Human Tech Tree
Current research2022–2026 · Present (2015 – Oct 2026)

Energy & Industry / Energy

Fusion: Ignition & Records

NIF ignites repeatedly (record 8.6 MJ, gain 4.13); Wendelstein 7-X, WEST and EAST set duration records; capital floods into start-ups.

Open in the interactive tree →

On 5 December 2022 the National Ignition Facility in California for the first time released more fusion energy than the laser delivered to the target (3.15 MJ from 2.05 MJ). Repeats with rising yield followed. In parallel, magnetic fusion machines with high-temperature superconductors are being built that are more compact than ITER.

As of October 2026

NIF reached a target gain of 4.13 on 7 April 2025 with 8.6 MJ yield from 2.08 MJ of laser energy - but the lasers themselves draw about 300 MJ from the grid for that shot, so this is not a power plant. In 2025 EAST (China, 1,066 s in January), WEST (France, 1,337 s in February) and Wendelstein 7-X (43 s long pulse with a record triple product in May) set duration or long-pulse records. In February 2026 Helion reported 150 million degrees C in Polaris and, as the first fusion start-up, operation with deuterium-tritium; the Fusion Industry Association says the sector raised a record $4.48 billion in the 12 months to July 2026 ($14.24 billion cumulative in its surveys, 56 companies).

Open steps

  • Plasma control and disruption avoidance High AI leveragePredict and avoid disruptions and hold burning plasmas steady for minutes, well beyond today's best pulses.
  • Tritium breeding blanket demonstration Medium AI leverageTest blankets that breed more tritium than a plant burns, with extraction systems, in a fusion-relevant neutron environment; none has been built at scale.
  • Materials that survive 14 MeV neutrons Medium AI leverageQualify structural steels and plasma-facing materials after years of neutron damage; no matching neutron source exists yet.
  • Efficient drivers and durable magnets Low AI leverageClose the gap between the 8.6 MJ NIF shot, which drew about 300 MJ from the grid, and a plant: efficient high-repetition lasers, or high-field magnets that last.

Where AI could help

Medium AI leverage. AI helps plasma control, simulation and design search, but magnets, materials, tritium and power-plant engineering are physical challenges.

  • Control plasma shape and instabilities with reinforcement learning trained in simulation
  • Run fast differentiable plasma simulators to explore SPARC scenarios before first plasma
  • Search stellarator and laser-target designs for higher gain
  • Predict disruptions in time to protect the machine

Shown so far

  • In October 2025, Google DeepMind and Commonwealth Fusion Systems announced a partnership using the open-source TORAX plasma simulator, reinforcement-learning scenario search and plasma control for the SPARC tokamak. source
  • In February 2022, a deep-reinforcement-learning controller trained in simulation shaped plasmas on the TCV tokamak without fine-tuning, including negative-triangularity and snowflake configurations. source

Prerequisites

Unlocks

Sources

More in Energy · Present

All 50 points in Energy →

Open in the interactive tree →