Fusion's Economics, Trapped Fuel, and 100-Millisecond Warnings

A Formula for Whether a Fusion Plant Can Actually Make Money

Nobody has built a commercial fusion power plant yet — but a research team just asked: when someone does, how will we know it was worth the money?

Before you open a restaurant, you run the numbers: will revenue beat costs? Fusion plants face the same question, and surprisingly nobody had written it down in a clean, universal way — until now. A team of researchers has derived something they call Q_econ, an economic gain factor for fusion power. Think of it like a renovation contractor's break-even rule: if the value of energy you sell over the plant's lifetime doesn't exceed what it cost to build and maintain the thing, you're running a very expensive charity. Q_econ ≥ 1 means you break even or better. Here's what makes this useful: the framework doesn't care how big your plant is, or which fusion technology you're betting on — magnetic confinement, laser-driven, or hybrids. The team normalised all costs and revenues to the surface area surrounding the fusion zone, measured in square metres. That makes the maths technology-agnostic. They identified ten parameters that control whether Q_econ clears the bar. The big ones: how long the plasma-facing components last before they need replacing, how efficiently you convert fusion energy to electricity, and something called the utilisation factor — the fraction of time the plant is actually running versus sitting idle for repairs. A plant that spends half its life being fixed is like a taxi that's always in the shop. It has a perfect engine and zero revenue. The catch: this is pure theory. There is no experimental data behind it, no real plant to validate against. It is a checklist, not a guarantee. The hard work is filling in those ten numbers from a machine that doesn't exist yet — and right now, for most of them, we're still guessing.

Simulations Show ITER's Tritium Fuel Will Hide in the Walls

Thirty-five grams — about the weight of seven nickels — of radioactive fuel, buried in the reactor walls after just ten days of operation.

Tritium is the rare, mildly radioactive form of hydrogen that powers fusion reactions alongside deuterium. You have to breed it inside the reactor, use it, and then recover as much as possible — both because it's scarce and because letting it accumulate in the structure is a regulatory headache. A team using a simulation framework called HISP modelled what happens to tritium inside ITER after ten days of full deuterium-tritium operation. The picture is sobering. About 35 grams end up stuck in the first wall and divertor — the components that face the plasma directly. Think of flour dust in a kitchen: it works its way into every crack, groove, and textured surface, and the more irregular the surface, the more it holds. About 80% of that trapped tritium doesn't sit in the tungsten metal walls themselves — it lives in thin boron layers that get deposited on surfaces over time. Boron is routinely sprayed into the plasma chamber to keep things clean, but it ends up coating every surface and trapping fuel hydrogen atoms inside it. The good news: baking the reactor — heating it up to drive out trapped hydrogen, like warming a damp sponge — removes about 88% of tritium from the tungsten components. The bad news: baking only removes about 30% from the boron-coated regions. And that's where most of the inventory lives. The catch: this is a simulation, not a measurement from a real DT machine. ITER has not run deuterium-tritium plasmas yet. These numbers depend on the model's assumptions about how boron layers form and how hydrogen diffuses through them. The real inventory won't be known until the machine actually runs — and that's still years away.

A Neural Network Spots Plasma Eruptions 100 Milliseconds Early

A tenth of a second sounds trivially short — but in a fusion reactor, it is enough time to fire a mitigation system and stop serious damage.

Fusion plasmas are held in a magnetic cage, but the edge of that cage is unstable. Periodically, the plasma edge throws off bursts of energy — called ELMs, or edge-localised modes — that slam into the reactor wall. In a machine like ITER, these repeated impacts could gradually erode the surface of expensive components. Catching them early is like having a smoke alarm in your kitchen: if you know the fire is starting, you can act before it spreads. A team working on the DIII-D tokamak at General Atomics in San Diego trained a neural network to predict the very first ELM after the plasma jumps into its high-performance state — a notoriously tricky moment to manage. Their model uses a deep-learning architecture called ResNetTransformer. It analyses 50 milliseconds of radar-like measurements of the plasma edge — a diagnostic called Doppler backscattering, which bounces microwaves off density ripples in the plasma to track turbulence. The model then outputs a probability: how likely is an ELM crash in the next 50ms, 100ms, or 150ms? The early results show the model can issue a reliable warning roughly 100 milliseconds before the crash. In practice, that is enough lead time for automated systems to respond — injecting gas or adjusting the magnetic field to soften the blow. The catch: this is an early proof of concept. The paper, which is still a preprint, does not yet publish the full performance numbers — false alarm rate, precision, recall — that would let you judge how trustworthy the system really is. And a model trained on DIII-D would need retraining before it could work on a different machine. The plasma physics changes with every device.

What connects these three papers? Each one is pressing on a problem that has to be solved not for fusion to be physically possible, but for it to be practically real. The economic framework asks: even if we achieve a burning plasma, will the plant ever pay its bills? The tritium simulation confronts a specific operational constraint: if you cannot recover your rare fuel efficiently, you cannot run continuously — and tritium that stays stuck in the wall is tritium you cannot burn. The ELM prediction work tackles the chronic wear problem: plasma eruptions that slowly grind down the surfaces you spent billions building. None of these papers announces a discovery. Together, they tell you the fusion community is starting to take the engineering and operational layer seriously — the layer that sits just above 'can we make it work' and asks 'can we make it work for thirty years at a profit.' That is a shift worth noticing.

ITER's first deuterium-tritium plasmas remain the biggest upcoming test of all three of these questions at once — but that milestone has slipped to the early 2030s, so don't hold your breath. In the nearer term, watch for updated ELM prediction results from DIII-D and the ASDEX Upgrade experiment in Germany, where disruption mitigation campaigns are ongoing through 2026. The question I'd most like answered: how much of that boron-trapped tritium can actually be removed without thermally damaging the surrounding wall structure — and does anyone have a better idea than baking?

• Wikipedia: Tritium — background on why fusion fuel recovery matters
• Wikipedia: Edge-localised mode — what ELMs are and why they're dangerous
• Wikipedia: Lawson criterion — the physics analogue of the economic gain factor derived in Story 1