LUNAROPS · OPERATIONAL UPLINK 100% UPTIME 1,247d POSTS 893 JEFF.MOON@LUNAROPS.DEV UTC --:--:--

Fusion, Honestly

15 min read Infrastructure of Civilization
fusionenergyplasma-physicsinfrastructurenuclearscience
Contents

Fusion is the energy source that has been thirty years away for seventy years, and the joke is starting to feel unfair — because the physics genuinely is moving now, faster than at any point in the field’s history. But “moving” and “almost here” are different claims, and the gap between them is exactly where press releases live. In December 2022 a US laser facility achieved “ignition” and the headlines said fusion had produced more energy than it consumed. That statement is true under one specific definition of “consumed” and wildly false under the definition a grid operator would use. Understanding which is which — separating the real, hard-won physics milestone from the accounting sleight-of-hand the headline performed — is the whole game when reading fusion news. This post is the honest version: what the milestones actually proved, what they didn’t, why the remaining problems are materials and engineering rather than physics, and what timeline a numerate skeptic should actually hold. Fusion is real, it is coming, and it is further away than the funding announcements imply. All three of those are true at once.

Why Fusion Is Hard: The Lawson Criterion

Fusion releases energy by forcing light nuclei together until the strong force binds them, which liberates the binding-energy difference. The reaction that matters for a first generation of reactors is deuterium plus tritium (two hydrogen isotopes), because it has by far the highest cross-section at the lowest temperature of any candidate. “Lowest” is relative: you still need to get to roughly 100-150 million degrees Celsius, ten times the core of the Sun, because the Sun cheats with crushing gravitational pressure and absurd density that no earthbound machine can match.

The difficulty is not reaching the temperature — we have done that for decades. The difficulty is holding a hot enough, dense enough plasma together for long enough that the energy released by fusion exceeds the energy required to create and confine it. That three-way trade is the Lawson criterion, usually expressed as the triple product of density (n), confinement time (tau), and temperature (T):

  Triple product for net energy (D-T fuel):

    n * T * tau   >=   ~ 3 x 10^21   keV * s / m^3

  where
    n   = plasma density            (particles / m^3)
    T   = temperature               (keV; 1 keV ~ 11.6 million K)
    tau = energy confinement time   (seconds)

  Two ways to win, at opposite extremes:

    MAGNETIC (tokamak):   modest n, big tau
      n   ~ 10^20 /m^3
      tau ~ seconds
      "thin gas, held for a long time"

    INERTIAL (laser):     enormous n, tiny tau
      n   ~ 10^31 /m^3
      tau ~ 10^-11 s  (100 picoseconds)
      "crush it so hard it fuses before it flies apart"

That single inequality explains the entire architecture of the field. You can satisfy it with a tenuous plasma confined by magnetic fields for seconds (the tokamak and stellarator route), or with a tiny pellet compressed to densities beyond the center of the Sun for a fraction of a nanosecond (the laser route). Density and confinement time trade off across thirty orders of magnitude, and every fusion machine ever built is a different bet on where to sit on that curve. The temperature term is roughly fixed by the fuel; the engineering fight is the product of the other two.

This is also why fusion is fundamentally harder than fission, which needs no such criterion — a critical mass of fissile material sustains a chain reaction at room temperature, as covered in how nuclear reactors work. Fission’s problem is slowing reactions down; fusion’s problem is making them happen at all, and keeping them happening.

The NIF Headline, Decoded: Target Gain vs. Wall-Plug Gain

The National Ignition Facility (NIF) at Lawrence Livermore takes the inertial route: 192 laser beams converge on a peppercorn-sized capsule of D-T fuel, compressing it until it fuses. In December 2022 NIF crossed a real threshold — for the first time, the fusion energy released exceeded the laser energy delivered to the target. The progress since has been genuine and rapid:

Date Laser energy to target Fusion yield Target gain (Q_target)
Aug 2021 (near-miss) 1.9 MJ 1.3 MJ 0.7
Dec 2022 (first ignition) 2.05 MJ 3.15 MJ 1.5
Feb 2025 2.05 MJ 5.0 MJ 2.44
Apr 2025 (record yield) 2.08 MJ 8.6 MJ 4.13
Oct 2025 (10th ignition) 2.07 MJ 3.5 MJ 1.74

A target gain above 4 is a spectacular scientific achievement and a vindication of decades of work. It is also where the headline quietly switches definitions. The gain in that table — call it Q_target — is fusion energy out divided by laser energy delivered to the capsule. It says nothing about the energy required to make the laser fire.

