I recently spoke with Lennart Bock at Proxima Fusion, and he got me thinking about stellarators. I’d just finished writing about laser fusion — one bet on how to hold a plasma together. Stellarators and tokamaks are a different fork: both use magnetic fields, and disagree about how to shape them.
What I actually wanted to know was which one wins. Turns out that’s the wrong question: the two designs aren’t competing to be better at the same thing. A tokamak buys compactness by driving a current through the plasma, which does most of the confinement work for free. A stellarator refuses that current and shapes the field entirely with a far more complex coil set. Every difference below traces back to that one decision.
Tap a row to see why it matters.
The trade, compressed
What a tokamak buys you: compactness. ARC does with a 3.3 m radius what Stellaris needs 12.7 m for, using coils that are just simple rings, the same shape the industry has built for seventy years. What it costs: the plasma current that makes that compactness possible also creates disruption risk. SPARC’s own team designs the machine around a 1-in-10 chance of one on every full-power pulse.
What a stellarator buys you: no disruptions, by design, and continuous operation instead of pulsing. What it costs: with no plasma current to help confine the plasma, you pay for that confinement entirely in coil complexity and machine size. Those twisted, individually-machined 3D coils are one of the hardest manufacturing problems in fusion.
Everything below is the case for why that trade looks the way it does.
Why Commonwealth Fusion Systems, not ITER
I picked CFS’s SPARC and ARC as the tokamak side on purpose. ITER is a government megaproject: different funding, different timeline, different mission (physics demonstration, not commercialization). CFS is the fairer match: venture-backed, betting on the same high-temperature superconducting magnets to shrink the machine, racing toward net energy this decade. Comparing Proxima to ITER is startup vs. nation-state. Comparing it to CFS is bet vs. bet.
The market is pricing both bets right now, three weeks apart. Proxima raised €411 million ($468 million) on July 7, 2026 at a €2.4 billion valuation, pushing its total raised past €650 million. CFS raised $1 billion on July 30, 2026, pushing its total past $4 billion. Real investors, real money, on both sides of a roughly 5x gap in total capital raised.
Size alone doesn’t tell the story
Stellaris’s major radius (12.7 m) is roughly 3.8x ARC’s (3.3 m), at nearly the same field strength (9 T vs. 9.2 T). That’s not quite the whole trade-off, though: bigger also means more plasma to fuse. Using the standard torus volume formula, Stellaris’s plasma volume works out to roughly 425 m³ against ARC’s roughly 79 m³, about 5.4x more, and Stellaris’s own published fusion power (up to 2.7 GW) is also about 5.4x ARC’s (more than 500 MW). Those two ratios landing on nearly the same number isn’t a coincidence: power output tracks plasma volume and confinement, not radius by itself. At a similar turbine efficiency (ARC assumes 40-50%), the bigger machine is producing proportionally more electricity, not just costing proportionally more to build. Cost per watt, not radius, is the real comparison, and this post doesn’t have the figures to make it.
What’s still true: a tokamak’s driven current does a lot of its confining for it, almost for free. A stellarator has to buy the same confinement entirely through coil shape.
SPARC’s own physics team designs the machine assuming a 1-in-10 chance of a plasma disruption on every full-power pulse. Not a skeptic’s estimate — the tokamak team’s own engineering assumption, baked into how they design the vessel. A stellarator, with no net plasma current, doesn’t have that failure mode at all. That’s the trade Proxima is betting on: give up the compactness a driven current buys you, and the disruption risk that comes bundled with it goes away too.
Is 3.8x real physics, or extra margin?
Both concepts have published, peer-reviewed scaling laws for confinement time, fitted to thousands of real discharges: IPB98(y,2) for tokamaks, ISS04 for stellarators. This isn’t a recomputation of ARC’s or Stellaris’s specific numbers — I don’t have either team’s assumed density or heating power, and neither company has published enough to reconstruct it. It’s what the published laws say happens to any tokamak or stellarator design as you scale it up. Do that geometrically — same field, density, and heating power, just bigger — and IPB98(y,2) says confinement improves as roughly size². Do the same to a stellarator, and ISS04 says it improves as roughly size³ (an exponent of 2.28 on minor radius alone, plus 0.64 on major radius). On paper, a stellarator gets a steeper payoff from bulking up.
