PPPL says the fusion design problem is geometry. Its supplied numbers stop at 4 million amps.

The Princeton Plasma Physics Laboratory has published a plain-language explainer arguing that, alongside the machinery and the temperatures, working out the best design for a fusion power plant comes down to geometry ^[19]. Taken at face value, that relocates the constraint in reactor design away from what magnets and steel can be made to do and towards how well anyone can search a space of shapes, which is a simulation and compute budget rather than a fabrication one. That second step is my inference, not the lab's claim.

The lineage PPPL offers is a list of shapes. Lyman Spitzer Jr., who founded the lab, built a figure eight ^[1][2]. The tokamak arrived in the 1960s as a doughnut that confines plasma using a central electrical current to generate the confining magnetic fields ^[3]. Other devices have been straight lines or twisty crullers ^[4]. The spherical tokamak is a doughnut compressed until the central hole is far narrower, closer to a cored apple ^[5], and PPPL says the shape has properties that could confine plasma energy more efficiently than a conventional tokamak ^[6].

The economics sit in one ratio. Beta is plasma pressure divided by magnetic pressure, and a high beta has been a long-standing goal because it indicates efficient use of the confining field ^[7]. Strong fields are expensive to produce, so a high-beta plant reaches the pressures it needs with less field, easing one of the costliest engineering demands ^[8]. The other lever is volume. According to Jack Berkery, deputy director for NSTX-U research, a larger-diameter conventional tokamak means much more steel, concrete and copper, so the relatively compact spherical tokamak is theoretically cheaper to construct ^[17][18].

The test article is the National Spherical Torus Experiment-Upgrade, PPPL's primary fusion experiment, the largest spherical tokamak in the United States and designed to be the most powerful in the world ^[9]. Its central magnet bundle is specified for up to 4 million amps to produce a field of 1 tesla ^[10]; PPPL's own comparisons are a 30,000-amp lightning bolt and a field 20,000 times Earth's surface value ^[11], which puts the magnet current at roughly 130 lightning bolts ^[20]. The stated aim is the highest plasma stored energy of any spherical tokamak to date ^[14], with diagnostics for plasma temperature and density ^[15] and measurements of the heat flows a plant would have to survive ^[13].

The compute half of the argument is where the material thins. PPPL says NSTX-U will use AI systems to improve performance ^[12], and that as an international user facility it will host public and private users testing how materials and components hold up against plasma and establishing trusted AI tools for real-time data analysis and control ^[16]. That is machine operation, not design-space search. Nothing in the supplied text costs a simulation campaign, counts design iterations, names a target beta, or gives an operating date or price ^[21]. The explainer also promises three categories of spherical tokamak advantage, small size and better performance among them, and the supplied text runs out inside the second ^[22][23].

Watch whether the stored-energy campaign produces beta figures a design code can actually consume ^[14][7], whether the AI tools validated on NSTX-U are ever pointed at geometry optimisation rather than real-time control ^[16], and whether the third advantage arrives with a number attached ^[23].