  Where the energy actually goes at NIF (approximate):

  grid power drawn   ~300-400 MJ
        |
        |  flashlamps + laser chain are ~1% efficient
        v
  laser to target    ~2 MJ  <-- "input" in the headline
        |
        |  fusion burn, Q_target ~ 4
        v
  fusion yield       ~8.6 MJ <-- "output" in the headline

  Headline gain:   8.6 / 2.0   = 4.3   (TARGET gain)
  Honest gain:     8.6 / ~350  = 0.025 (WALL-PLUG gain)

The NIF lasers are about 1% efficient at converting grid electricity into light on target, so producing 2 MJ at the capsule costs hundreds of megajoules from the wall. Measured end to end — the only measure a power plant cares about — NIF returns roughly 2-3% of the energy it draws. This is the Q_plasma versus Q_total distinction, and it is the single most important thing to internalize about fusion news. Q_plasma (or Q_target) is a physics number about the burn; Q_total is an engineering number about the whole plant. A grid-relevant reactor needs Q_total well above 1, which implies Q_plasma in the range of 10 or higher to leave margin for every inefficiency in between. NIF was never designed to be a power plant — it is a nuclear weapons science facility, and ICF for energy is a different machine that does not yet exist. The milestone was real. The headline borrowed a power-plant frame the experiment was never built to satisfy.

The Confinement Zoo: Tokamaks, Stellarators, and Lasers

On the magnetic side, the dominant design is the tokamak: a doughnut-shaped vacuum chamber wrapping the plasma in helical magnetic fields. Part of that field comes from external coils; crucially, part comes from a large electric current driven through the plasma itself. That self-current is the tokamak’s strength and its curse — it makes for efficient confinement, but a plasma carrying millions of amps is prone to sudden instabilities called disruptions, where confinement collapses in milliseconds and dumps enormous energy into the chamber walls. Tokamaks are also, because of that current, naturally pulsed rather than steady-state, which a power plant dislikes.

The stellarator answers the same problem differently: it produces the entire twisting field from external coils alone, with no plasma current, by making those coils fantastically complicated three-dimensional shapes. This buys inherent steady-state operation and freedom from current-driven disruptions, at the cost of coils so geometrically brutal that the leading example, Germany’s Wendelstein 7-X, was only buildable once computational optimization matured. W7-X has demonstrated long-pulse, high-performance plasmas that validate the concept; the trade is manufacturability and a historically lower performance ceiling than tokamaks, a gap that is narrowing.

Inertial confinement (NIF and the laser startups) is the third branch — no magnetic bottle at all, just compress and burn before the fuel disassembles. For energy rather than weapons science it faces a brutal additional requirement: a power plant must do this several times per second, with a fresh perfectly-manufactured target each shot, and capture the energy. NIF fires roughly once per several hours.

Approach Confinement Strength Hard problem
Tokamak Magnetic, plasma current Best-proven performance; ITER, SPARC Disruptions; naturally pulsed
Stellarator Magnetic, external coils only Inherently steady-state, disruption-free Insanely complex coils; manufacturability
Laser ICF Inertial (compress + burn) Highest density; NIF proved ignition Rep-rate (shots/sec), target cost, efficiency
Compact / alternative Various (FRC, Z-pinch, mirror) Cheaper, faster iteration Mostly unproven at gain

There is real intellectual honesty in noting that no one knows which branch wins, and possibly none of them do in their current form. The field is a portfolio of bets, not a march down one road.

ITER: What the Big Machine Will and Won’t Prove

ITER, under construction in southern France, is the largest scientific collaboration on Earth — 35 nations building a tokamak designed to produce 500 MW of fusion power from 50 MW of input heating, a Q_plasma of 10. It is the machine meant to prove that a burning plasma — one heated primarily by its own fusion reactions rather than external heating — can be sustained and controlled at power-plant scale.

