning stability sets a beta limit: you cannot compress the plasma pressure beyond a threshold relative to the magnetic pressure. The theoretical no-wall limit is β < (4 μ₀/B²) × (pressure), which typically allows β ~ 3–5% in standard tokamaks. Modern tokamaks operate near the ballooning limit—they’ve learned to balance on the edge of instability by carefully shaping the pressure profile (broader in the core, steeper at the edge) and using magnetic shear to stabilize the pedestal.

3. Edge Localized Modes (ELMs)

At the edge of the plasma, the temperature and density gradients are very steep. This steep gradient region is called the pedestal. On the pedestal, a new instability emerges: Edge Localized Modes (ELMs).

ELMs are violent, localized bursts of energy and particles that escape from the edge region in milliseconds. They release accumulated plasma energy in discrete chunks—like a pressure relief valve opening and closing. While ELMs seem destructive, they also regulate density and impurities by expelling edge material. Modern tokamaks intentionally trigger ELMs to control the plasma state.

However, uncontrolled ELMs can damage the divertor (the material at the bottom of the reactor where plasma particles escape). This is a major challenge for ITER and future reactors.

4. Disruptions: Sudden Confinement Loss

A disruption is a catastrophic, sudden collapse of the plasma current and magnetic confinement. Within 10–100 milliseconds, the plasma loses all its energy and confinement is lost; the plasma column cools rapidly and expands outward, striking the vessel wall. Disruptions dump massive thermal energy (500 MJ in ITER) and electromagnetic stress into the structure in microseconds—equivalent to a controlled explosion.

Disruptions are triggered by a variety of underlying instabilities:

Tearing modes (m=2, m=3): Magnetic field line reconnection at rational surfaces (where n/m matches local q). These grow slowly but can trigger secondary instabilities.
Vertical instability: The plasma column becomes vertically unstable and drifts toward the top or bottom of the vessel.
Density limit disruptions: As density increases, turbulent transport increases, stored energy decreases, the plasma cools, and confinement is lost catastrophically.
Beta-limit disruptions: Ballooning instability triggers a mode that grows and disrupts the current.

The sequence is usually: a precursor instability grows for tens of milliseconds, then a trigger (often a secondary mode) crosses a threshold, and confinement collapses in one final burst of magnetic reconnection.

Predicting and preventing disruptions is one of the most important challenges in tokamak control—large devices like ITER could be damaged by even one uncontrolled disruption. This is where AI has made recent breakthroughs: DeepMind and PPPL have trained neural networks that predict disruption precursors 300–500 milliseconds in advance, giving the control system time to adjust parameters and avoid the event entirely.

Part 7: Magnetic Field Topology and Superconductors

A tokamak cannot function without extraordinarily strong and stable magnetic fields. Generating and maintaining these fields is an engineering feat.

Superconducting Magnets: Why They Changed Everything

Classical copper magnets, cooled to liquid nitrogen temperatures (77 K), have resistivity but not zero resistance. Running large currents through them generates enormous heat. For a tokamak to operate for minutes or hours, you need superconducting magnets—materials with exactly zero electrical resistance below a critical temperature.

For decades, tokamaks used low-temperature superconductors (LTS): niobium-titanium (NbTi) alloys, cooled to 4.2 K in liquid helium. NbTi is reliable and well-understood, but the operational ceiling is around 8 Tesla. Above that, the superconductor quenches (loses superconductivity).

The game-changer has been high-temperature superconductors (HTS), particularly Rare-Earth Barium Copper Oxide (REBCO). REBCO can operate at 10–11 Tesla or higher while being cooled to only 20–30 K (much cheaper than liquid helium). More importantly, REBCO’s magnetic field density is much higher—you can achieve 20+ Tesla with HTS magnets.

This is why SPARC (Commonwealth Fusion Systems) and other private fusion companies are using HTS: the stronger fields allow smaller, cheaper reactors while maintaining the same fusion power. A compact tokamak with 20 Tesla fields can be far more economical than a massive ITER-scale device with 8 Tesla fields.

The Magnet Systems

A tokamak has several magnet systems:

Toroidal field coils (TF coils): 18–24 massive D-shaped coils arranged around the torus
Poloidal field coils (PF coils): Located inside and outside the torus to induce the plasma current and provide vertical stability
Central solenoid (CS): The transformer primary that drives the plasma current
Trim coils: Small coils for fine tuning the field

Each system must be precisely aligned and synchronized. In ITER, the toroidal field coils alone cost billions and took years to qualify.