But that size² number flatters the tokamak, because it assumes plasma current stays fixed while the machine grows. Real reactor studies don’t do that — current scales up with size too, at a fixed safety factor. Fold in IPB98(y,2)’s own 0.93 exponent on current and the tokamak’s effective size scaling jumps to size^2.9, almost exactly matching the stellarator’s size^2.92. Scale both machines the way an actual design study would, and the tokamak’s size advantage nearly disappears. The compactness isn’t coming from some confinement-efficiency edge — it’s coming from the same driven current that also causes the disruption risk above. Same source, both effects.
That’s a real, sourced number, not two press releases side by side — with the caveat every scaling law like this carries: these are fits to present-day, much smaller experiments, extrapolated to reactor scale with real uncertainty (typically ±20%, even inside the fitted range). They explain the shape of the trade-off. They don’t pin down Stellaris’s exact radius.
What’s published isn’t the same as what’s built
CFS has put a number on ARC’s net electric output: 270 MWe. Proxima’s Stellaris paper reports fusion power (up to 2.7 GW) and thermal power (about 3.1 GW), but no net electric figure. Proxima’s own framing of the paper is that it shows a QI stellarator can satisfy a power plant’s physics and engineering constraints all at once, not that it’s a cost-optimized design point. Those are different kinds of documents, so the two net electric numbers, one published and one not, aren’t quite comparable yet.
Proxima’s own construction timeline, meanwhile, is concrete. In February 2026, Proxima signed a framework agreement with the Free State of Bavaria, RWE, and the Max Planck Institute for Plasma Physics to build Alpha at Garching, a project priced at €2 billion. SPARC is further along, about three-quarters built as of April 2026 with first plasma targeted for 2027. Alpha’s next milestone, the Stellarator Model Coil, is due the same year, ahead of Alpha’s own target of reaching net energy in 2031.
Both bets are still bets
Neither SPARC nor Alpha makes electricity yet. SPARC exists to prove a compact, high-field tokamak can hit strong energy gain despite the disruption risk. Alpha exists to prove a stellarator can hit net energy without ever taking that risk on. Whoever proves their machine works first gets to build the commercial plant with a straight face. Until then, stellarator vs. tokamak isn’t a verdict — it’s two bets on which problem is worth solving first.
If the physics really does converge the way this post argues, the comparison that actually matters is construction and operating cost against energy produced, not radius. Whyte, Lo, and colleagues at MIT just published a framework for scoring exactly that, across any confinement concept, on ten normalized parameters including power density and component lifetime, an economic analog to the Lawson criterion. That’s the right next place to look.
Sources: Proxima Fusion, Stellaris power plant concept announcement (Feb 2025) · World Nuclear News on the Stellaris design · Wikipedia, SPARC (tokamak) · Wikipedia, ARC fusion reactor · Sweeney et al., “MHD stability and disruptions in the SPARC tokamak,” Journal of Plasma Physics · ITER Physics Expert Group, “Chapter 2: Plasma confinement and transport,” Nuclear Fusion 39 (1999) · Yamada et al., “Characterization of energy confinement in net-current free plasmas using the extended International Stellarator Database,” Nuclear Fusion 45 (2005) · Proxima Fusion, €411M funding announcement (July 2026) · CFS, $1B funding announcement (July 2026) · ANS/Nuclear Newswire on the Proxima/Bavaria/RWE/IPP Alpha agreement · Neutron Bytes on Alpha’s €2B cost · CFS, SPARC construction progress (April 2026) · Whyte, Lo et al., “Criteria for the economic viability of fusion power plants,” Journal of Fusion Energy (2026)