⚡ Energy
Fusion Reactors Could Be 20% Smaller Than Planned. Two New Papers Explain Why.
Simulations of ITER, SPARC, and ARC show that fusion-born alpha particles suppress plasma turbulence instead of worsening it, boosting output by 18–25%. Run the numbers backward through tokamak scaling laws and each commercial reactor could shed $600 million to $1.2 billion in construction costs.
Twenty-five percent. That's the number that landed in fusion physicists' inboxes on May 11, buried inside a 16-author simulation paper posted to arXiv. According to Di Siena et al., fusion-born alpha particles in SPARC don't just heat the plasma. They actively suppress turbulence, creating a self-reinforcing feedback loop that increases the reactor's fusion output by a quarter beyond what standard models predict.
A month later, a separate group at MIT and Commonwealth Fusion Systems confirmed the same mechanism in ARC, the commercial reactor SPARC is designed to validate. The implications extend beyond a nice bump in Q factor. Work the numbers backward through tokamak scaling laws, and each commercial fusion plant could be meaningfully smaller, cheaper, and closer to grid parity than anyone's current financial model assumes.
What Actually Happened in the Simulations
Fusion reactions produce alpha particles, which are helium-4 nuclei carrying 3.5 MeV of kinetic energy apiece. In a deuterium-tritium reactor, these alphas are supposed to be the self-sustaining spark: they deposit their energy into the surrounding plasma, keeping it hot enough to fuse more fuel without external heating, and that part was always the plan.
What wasn't in the plan was this: the alpha particles weakly destabilize a class of plasma waves called toroidal Alfvén eigenmodes, or TAEs, and for decades, physicists worried this destabilization would scatter the alphas outward, robbing the plasma of their energy and degrading confinement so severely that billions of dollars in reactor design have been spent with that fear as a background assumption.
Di Siena's team at the Max Planck Institute for Plasma Physics ran the first fully self-consistent simulations that evolve microturbulence, alpha-particle heating, and macroscopic plasma profiles simultaneously to steady state, something nobody had done with all three coupled together before. What emerged was counterintuitive: the TAEs, rather than scattering alphas outward, nonlinearly enhance zonal flows, which are large-scale plasma currents that shear apart smaller turbulent eddies. Less turbulence means less heat escaping the core, which means a hotter core, which means more fusion, which means more alphas, establishing what amounts to a positive feedback loop rather than the loss mechanism the field had feared for decades.
"What we see is that you can enter in a type of positive feedback loop," Di Siena told Science News. The effect increased alpha-particle heating by up to 25% in SPARC and up to 18% in ITER.
Independent Confirmation Arrived in 34 Days
On June 14, Hall et al. posted a separate study using CGYRO gyrokinetic simulations of an ARC-class fusion power plant, built with different simulation code by a different team at MIT and CFS, and arrived at the same finding: "significant reduction in ion-scale turbulent heat and particle fluxes" in the inner core, driven by multiscale interactions between fast-ion-destabilized modes, zonal flows, and background turbulence. They found the suppression scales beneficially with alpha particle density and plasma βe, meaning the effect gets stronger in more reactor-relevant conditions.
This wasn't just prediction confirming prediction in a vacuum, because experimental evidence had been accumulating quietly at two different facilities. A 2024 study at the Joint European Torus in England, during its final deuterium-tritium campaign before decommissioning, observed alpha particles driving axisymmetric Alfvén eigenmodes at the plasma edge, consistent with the simulated mechanism. A 2025 study at the DIII-D tokamak in San Diego found similar turbulence effects to Di Siena's simulation. Neither experiment could replicate full burning-plasma conditions, but both pointed in the same direction.
"Maybe this thing, which seems sort of magical and fanciful, could really work positively," said William Heidbrink, a plasma physicist at UC Irvine who was not involved in either study.