Part 8: Plasma Control and Feedback

A tokamak plasma is fundamentally unstable and must be actively controlled. A naive approach—running fixed coil currents and RF power—would result in a disruption within seconds. Instead, tokamak operators use feedback control: measurements of the plasma state (position, shape, density, temperature) feed into real-time algorithms that adjust coil currents and heating power to stabilize the plasma.

Classical Control: PID and Model-Based Schemes

Historically, tokamak control used proportional-integral-derivative (PID) feedback and model-predictive control. The plasma position is measured by magnetic sensors around the vessel. If the plasma drifts outward, the poloidal field coil currents are adjusted to push it back inward. If the plasma is too cold, NBI or RF power is increased.

This works, but it requires careful tuning and good physics models. When new plasma regimes are explored, controllers must be redesigned.

AI and Reinforcement Learning: The Game-Changer

In 2021–2022, DeepMind and Princeton Plasma Physics Lab (PPPL) demonstrated that deep reinforcement learning (DRL) could learn tokamak control policies that outperform classical controllers. This was a watershed moment for the field.

The DeepMind approach, published in Nature in 2021, trained a neural network on thousands of simulated tokamak discharges using the TORAX code (a fast, reduced-physics tokamak simulator). The network receives 90 simultaneous measurements (plasma shape, position, density, temperature, current profile measurements) and produces outputs: voltages for 19 poloidal field coils, updated 10,000 times per second. The training objective was to maximize confinement time while avoiding disruptions and maintaining desired plasma profiles—a multi-objective optimization problem that classical controllers struggle with.

The resulting policy—a deep neural network mapping observations to control actions—was then deployed on the TCV (Tokamak à Configuration Variable) tokamak in Switzerland. The AI controller successfully stabilized a wide variety of plasma shapes and configurations—including exotic “snowflake” plasmas (with four X-points instead of two) and “negative triangularity” configurations (where the plasma is concave rather than convex)—without being explicitly programmed for those shapes. This generalization ability was remarkable: the network had learned a general principle of tokamak confinement, not memorized specific scenarios.

PPPL separately developed AI models that predict plasma instabilities—specifically, tearing mode instabilities—up to 300 milliseconds in advance. This prediction window is crucial: it’s long enough for the controller to adjust parameters (poloidal field currents, heating power) and prevent the instability before it develops into a disruption. PPPL’s models were trained on experimental data from EAST, ITER simulations, and machine learning pipelines that identify instability precursors in time-series data.

The significance: AI transforms tokamak control from a reactive, tuned-for-a-specific-scenario problem into a proactive, generalizable system. If this approach scales to future large tokamaks (ITER, BEST, SPARC), it could reduce the iteration time for developing new operating scenarios from weeks to hours.

Part 9: Current Tokamak Landscape and Recent Breakthroughs

The tokamak is no longer purely a scientific experiment. Major projects are moving toward demonstration of net fusion power and commercial operation.

ITER (France, Under Construction)

ITER (International Thermonuclear Experimental Reactor) is the world’s largest tokamak under construction in Cadarache, France. Specifications:

Major radius: 6.2 m
Toroidal field: 11.75 T
Fusion power output: 10 MW (target)
Plasma volume: ~840 m³
Superconductor: NbTi (LTS)

ITER’s revised schedule targets first plasma in 2033–2034, with deuterium-deuterium (D-D) operations. Full deuterium-tritium (D-T) operations would begin later. ITER is a demonstration machine: it’s designed to prove that a tokamak can achieve sustained, controlled fusion reactions on a net-power-relevant scale. It is not designed to produce electricity; that comes after ITER.

Status (2026): ITER is in the midst of major construction milestones. The first toroidal field magnet qualification and assembly is ongoing. The project has faced delays and budget overruns, but remains the centerpiece of international fusion research.

EAST (China, Operating)

China’s Experimental Advanced Superconducting Tokamak (EAST) has become the world’s leading facility for sustained plasma confinement records:

2025 achievement: Sustained plasma for 1,066 seconds (17.7 minutes) in a steady-state-relevant mode
2026 achievement: Reached line-averaged electron density of 1.3–1.65 × 10²⁰ m⁻³, breaking the conventional density limit and accessing a new “density-free regime”

These achievements are remarkable because they were predicted theoretically but never realized until now. The high-density regime allows tokamaks to operate at higher fusion gain (more reactions for a given confinement time). EAST’s success validates the physics and engineering needed for future commercial plants.

