A blueprint for scalable fusion power

The first thing the episode does, and the most useful, is take apart a milestone that gets reported as one thing and is really two. At the national ignition facility at Livermore, roughly 300 megajoules were stored in a capacitor bank, a laser drove about 2 megajoules into a fuel target, and about 5 megajoules came back out, since improved to nearer 8. Draw a box around the target and you have more energy out than in, which is ignition, and it is a real and hard-won result. Draw the box around the entire machine and you recover roughly 1.5% of what was stored. The milestone still ahead, which von Muench calls net facility gain, is getting more energy out than everything required to run the machine, and she defines “everything” as all the energy stored in the system. Nobody has done it. Several companies say they will within a couple of years; Pacific Fusion says it is on track for 2030. Practical power, by her estimate, needs roughly 5x on that whole-machine basis, which she calls a ballpark that depends on the efficiency of the supporting systems.

Why the Livermore machine cannot be scaled into a power plant is the hinge. It was built for a different mission and was never designed to return more energy than it drew, so its driver, the machinery that delivers energy to the fuel, is relatively inefficient. Her argument is that the fuel does not care how it reaches ignition conditions, so a more efficient driver can reach the same physics with far less loss. Pacific Fusion uses a capacitor bank to make a large electrical current that compresses the target directly, removing the lossy laser step in between. That places them in the inertial branch of the field, alongside Livermore’s laser-driven work and Sandia’s pulser-driven work, rather than the steady-state branch of tokamaks and stellarators, whose best experimental result dates from a 1990s tokamak and sits a factor of a few short of the scientific threshold. Approaches outside those two families, she says, are a factor of a thousand or more away.

The cost argument is where the episode gets genuinely interesting, because Kann puts the strongest available objection and does not entirely get an answer. Von Muench’s case is modularity: the driver holds most of the capital cost and footprint, and it consists of 156 identical modules, each roughly a shipping container in size, each producing more than a terawatt of peak power, and each built from oil, plastic, metal and water rather than anything rare. Cheap materials keep the required markup over the bill of materials achievable with today’s manufacturing, and identical units mean you iterate from one generation to the next quickly. Kann’s objection is drawn from solar and batteries, where exactly this happened: the modular cells got dramatically cheaper, and then everything else, the balance of the plant and the soft costs of building it, came to dominate installed cost because it never fell as fast. So how much of the system is the modular part? She answers that it is much more than 10% and a very large share today, then declines to give a percentage, calling it premature because balance-of-plant equipment is in a seller’s market and the numbers are moving. The strongest part of her reply is a different mechanism entirely: in an inertial system, nameplate output is set by gain per shot multiplied by shot rate, so improved fuel targets and chamber upgrades can raise the output of a facility that already exists, lowering its effective cost without building anything new. Whether that outruns the balance-of-plant problem is not resolved here.

Two things separate the blueprint from a power plant, and she names both. The demonstration system will fire about one shot per day; a power system has to fire about once a second, and the pulsed hardware has to survive that for thirty years, which she calls a meaningful reliability engineering challenge and expects to work out at the module level before committing to a plant. And every fusion design, hers included, needs what she calls a sustainable tritium economy: a small starting inventory of tritium, and an operating plant that breeds more of it than it burns, so it can start the next one. She describes this as a field-wide unsolved problem with people working on it, differing across approaches only in how much tritium is needed to start and how efficiently it can be bred. That leaves the financing, which is the part most transferable outside fusion. Pacific Fusion raised roughly a billion dollars structured as capital called against pre-agreed technical milestones, a model borrowed from biotech, and von Muench is explicit that it works only because two conditions hold: the market has high conviction about what the end milestone is worth, and no conviction about what the intermediate ones are worth. Underneath that sits the company’s own assessment that its path carries a large execution burden but little binary scientific risk. That assessment is the load-bearing claim of the whole episode, and it is the company’s own.