Translating Physics Into Steel and Concrete
Here's where it gets interesting for anyone writing checks. Tokamak fusion power scales approximately as Pfusion ∝ β² × B⁴ × R³ × κ, where R is the major radius, B is the magnetic field strength, β is normalized pressure, and κ is plasma elongation. The key variable for cost is R, because nearly every reactor subsystem scales with the size of the machine.
If alpha particles deliver 25% more fusion power than current models predict from the same plasma geometry, reactor designers face a choice. Option one: accept the bonus and get more power from the same reactor. Option two: build a smaller reactor that hits the same power target.
Option one is straightforward: SPARC's nominal Q jumps from about 11 to roughly 13.8, ITER's Q rises from 10 to 11.8, and ARC's net electricity output climbs from 400 megawatts to approximately 500 megawatts, enough to power an additional 70,000 homes from the same machine.
Option two is where the economics get dramatic. Holding fusion power constant at 1.1 gigawatts for ARC, a 25% efficiency gain allows the major radius to shrink by about 7%: from 4.6 meters to 4.27 meters. That sounds modest until you remember volume scales as R³. The plasma chamber shrinks by roughly 20%, the first-wall area drops by 14%, and magnet mass follows suit. Tokamak construction costs scale between R² and R³ depending on the subsystem, so a conservative aggregate estimate is 15% lower capital expenditure per plant.
| Reactor | Without Alpha Boost | With 25% Alpha Boost | Delta |
|---|---|---|---|
| SPARC (fusion power) | 140 MW, Q ≈ 11 | 175 MW, Q ≈ 13.8 | +25% |
| ITER (fusion power) | 500 MW, Q = 10 | 590 MW, Q ≈ 11.8 | +18% |
| ARC (net electricity) | 400 MWe from 4.6 m radius | 400 MWe from 4.27 m radius | −20% volume |
| ARC (est. capex, $B) | $4–8B | $3.4–6.8B | −$0.6–1.2B |
For a fleet of 50 ARC-class reactors supplying 20 gigawatts of baseload power, the cumulative savings range from $30 billion to $60 billion. That's real money in an industry where the difference between "investable" and "uninvestable" has historically been measured in dollars per megawatt-hour.
"Everything Is About the Alphas"
Jacobo Varela, a plasma physicist at the University of Texas at Austin, put it bluntly: "If you don't know how the alphas will behave, there is no way to make an economically viable reactor. In a reactor, everything is about the alphas and how they behave." That assessment, made to Science News, captures why this finding resonates beyond academic journals. Every business plan in the private fusion sector contains an assumption about alpha-particle behavior. If that assumption has been too pessimistic, the economics shift, not by some abstract, hand-wavy amount, but by billions of dollars per plant.
CFS, which partly funded the Di Siena study, is already building SPARC in Devens, Massachusetts. The machine is 75% complete with 2 of 18 toroidal field magnets installed. First plasma is scheduled for late 2026, with Q > 1 demonstration planned for 2027. ARC, the commercial follow-on validated in a peer-reviewed study published in the Journal of Plasma Physics this month, is designed to produce 1.1 gigawatts of fusion power and deliver 400 megawatts of net electricity to the grid. Both machines are tokamaks using high-temperature superconducting magnets that CFS says achieve 20 tesla in a package one-fortieth the volume of ITER's conventional superconductors.
If the alpha particle effect holds, SPARC's margin of safety widens considerably. Instead of needing Q > 2 to validate ARC's physics basis, the machine might demonstrate Q > 7 or higher even under conservative confinement assumptions. That kind of headroom changes conversations with regulators, grid operators, and the institutional investors who have historically treated fusion as a charitable write-off rather than infrastructure.
Limitations
Several caveats deserve their full weight. Phil Snyder, vice president of plasma physics at CFS, cautioned that the specific numbers should be taken "with something of a grain of salt." The 25% boost comes from gyrokinetic simulations using specific plasma profiles and assumptions, not from a burning plasma experiment (which has never been conducted at the relevant conditions). No existing tokamak produces enough alpha particles to directly verify the effect at scale. SPARC will be the first machine capable of testing this in a burning plasma, and that test is still one to two years away.