China’s broader goal: Build the BEST (Burning Plasma Experimental Superconducting Tokamak) to demonstrate net fusion power gain and electricity generation by 2030. BEST will be larger than EAST but smaller and cheaper than ITER, using lessons from EAST.

WEST (France, Operating)

The WEST tokamak (Tungsten Environment in Steady-state Tokamak) holds the tokamak plasma duration record: 22 minutes and 8 seconds (1,327 seconds) achieved in February 2025. WEST is smaller than ITER or EAST but designed for steady-state operation with advanced wall materials (tungsten).

This record is crucial: it proves that tokamaks can sustain plasma for long durations without disruption, a prerequisite for commercial power plants.

NSTX-U (USA, Restarting in 2026)

The National Spherical Torus Experiment (NSTX-U) at Princeton is not a conventional tokamak; it’s a spherical tokamak—a torus with a very small major radius relative to the minor radius, making it look more like a squat sphere than a donut. Spherical tokamaks have some advantages: higher beta (pressure ratio), better confinement scaling, and potentially lower cost.

NSTX-U was shut down for upgrades and is scheduled to restart in 2026. Recent results from similar spherical tokamaks (Mast-U in UK, KTMQ-1 in China) show that spherical tokamaks may have even better confinement than conventional tokamaks, making them attractive for commercial fusion.

Commonwealth Fusion Systems (SPARC, Under Construction)

Commonwealth Fusion Systems (CFS), an MIT spinoff, is building SPARC—a compact, high-field tokamak using REBCO superconductors:

Major radius: 1.85 m (very compact)
Toroidal field: 20 T (using HTS)
Design goal: Achieve Q > 1 (net fusion gain) in 2027
Fusion output: 70+ MW for 10-second pulses

SPARC is revolutionary because of its size and cost. By using 20 T HTS magnets, CFS can achieve fusion-relevant conditions in a reactor one-tenth the size of ITER, at a fraction of the cost. If SPARC succeeds (achieves Q > 1), it would be a watershed moment for fusion—proof that net gain is feasible in a compact, privately-funded device.

Follow-on: If SPARC succeeds, CFS plans to build ARC (Affordable, Robust, Compact), a 400 MW commercial fusion power plant to be built near Richmond, Virginia, targeting the early 2030s.

2026 Status: CFS installed the first of 18 toroidal field magnets in January 2026. Construction is on track. CFS is also partnering with Nvidia and Siemens to build a digital twin of SPARC for control optimization and training AI controllers.

Part 10: The Physics-to-Engineering Gap

Understanding tokamak physics is one thing; building a working reactor is another. The gap between theory and practice is where engineering breakthroughs matter.

