The Physical Stack Behind Fusion

An account of what it takes to turn fusion into electricity: the machines, materials, fuel cycle, economics, and companies building each layer.

Research through 16 September 2026
Figure 1 · Historical Energy Balance & Commercial Trajectory
Metric:
Timeline:
Trajectory:
The Positive Downtrend in Fusion Energy Balance (1958–2026) with Commercial BenchmarkLine chart displaying historical fusion energy milestone experiments against scientific breakeven (Q = 1.0) and the commercial power plant engineering benchmark (Q = 10 to 30) on a 14-order-of-magnitude logarithmic scale.1958197019801990200020102020202610¹²10⁻¹²10¹⁰10⁻¹⁰10⁸10⁻⁸10⁶10⁻⁶10⁴10⁻⁴10²10⁻²1.01.00.10100.033300.01100Scientific Breakeven threshold (Q = 1.0, R = 1.0)▼ Net Energy Extracted / Target Surplus Zone (Q > 1, Eout > Ein)Burning Plasma / ITER Benchmark (Q ≥ 10, R ≤ 0.10)★ Commercial Power Plant Benchmark (Q ≥ 10–30, R ≤ 0.10–0.033): target for net electricityClosed-loop engineering breakeven (covers driver wall-plug efficiency & recirculating plant house loads: Qeng > 1)Inverse Burden Ratio, R = Ein / Eout (log₁₀ scale; lower is better)→ Scientific Gain Qsci (log₁₀)Research cut-off (Sep 2026)
Laser Inertial (ICF)All-Time Milestone Record

NIF (Record Apr 7, 2025)

Apr 7, 2025 · ICF (Nd:Glass)
Net Energy Extracted+6.52 MJThermonuclear surplus (Eout > Ein)
Target Input (Ein)
2.08 MJ
Fusion Yield (Eout)
8.60 MJ
Scientific Gain (Qsci)
4.135 (Q > 1)
Inverse Burden (Ein / Eout)
0.242 (Surplus) (-0.62 log₁₀)
Commercial Benchmark Standing (Target: Q ≥ 10–30):

Scientific Breakeven achieved (Q = 4.135, +6.52 MJ). Remaining gap: 7.3× increase in target gain needed to achieve the Commercial Closed-Loop Power Plant Benchmark (Q = 30, R = 0.033).

Key milestone: All-Time Milestone Record: Target gain > 4; continuous diamond doping (+6.52 MJ net target surplus, Rburden = 0.242)

Figure 1: The positive downtrend in fusion energy balance plotted on a 14-order-of-magnitude logarithmic scale (1958–2026), showing the collapse in the inverse energy burden ratio (Rburden = Ein / Eout = 1 / Qsci). The horizontal green line marks scientific breakeven (Qsci = 1.0, Rburden = 1.0), first crossed by NIF in December 2022 and extended to a record Qsci = 4.135 (+6.52 MJ surplus) on April 7, 2025. The amber benchmark line and band (Q ≥ 10–30, R ≤ 0.10–0.033) mark the target level required for commercial closed-loop engineering breakeven, overcoming driver efficiency and auxiliary plant loads to generate net electricity for the grid (Qeng > 1). Beside the chart, the readiness card details the five finish lines and the ~7.2× factor increase in target gain remaining to reach commercial operation.
Source: fusion.md, “Historical Trajectory of Nuclear Fusion Energy: Experimental Input-Output Balances, Gain Ratios, and the Transition to Net Ignition (1958–September 2026).” Evidence & definitions

What the record shows

  • Fusion has crossed a scientific threshold, but not the power-plant threshold. NIF produced 8.6 MJ from 2.08 MJ delivered to its target: a target gain of 4.13, excluding the electricity consumed by the facility.
  • High-field magnets are becoming real hardware. A full-scale MIT/CFS REBCO magnet reached 20 T; SPARC is reported to be almost 80% complete.
  • The least mature common layers are the breeding blanket, closed tritium cycle, and neutron-resistant materials. A plasma experiment does not validate these plant systems.
  • Published first-of-a-kind cost estimates span $1.4m to $43m per net MWe. Grid-value modelling suggests a target of $2.7m to $7.5m/MWe for substantial deployment.