It is also the cautionary tale that should temper every optimistic timeline. ITER’s schedule has slipped repeatedly, and the 2024 rebaselining was brutal: first plasma, once promised for 2025, is now 2034; full-power deuterium-tritium operation slid to 2039; and the cost overrun for this single revision was around 5 billion euros on a project already estimated well north of 20 billion. The causes are instructive and not unique to ITER — pandemic disruption, manufacturing defects (cracks discovered in cooling pipes and thermal shields), and the irreducible difficulty of building a first-of-a-kind machine to nuclear tolerances across a 35-nation supply chain where no one is fully in charge.

What ITER will prove, if it works: that a burning plasma is controllable, that the physics of self-heating scales as predicted, and that the integrated engineering of a reactor-scale tokamak is feasible. What it will not do: generate a single watt of electricity. ITER has no turbine; its heat goes to cooling towers. It is a physics experiment, and the demonstration power plant meant to follow it (DEMO, in various national flavors) exists mostly on paper with operation dates in the 2040s-2050s. Anyone citing ITER as evidence that grid fusion is near has the chronology backward — ITER is evidence of how long the careful version takes.

The Private Surge: Faster, Riskier, Divergent Bets

The most genuinely new thing in fusion is the wave of private companies betting that they can move faster than the intergovernmental behemoth by taking on more risk and exploiting technologies ITER’s design predates — chiefly high-temperature superconducting (HTS) magnets, which produce far stronger fields in far smaller machines, shrinking the reactor and slashing cost. The capital is real: multiple companies have raised over a billion dollars, and the sector’s total now runs to the tens of billions.

The bets diverge sharply, which is the interesting part:

  • Commonwealth Fusion Systems is the HTS-tokamak frontrunner, an MIT spinout. Its SPARC machine aims to demonstrate Q_plasma > 1 — net plasma gain — with first plasma targeted for 2027, and a follow-on commercial plant, ARC, promising roughly 400 MW in the early 2030s. CFS is the most “conventional physics, aggressive engineering” bet: a known design made small and cheap by better magnets.
  • Helion takes the field-reversed-configuration route with a pulsed, non-tokamak machine and the most aggressive public timeline in the industry — electricity by 2028, backed by a power-purchase agreement with Microsoft. Helion also bets on deuterium-helium-3 fuel, sidestepping some neutron problems while creating a helium-3 supply problem.
  • TAE Technologies pursues an even harder fuel, proton-boron-11, which is aneutronic — it releases its energy as charged particles rather than neutrons, which would eliminate the materials and tritium nightmares described below. The catch is that p-B11 needs far higher temperatures than D-T, a steep physics price for the engineering relief.
  • Zap Energy bets on the sheared-flow-stabilized Z-pinch: no expensive magnets, no lasers, just a cleverly stabilized current pinch, aiming for radical cost reduction if the stability holds.

The honest read on the private surge: the capital and engineering have moved the field meaningfully forward, HTS magnets are a genuine step-change, and the iteration speed beats the government labs. But as of 2026, no private company has produced net energy by any definition, none has put fusion electricity on a grid, and the aggressive milestones have a consistent history of slipping. The right posture is neither dismissal nor belief — it is watching for the specific, unfakeable milestone of Q_plasma > 1 in a machine designed to scale, and treating everything before that as promising preliminary.

The Problems Physics Doesn’t Solve: Materials and Tritium

Suppose the plasma physics is fully cracked tomorrow. A D-T reactor still faces two engineering problems that no amount of plasma cleverness touches, and they may be the actual long poles.

Neutron embrittlement. D-T fusion releases 80% of its energy as 14.1 MeV neutrons — far more energetic than fission neutrons, and electrically neutral, so no magnetic field can contain them. They slam into the reactor’s first wall and structure, and at those energies they don’t just heat the material; they knock atoms out of the crystal lattice and transmute nuclei, causing swelling, embrittlement, and the production of helium gas inside the metal that blisters it from within. Over a reactor’s life the first wall absorbs a neutron dose that no existing structural material is qualified to survive while staying strong and low-activation. Developing and testing such materials requires a dedicated 14 MeV neutron source (the long-delayed IFMIF/DONES facility) that itself does not yet operate at full capability. This is a multi-decade materials-science program running in parallel, and it is genuinely unsolved.