Key Engineering Challenges

Magnet reliability and quench protection: Superconducting magnets must operate stably for years without quenching. A quench—sudden loss of superconductivity, often triggered by a small resistive region or mechanical disturbance—causes the magnet to rapidly heat and can destroy the coil through Joule heating and mechanical stress. ITER’s magnets require redundant quench detection systems and fast discharge circuits that can dump the magnet’s energy (~80 GJ total in ITER) into external resistors within seconds.
Plasma material interaction (PMI): The plasma touches the divertor—the material exhaust system at the bottom of the reactor—at extremely high power density (10–100 MW/m² in ITER). Materials must withstand repeated exposure to 1–10 keV electrons and ions, 14 MeV neutron bombardment, and transient heating from disruptions (100 MW/m² for microseconds). Tungsten is the current choice for the divertor strike zone (used in ITER, EAST, and most modern tokamaks) because of its high melting point and low sputtering rate, but even tungsten develops grain growth and embrittlement under neutron irradiation.
Neutron activation and transmutation: Fusion reactions produce 14.1 MeV neutrons. These neutrons are absorbed by the reactor structure (steel, copper, concrete), activating it and making it radioactive. This is actually an advantage for fusion compared to fission (shorter-lived radioactive waste, less long-term hazard), but it requires specialized materials engineering and remote handling equipment for maintenance. The activation cross-section and induced radioactivity depend on the material composition; designers use “low-activation” steels (reduced Mo, Nb, Ni) to minimize long-lived isotope production.
Heat extraction and conversion efficiency: The 17.6 MeV per D-T reaction is partitioned: 3.5 MeV goes to the alpha particle (4He nucleus), which heats the core plasma; 14.1 MeV goes to the neutron, which escapes the plasma and must be captured externally. Neutrons are absorbed in a breeding blanket surrounding the reactor, depositing their energy as heat. This heat is extracted by coolant (liquid lithium, molten salt, or high-pressure helium) and converted to electricity via a standard Rankine cycle. Current thermal-to-electric conversion is ~40% efficient (limited by thermodynamics); fusion plants need ~50% to be economically competitive with fission. This requires high blanket outlet temperatures (600–700 K), which demands advanced materials.
Tritium breeding and the tritium fuel cycle: D-T fusion consumes tritium (³H, radioactive with 12-year half-life). Tritium is vanishingly rare on Earth (parts per trillion in atmosphere). A fusion reactor must breed its own tritium from lithium-6 in the blanket via the reaction: ⁶Li + n → ⁴He + ³H + 4.8 MeV. This breeding ratio must exceed 1 (produce more tritium than consumed) for the reactor to be self-sustaining. The challenge: maximize neutron multiplication in the blanket while minimizing tritium losses and inventory. Tritium is volatile, soluble in metal coolants, and permeates through steel at high temperatures. Containing and recovering tritium requires advanced material barriers and chemical processing.
Scale-up uncertainties and turbulent transport: Physics that works at 1 m scale may not work at 10 m scale. Plasma turbulence—driven by micro-instabilities like electron temperature gradient (ETG) and ion temperature gradient (ITG) modes—scales with gyroradius and collisionality. In larger, hotter devices, turbulence can behave very differently. ITER is ~10× larger than current tokamaks; the scaling laws are empirical, and there’s always concern that anomalous transport will be worse than predicted. This is where DeepMind’s AI approach has potential value: instead of relying on hand-tuned models, learning-based controllers can adapt to the actual transport in a new device.

Why This Matters for SPARC and ITER

SPARC’s advantage is that it’s testing high-field, compact physics in a real device. ITER’s advantage is that it’s large enough to test scale-up physics and full D-T reactions. Both are necessary; they’re complementary.

Part 11: The Path Forward—2030 and Beyond

The tokamak roadmap is accelerating:

2026–2027 Milestones

NSTX-U restart (2026): Spherical tokamak physics validation
SPARC first plasma (2026): Compact, high-field tokamak commissioning
SPARC achieve Q > 1 (2027, target): Net fusion gain in a private device
EAST sustained records: Continue plasma duration and density records
WEST steady-state operations: Validate long-pulse tokamak engineering

2030–2035 Targets

ITER D-T operations (post-2034): Full fusion reactions and alpha heating
BEST net power gain (2030, target): China’s commercial demonstration tokamak
ARC construction begins: CFS commercial plant (if SPARC succeeds)
First commercial fusion electricity (likely from a private company, not ITER)

The physics is solved. The remaining challenges are engineering—materials, control systems, supply chains, and the eternal fusion mantra: “Fusion is 30 years away, and always will be.” But that joke is getting old. Real progress is accelerating.

Conclusion: A Century of Plasma Physics Compressed into a Reactor

The tokamak is a marvel of 20th-century physics and 21st-century engineering. It elegantly solves the confinement problem using toroidal geometry and carefully tuned magnetic fields. It faces real instabilities and engineering challenges, but each is being systematically addressed.

Recent breakthroughs—EAST’s density records, WEST’s duration record, DeepMind’s AI control, Commonwealth’s compact HTS design—are not silver bullets. They’re incremental steps on a well-mapped roadmap. But incremental steps that are accelerating.

The next 5 years will be decisive. SPARC will prove or disprove whether compact, high-field fusion can achieve net gain. ITER will show whether large-scale, first-principles tokamak physics scales as predicted. EAST and China’s BEST will demonstrate whether Asian fusion engineering can lead. If all three succeed—and there’s no fundamental reason they shouldn’t—then fusion stops being a distant dream and becomes a coming energy revolution.

The physics of confining 150-million-degree plasma in a torus is complex, but it’s no longer mysterious. The next century of fusion is about making it economical. That engineering challenge is harder than the physics. But for the first time, it’s the binding constraint—not a fundamental law of nature.

References and Further Reading

Author’s Note: This post reflects tokamak physics as of April 2026. The landscape is moving fast. By the time you read this, there may be new records or unexpected setbacks. That’s the nature of frontier science. What remains constant is the fundamental physics: the toroidal geometry, the magnetic field configuration, the instabilities, and the triple product. Those don’t change; only the engineering maturity and the timeline evolve.