The supplied research records approximately $14.24bn of cumulative private investment in the 2026 Fusion Industry Association survey. Reported results, company announcements, modelled costs, and analyst judgements remain separate throughout this page.

How the system fits together

A fusion power plant is an energy-conversion chain. Magnets confine a plasma; heating and fuel systems sustain reactions; a blanket captures neutron energy and breeds fuel; coolant carries heat to a turbine. The electricity left after powering the plant is what reaches the grid.

The sections below follow that chain, then examine the competing architectures, their schedules, and the cost of commercially useful power.

In this report
Part I

Build the power plant

The plasma is only the beginning. Each layer has to work on its own, and then work with every other layer, repeatedly, for years.

01Magnets & superconductors

A magnetic fusion plant begins with a field strong enough to keep an extraordinarily hot plasma away from the walls. Superconducting coils carry enormous currents; cryogenics keeps the conductors below their operating limits.

Confine the plasma

Conventional D-T magnetic plant
Simplified tokamak and energy-conversion schematicConcentric plasma, lithium blanket, and external superconducting coils connect to a heat exchanger, turbine generator, and electricity grid. The diagram is conceptual, not to scale.Superconducting coilsLithium blanketD-T plasmaVacuum vesselHeat exchangerTurbineGrid

Superconducting coils create a magnetic cage. Stronger fields can shrink the machine, but neutron shielding and structural loads set physical limits.

20 T · Full-scale MIT/CFS HTS magnet result
Figure 2: From magnetic confinement to net electricity. Select a stage to trace the conversion chain. Conceptual schematic, not to scale; design values are not measured plant output.

Higher field can make a smaller machine. The MIT/CFS programme demonstrated a full-scale REBCO magnet at 20 T. ITER’s conventional superconducting system stores about 51 GJ. Compact high-field designs still need structural support, quench protection, and enough shielding to protect their coils from neutrons.

What’s inside: components, evidence, and interfaces
Components
REBCO or Nb₃Sn/Nb-Ti conductor, toroidal and poloidal coils, structural cases, joints, cryostat, power supplies.
Key metrics
Peak field (T), stored energy (GJ), critical current, operating temperature, strain, and joint resistance.
Readiness
TRL 6-8 for low-temperature superconducting systems; approximately 5-6 for compact HTS fusion systems.
Interfaces
Cryogenics → conductor → coil → plasma control; neutron shielding protects the magnet system.
Principal risk
Mechanical stress, radiation damage, detection of quenches, and economical production of fusion-grade REBCO tape.

Adapted from the physical-stack section of fusion.md. Technology readiness levels are analyst mappings, not official programme ratings.

Research & industry participants5 organisations

Commonwealth Fusion Systems / MIT

ITER Organization

Tokamak Energy

Proxima Fusion

Thea Energy

Programmes associated with this layer in the supplied report; inclusion does not imply a supplier contract.

02Vacuum & plasma heating

The vessel must hold a clean, near-empty environment while fuel flows in and helium ash flows out. Neutral beams and radio-frequency systems then heat the plasma and, in many tokamaks, help sustain its current.

Industrial experience does not yet equal power-plant availability. ITER’s systems evacuate a 1,400 m³ vessel and an 8,500 m³ cryostat. Each heating neutral beam is designed to deliver 16.5 MW of 1 MeV deuterium for up to 3,600 seconds. A commercial plant must do this efficiently and reliably while handling tritiated exhaust.

What’s inside: components, evidence, and interfaces
Components
Vacuum vessel, mechanical pumps, cryopumps, neutral-beam injectors, RF sources, transmission systems, and plasma controls.
Key metrics
Base pressure, pumping speed, leak rate, injected MW, wall-plug efficiency, and pulse length.
Readiness
Vacuum systems approximately TRL 7-8; heating approximately 7-8 experimentally, 5-6 at plant efficiency and availability.
Interfaces
Fuel cycle, plasma exhaust, cryogenics, electrical supplies, vessel seals, and shielding.
Principal risk
Tritiated pumping, remote maintenance, component lifetime, coupling losses, and high recirculating power.

