The Sun confines its fusion reactor with the gravity of 2 × 10^30 kilograms of matter. Earth-bound fusion has to do the same job without the planet-crushing mass — and that single missing ingredient is why the field has been "thirty years away" for seventy years. The physics of fusion is settled. The engineering problem is confinement: holding a plasma hotter than the Sun's core, densely enough, for long enough, inside a machine that survives the experience.
Three ways to hold a star
Magnetic confinement shapes the plasma with fields, since a 100-million-kelvin gas of charged particles will follow field lines but destroy any material wall it touches. Tokamaks and stellarators are the leading geometries. The flagship is ITER, a 35-nation project in southern France designed to produce ten times the power it puts into heating the plasma. After a 2024 rebaseline, first plasma is now expected in the mid-2030s — a timeline that has slipped before and should be read as a projection, not a promise.
Inertial confinement skips steady-state entirely: crush a peppercorn-sized fuel capsule so fast that fusion completes before the fuel flies apart. On December 5, 2022, the National Ignition Facility achieved ignition — about 3.15 MJ of fusion energy out against 2.05 MJ of laser energy delivered to the target. The honest asterisk: the lasers drew far more energy from the wall than they delivered to the capsule, roughly two orders of magnitude more. Scientific breakeven at the target, nowhere near breakeven at the plug.
Private ventures are betting that new technology shortens the path. Commonwealth Fusion Systems is building SPARC around high-temperature superconducting magnets, which allow far stronger fields in a far smaller machine; Helion is pursuing pulsed field-reversed configurations with direct electricity recovery; TAE is working on beam-driven plasmas with an eye toward aneutronic fuel. As of 2026, none has produced net electricity. That is the current scoreboard, stated plainly.
Why it is this hard
Three constraints compound. The plasma must exceed 100 million kelvin — hotter than the Sun's core, because we cannot match the Sun's density or wait millennia for reactions to accumulate. It must be confined long enough at high enough density for fusion output to exceed losses (the Lawson criterion), while the plasma itself fights confinement with a zoo of instabilities. And the reactor structure must survive years of 14 MeV neutron bombardment that displaces atoms in any material we know how to make — arguably the least glamorous and most unsolved problem of the three.
Not the Sun's fusion
Terrestrial machines do not run the Sun's reaction. The Sun fuses bare protons through the proton-proton chain — a process so slow (its first step waits on a rare weak-force decay) that a kilogram of solar core produces less power than a compost heap. The Sun compensates with volume and gravity-funded patience. We have neither, so we burn deuterium-tritium, the reaction with the fattest cross-section, at temperatures the Sun never needs. The Sun also self-regulates: run hot and the core expands and cools, run cold and it contracts and reheats — a passive feedback loop with 10-billion-year uptime. Every tokamak control system is an attempt to engineer, with sensors and actuators, the stability gravity provides the Sun for free.
Why it matters to a builder
Fusion is the clearest live example of a problem where the science is proven and everything remaining is engineering — and the engineering is the hard part. The lessons transfer. Metrics can mislead: "net energy gain" meant something specific and narrow at NIF, and knowing which boundary a number is measured at is the difference between insight and hype. Constraints compound: temperature, confinement, and materials are not three problems but one coupled system, like latency, consistency, and cost in distributed systems. And nature's reference design shows what to steal: not the Sun's reaction, which we cannot afford, but its self-regulation. The best systems hold their setpoint without a controller.