Tritium breeding. Tritium does not exist in nature in usable quantities — it has a 12.3-year half-life and must be manufactured. The plan is for the reactor to breed its own: surround the plasma with a “blanket” containing lithium, and let the fusion neutrons transmute lithium into tritium. For the reactor to be self-sufficient, the tritium breeding ratio (TBR) must exceed 1 — every tritium burned must produce more than one to replace it plus cover losses and decay. On paper this is achievable; in practice no one has ever demonstrated a closed tritium fuel cycle with TBR > 1 in an integrated reactor, the neutron multiplication needed to get there competes with every other thing you want the blanket to do (capture heat, shield the magnets), and the world’s total civilian tritium stockpile is only about 25 kilograms, much of it from aging fission reactors that are retiring. A first reactor needs kilograms to start; the bootstrap problem is real.

  The D-T fuel cycle that MUST close for grid fusion:

   plasma (D + T) --fuse--> He-4 + neutron (14.1 MeV)
                               |
                               v
              lithium blanket: n + Li -> T + He-4
                               |
                       breed T, TBR must be > 1
                               |
                               v
              extract T  -->  refuel plasma  --> (loop)

   Unsolved at integrated scale:
     - TBR > 1 demonstrated in a real reactor: never
     - structural material surviving the neutron dose: none qualified
     - closed tritium handling at kg scale: not done

Neither problem is a showstopper in principle. Both are the kind of grinding, expensive, decade-scale engineering that does not generate press releases and does not get solved by a magnet breakthrough or a record shot. They are why “we proved the physics” and “we have a power plant” are separated by a very long bridge.

A Realistic Clock

So when does fusion put power on the grid? A numerate, non-cynical estimate, separating the milestones:

  • Q_plasma > 1 in a scalable private machine: plausibly this decade. SPARC and peers have a real shot, and HTS magnets make it credible. This is the milestone worth watching for.
  • Q_total > 1 (true wall-plug net energy), demonstrated: 2030s at the earliest, and “demonstrated” is not “economical.”
  • First fusion electricity onto a grid, at any cost: most credibly the 2030s for an optimistic private effort, 2040s for the careful ITER-DEMO lineage.
  • Fusion as a meaningful fraction of grid generation: 2050s and beyond, gated not by physics but by tritium supply, materials qualification, regulatory frameworks, and the unforgiving economics of competing against solar-plus-storage that keeps getting cheaper (solar PV from photon to inverter and battery chemistry compared).

That last point is the one fusion optimists most often skip. Fusion is not competing against the energy market of 1995; it is competing against an electricity system being rebuilt around cheap renewables and storage, integrated into a grid (how the power grid works) that increasingly rewards flexibility over baseload. Even a working fusion plant must then be cheaper than the alternatives, and a first-generation fusion plant — a neutron-bombarded, tritium-handling, superconducting-magnet machine — will not be cheap. Its first markets may be niches where energy density or siting matters more than price.

Verdict

Fusion deserves neither the breathless “it’s here” of the funding announcements nor the weary “it’ll never work” of the cynics, and the honest position takes real effort to hold because it requires believing two things that feel contradictory: the physics is being solved, and grid fusion is still decades away. When you read the next fusion headline, do three things. First, find out whether the gain quoted is Q_target/Q_plasma (a physics number) or Q_total (a plant number) — the difference is usually a factor of 100 and it is almost always elided. Second, ask whether the milestone is repeatable and scalable or a hero shot in a machine never meant to be a power plant. Third, remember that even a fully solved plasma leaves neutron-resistant materials and a closed tritium cycle unsolved, and those are decade-scale grinds that no magnet buys you out of. Fusion is one of the few genuinely civilization-altering technologies actually under construction, and the progress of the last few years is real and accelerating. It is also further away than anyone selling it wants you to believe. Hold both. That is what honest looks like here.

Sources

Comments