Adapted from the physical-stack section of fusion.md. Technology readiness levels are analyst mappings, not official programme ratings.

Research & industry participants4 organisations

ITER Organization

UKAEA

TAE Technologies

QST / Fusion for Energy

Programmes associated with this layer in the supplied report; inclusion does not imply a supplier contract.

03Fuel & the tritium cycle

Deuterium is widely available. Tritium is not. A deuterium-tritium plant must recover unburned fuel and breed replacement tritium from lithium, closing a cycle that current experiments do not demonstrate at commercial scale.

A reactor cannot simply buy decades of fuel. A 1 GW fusion source consumes about 55.8 kg of tritium per full-output year. Tritium also decays, with a 12.32-year half-life. Breeding must exceed consumption with enough margin for decay, processing inventory, extraction delays, and losses.

What’s inside: components, evidence, and interfaces
Components
Fuel injection, isotope separation, storage beds, detritiation, blanket extraction, and fuel-accounting systems.
Key metrics
Tritium breeding ratio, inventory, burn-up fraction, processing residence time, permeation, and losses.
Readiness
TRL 6-7 for tritium handling; approximately 3-4 for an integrated closed commercial cycle.
Interfaces
Lithium blanket → extraction → isotope separation → storage → plasma → exhaust recovery.
Principal risk
Scarce start-up inventory, insufficient breeding margin, and tritium permeation into coolant and structures.

Adapted from the physical-stack section of fusion.md. Technology readiness levels are analyst mappings, not official programme ratings.

Research & industry participants4 organisations

UKAEA / LIBRTI

ITER Organization

EUROfusion DEMO

BABY experimental programme

Programmes associated with this layer in the supplied report; inclusion does not imply a supplier contract.

04Blanket, first wall & materials

The blanket has three jobs: turn neutron energy into heat, breed tritium, and shield the rest of the machine. The first wall and divertor must withstand extreme heat and particle loads while remaining replaceable.

This is one of the least mature integrated layers. D-T neutrons carry 14.1 MeV and progressively damage structural materials. ITER’s blanket can remove roughly 736 MW of heat, but breeding is tested in modules rather than a self-sufficient fuel cycle. Heat removal alone is not a validated power-plant blanket.

What’s inside: components, evidence, and interfaces
Components
Lithium breeder, tungsten plasma-facing surfaces, structural steel, neutron shielding, coolant channels, and divertor.
Key metrics
Breeding ratio, heat flux (MW/m²), neutron wall load, displacements per atom, coolant temperature, and replacement interval.
Readiness
Approximately TRL 3-5 for the integrated blanket and breeder.
Interfaces
Plasma → first wall → blanket → coolant and tritium plant; blanket → shielding → magnets.
Principal risk
Cracking, erosion, embrittlement, corrosion, tritium permeation, and remote replacement of activated components.

Adapted from the physical-stack section of fusion.md. Technology readiness levels are analyst mappings, not official programme ratings.

Research & industry participants4 organisations

ITER Organization

EUROfusion DEMO

IFMIF-DONES

UKAEA

Programmes associated with this layer in the supplied report; inclusion does not imply a supplier contract.

05Heat, turbines & net power

Hot coolant transfers the captured energy to a power cycle. Steam turbines are mature technology, but integrating them with a fusion heat source, tritium isolation, and large auxiliary loads is a new problem.

Gross generation is not what reaches the grid. The ARC physics basis gives an illustrative chain of about 1.18 GW thermal input, 500 MWe gross generation at 42% efficiency, and 400 MWe net after approximately 100 MW of house loads. These are design quantities, not measured plant output.

What’s inside: components, evidence, and interfaces
Components
Primary coolant loop, heat exchangers, steam generator, turbine, condenser, generator, and grid connection.
Key metrics
Thermal efficiency, gross and net MWe, coolant outlet temperature, recirculating load, and capacity factor.
Readiness
TRL 9 for conventional steam Rankine equipment; approximately 4-6 for an integrated fusion plant.
Interfaces
Blanket → primary coolant → heat exchanger → turbine → generator → grid and plant auxiliaries.
Principal risk
Coolant temperature limits, pulsed heat input, tritium isolation, and electricity consumed by the machine itself.