The reactor-sizing calculation presented above assumes the alpha particle effect is the only variable that changes, which is a deliberate simplification. In practice, shrinking the major radius by 7% alters divertor heat loads (higher power density on a smaller target), neutron wall loading (potentially exceeding material limits), and tritium breeding geometry (less blanket volume per unit fusion power), each of which has to be rechecked before anyone pockets the savings. The savings estimate of $600 million to $1.2 billion per plant uses CFS's implied capital cost range from Series B financing disclosures and scales it by R² to R³, which is standard tokamak costing methodology but has never been validated against an actual commercial fusion reactor because none exist.
The two arXiv papers also share methodological limitations. Both use local gyrokinetic simulations, which calculate turbulence in a slice of the plasma and may miss global effects. Hall et al. note that "the suitability of local gyrokinetics" at the affected radii is an open question. And the Di Siena paper's feedback loop depends on the alpha particles remaining well-confined; if anomalous transport at the edge turns out to be worse than modeled, the loop breaks.
Strongest Counterargument
The strongest case against getting excited about these results comes from the history of fusion modeling itself. Tokamak simulation has a long record of predicting phenomena that fail to materialize at predicted magnitudes when hardware catches up. The "isotope effect" in tokamaks, where deuterium-tritium plasmas were expected to confine dramatically better than pure deuterium, turned out to be roughly half as large as simulations suggested when JET ran its DTE2 campaign in 2021. ITER's own physics basis document has been revised downward multiple times as experimental reality accumulated.
The alpha particle turbulence mechanism is particularly hard to validate incrementally because it requires a burning plasma, and you cannot scale down a burning plasma the way you can scale down a transformer model: either the alphas dominate the heating, or they don't, with no halfway experiment possible. SPARC is designed to cross this threshold, but if the machine encounters unexpected issues with plasma startup, magnet quench management, or tritium handling, the alpha particle question remains theoretical for another decade. The physics community has learned, painfully, that simulations that agree with each other are not the same as simulations that agree with reality.
What You Can Do
If you're a fusion investor: ask your portfolio companies whether their performance projections include or exclude alpha-particle turbulence suppression. If they exclude it (as most current models do), the upside case just got materially larger. If they include it, demand to see which simulation code they're using and what boundary conditions they've assumed, because the effect is profile-dependent and disappears outside the inner half of the plasma.
If you're an energy policy analyst: update your fusion LCOE models. Most published estimates (including ours from April) use Q = 10 as the baseline physics assumption. A Q ≈ 12 scenario with alpha-particle turbulence suppression shifts the break-even LCOE from roughly $90–130/MWh to $75–110/MWh, narrowing the gap with nuclear fission ($70–100/MWh) and potentially crossing into competitiveness with offshore wind ($80–120/MWh) for baseload applications.
If you're watching from the sideline: SPARC's first plasma, expected in late 2026, will be the most important data point in fusion physics in decades. Not because it will settle the alpha-particle question immediately (the machine runs in hydrogen and helium initially), but because the subsequent D-T campaign in 2027 will be the first direct test of burning-plasma turbulence dynamics. Watch the Q number. If SPARC demonstrates Q > 5, the alpha-particle feedback loop is almost certainly real. If it demonstrates Q > 10, every commercial fusion timeline shortens.
The Bottom Line
Fusion's biggest open question, how alpha particles interact with plasma confinement, has received its most comprehensive answer yet. Two independent simulation groups find the same mechanism: alpha particles don't degrade the plasma. They strengthen it. The magnitude is debatable, the direction isn't. For an industry that has spent $10 billion without generating a single commercial kilowatt-hour, a 25% improvement in the core physics could be the difference between "perpetually 20 years away" and a working power plant before 2035. We'll know when SPARC burns D-T fuel next year.