Adapted from the physical-stack section of fusion.md. Technology readiness levels are analyst mappings, not official programme ratings.

Research & industry participants4 organisations

Commonwealth Fusion Systems

EUROfusion DEMO

UK Fusion Energy / STEP

Type One Energy

Programmes associated with this layer in the supplied report; inclusion does not imply a supplier contract.

06Maintenance & availability

A successful pulse lasts seconds. A useful asset must operate for decades. Activated internal components need remote inspection and replacement, with access designed into the machine from the start.

Component lifetime becomes an electricity-cost problem. Once a plasma is comfortably burning, a longer-lived blanket or a shorter maintenance outage can be more valuable than a small improvement in plasma gain. Remote handling, replaceable modules, and material qualification are essential parts of the plant architecture.

What’s inside: components, evidence, and interfaces
Components
Remote manipulators, inspection systems, replaceable blanket and divertor modules, hot cells, and material qualification facilities.
Key metrics
Replacement interval, outage duration, net annual capacity factor, inspection coverage, and replacement cost.
Readiness
Industrial remote-handling experience exists; a commercially available integrated fusion maintenance regime remains unproven.
Interfaces
Vessel access, magnets, activated blanket and divertor, tritium containment, and plant operations.
Principal risk
Long unscheduled outages, inaccessible components, activation, and repeated replacement costs.

Adapted from the physical-stack section of fusion.md. Technology readiness levels are analyst mappings, not official programme ratings.

Research & industry participants4 organisations

UKAEA

EUROfusion DEMO

ITER Organization

IFMIF-DONES

Programmes associated with this layer in the supplied report; inclusion does not imply a supplier contract.

Part II

Different machines, shared constraints

Every architecture trades one difficult problem for another. The useful comparison is the complete power-production boundary.

High-field tokamak

Stronger magnets make the machine smaller. The deepest magnetic-confinement evidence base still leaves disruptions, neutron shielding, tritium breeding, and plant availability to solve.

Stellarator

Complex coils provide naturally steady-state confinement without a large driven plasma current. The trade is demanding geometry, manufacturing, and access for maintenance.

Laser inertial fusion

NIF has demonstrated target ignition. A power plant needs efficient lasers, inexpensive precision targets, several shots per second, and a chamber that survives repeated neutron loads.

Pulsed and direct-conversion concepts

Z-pinches, magnetised targets, FRCs, and magnetic mirrors simplify parts of the conventional plant. They introduce demanding confinement, repetition-rate, fatigue, or energy-recovery requirements.

Removing a component from a reactor diagram does not remove its job from the power plant.

Who is building what

Public experiments establish the physics and integration base. Private programmes pursue shorter design-build-test cycles across multiple architectures.

19 programmes
Commonwealth Fusion SystemsSPARC → ARCHigh-field tokamak
Reported evidence
SPARC reported almost 80% complete
Announced next step
~400 MWe, early 2030s
Unresolved risk
Blanket, fuel cycle, and availability beyond SPARC
HelionPolaris → OrionPulsed FRC / direct conversion
Reported evidence
Polaris operating; company reports D-T fusion
Announced next step
≥50 MW, originally 2028
Unresolved risk
Net electricity at commercial repetition rate
TAE TechnologiesDa VinciBeam-driven FRC
Reported evidence
Five reactor generations operated
Announced next step
First operations, 2031
Unresolved risk
Confinement and recirculating beam power
Proxima FusionAlpha → StellarisHTS stellarator
Reported evidence
Alpha consortium; €411m financing in 2026
Announced next step
Demonstrator, early 2030s
Unresolved risk
Burning plasma, complex coils, and maintenance
Type One EnergyInfinity TwoStellarator
Reported evidence
400 MWe design with TVA interest
Announced next step
400 MWe design capacity
Unresolved risk
Whole-plant validation beyond existing stellarators
Thea EnergyEos → HeliosPlanar-coil stellarator
Reported evidence
Full-size Eos-spec coil operated in 2026
Announced next step
Helios preconcept reviewed
Unresolved risk
Reactor-scale plasma and neutron geometry
Tokamak EnergyST40 / Demo4Spherical tokamak / HTS
Reported evidence
Demo4 reached 13.7 T; STEP magnet partner
Announced next step
HTS and plasma development
Unresolved risk
Centre-column shielding and plant engineering
General FusionLM26Magnetised-target fusion
Reported evidence
Compressional heating demonstrated
Announced next step
Lawson milestones through 2028
Unresolved risk
Compression symmetry, fatigue, and repetition
Zap EnergyCenturySheared-flow Z-pinch
Reported evidence
1,080 consecutive ≥100 kA plasmas
Announced next step
~50 MWe/module preconcept
Unresolved risk
High-current stability and electrode lifetime
Pacific FusionDemonstration SystemPulsed-magnetic inertial fusion
Reported evidence
Campus broke ground in August 2026
Announced next step
Net facility gain, 2030
Unresolved risk
High-yield targets and repetitive chamber systems
Xcimer EnergyPhoenix → Vulcan → AthenaExcimer-laser inertial fusion
Reported evidence
Phoenix operating; >1 kJ source
Announced next step
400 MWe, mid-2030s
Unresolved risk
Efficient, durable, repetitive multi-MJ lasers
Marvel FusionATLASLaser inertial / fast ignition
Reported evidence
$150m laser demonstrator with CSU
Announced next step
Early operation around 2027
Unresolved risk
Target gain, laser efficiency, and optics lifetime
Focused EnergyGerman-US laser programmeLaser inertial fusion
Reported evidence
Biblis concept reported in research
Announced next step
Mid-2030s ambition
Unresolved risk
Driver, target, and chamber integration
Realta FusionWHAMAxisymmetric magnetic mirror
Reported evidence
Plasma kinetic-energy conversion in 2026
Announced next step
Confinement and conversion development
Unresolved risk
End losses and reactor confinement
ITER OrganizationITERSuperconducting tokamak
Reported evidence
Assembly and integration
Announced next step
D-T operation, 2039; no electricity
Unresolved risk
Full-scale first-of-a-kind integration
UK Fusion Energy / UKAEASTEPSpherical tokamak
Reported evidence
Engineering and development programme
Announced next step
First operations, 2040
Unresolved risk
HTS, shielding, and maintainability
EUROfusionDEMOD-T tokamak
Reported evidence
Concept and technology development
Announced next step
300-500 MWe design
Unresolved risk
Closed fuel cycle and plant availability
LLNLNational Ignition FacilityLaser inertial fusion
Reported evidence
8.6 MJ yield; target gain 4.13
Announced next step
Ignition research; no electricity plant
Unresolved risk
Target gain excludes facility electricity
Max Planck IPPWendelstein 7-XStellarator
Reported evidence
1.8 GJ energy turnover over 360 s
Announced next step
Long-duration confinement research
Unresolved risk
Extrapolation to burning-plasma conditions

Programme status and target dates are reproduced from the supplied September 2026 research. A target is not a forecast; a design review is not hardware validation.

There is more than one finish line

“Net energy” can describe entirely different boundaries. Crossing one does not mean the next has been crossed.

  1. 1

    Scientific breakeven

    Fusion output exceeds energy delivered directly to the target or plasma.

    Demonstrated at NIF’s target
  2. 2

    Facility breakeven

    Output also covers the energy consumed by the driver and machine.

    Not demonstrated in this record
  3. 3

    Net electricity

    Electrical generation exceeds all plant auxiliary loads.

    Not demonstrated
  4. 4

    Repeatable pilot operation

    Meaningful electrical output, maintainability, and sustained availability.

    Analyst outlook: 2030s
  5. 5

    Financeable commercial power

    Reliable generation at a capital and operating cost buyers can finance.

    Analyst outlook: a harder 2040s problem
Figure 3
Announced first-plant net electrical capacityCompany targets, not observed generation: Helion at least 50 MW in 2028; TAE 50 MWe in 2031; CFS approximately 400 MWe in the early 2030s; Xcimer 400 MWe in the mid-2030s. All are future announcements.Net electrical capacity, MWe (announced)01002003004005002026202820302032203420362038Research cut-offHelion Orion2028 targetTAE Da Vinci2031 targetCFS ARCEarly 2030sXcimer AthenaMid-2030sNo demonstrated net electricity

Announced net electrical capacity

Helion Orion2028 target50MWe
TAE Da Vinci2031 target50MWe
CFS ARCEarly 2030s400MWe
Xcimer AthenaMid-2030s400MWe

Company targets. No demonstrated net fusion electricity in the supplied record.

Figure 3: Selected announced first-plant net electrical capacities. Open markers are company targets, not operating plants. Horizontal ranges show approximate early- and mid-2030s windows; they are not additional commitments. No net fusion electricity has been demonstrated in the supplied record.
Source: fusion.md, “How close fusion is to deployment.” Evidence & definitions
View the report’s probability ranges
MilestoneDateAnalyst probability
Private magnetic plasma Q > 1203065-80%
>10 MWe net exported to a grid203020-35%
Repeated ≥50 MWe pilot operation203545-65%
≥300 MWe plant, >50% annual capacity factor204020-35%
Financeable repeat-order plant below ~$7.5m/MWe204515-30%
>1 GWe operating commercial fusion globally205045-70%

Judgements from the report, not project-published probabilities or statistical confidence intervals.

Part III

The cost of useful electricity

The denominator matters. A 500 MW fusion plasma with no turbine has zero electrical capacity. A 500 MWe generator using 100 MW to run its own plant exports 400 MWe.

Figure 4

What it may cost. What the grid can afford.

First-of-a-kind estimates$1.4-43mper net MWe · Literature / expert estimates
Detailed mature-plant study$8.8-22.2mper net MWe · Bottom-up model
Grid-value threshold$2.7-7.5mper net MWe · Market-value model
These ranges answer different questions. The grid-value range is a deployment threshold, not a construction estimate. Studies: Nature Energy (2026), Applied Energy (2025), and Princeton/Joule, as cited in fusion.md.

Cheap fuel alone does not make cheap fusion power. Financing, replacement blankets and divertors, recirculating electricity, and unscheduled downtime can dominate the economics.

Figure 5 · Interactive model

What would the electricity cost?

Change the assumptions. The result is a scenario, not a price forecast.

Early fleet$170/MWh

Levelised cost of electricity

Capital recovery
$110
O&M + replacements
$51
Variable / fuel cycle
$8

35-year life; fixed O&M 3% of CAPEX/year; variable cost $8/MWh. All capital costs use net electrical capacity. Real 2026 dollars; excludes tax credits, construction-period interest, and decommissioning.

Calculation & assumptions

LCOE = CAPEX × (capital recovery factor + fixed O&M + replacement reserve) ÷ (8.76 × capacity factor) + variable costs.

CAPEX is converted to $/kWe; annual percentages and capacity factor are expressed as fractions. Capital recovery uses the selected real discount rate and scenario lifetime. Reset a scenario to restore its original inputs.

Model adapted from the four analytical scenarios in fusion.md. FOAK means first-of-a-kind; NOAK means a mature, repeated-build plant.

The report suggests experience rates of 2-8% per cumulative-capacity doubling, closer to large, customised thermal plants than mass-produced electronics. Standardisation and easier maintenance matter as much as a smaller reactor core.

About the record

This page is an editorial adaptation of Fusion Energy: The Physical Stack, Economics and Path to Commercial Power, with a research cut-off of 16 September 2026. Monetary figures are US dollars unless labelled otherwise.

Reported / measured
An experimental result or programme status recorded in the research.
Announced / target
A company or programme’s intended future performance or date.
Modelled / derived
A study result or calculation conditional on stated assumptions.
Analyst judgement
An interpretation of readiness, timing, or probability, not a measured result.

Source limitation: the supplied Markdown includes citation identifiers, but no resolvable bibliography for those identifiers. Claims here are attributed to that report and have not been independently reverified.