On 5 December 2022, a fusion experiment at the National Ignition Facility in California produced 3.15 megajoules of fusion energy from 2.05 megajoules of laser input. This was correctly reported as a scientific milestone — the first time a fusion reaction on Earth had produced more energy than was delivered to its target. It was also, less correctly, reported as the breakthrough that changes everything about energy. The lasers delivering the 2.05 MJ required roughly 300 MJ of electricity to produce. Net energy efficiency, including the energy cost of the apparatus, was well under one percent. This is not a criticism of the experiment, which did exactly what it was designed to do. It is a criticism of the coverage, which mostly did not.
This page treats two stories together because they are, increasingly, one story. The fusion story is about an energy technology that might arrive in decades and is surrounded by hype that outpaces its current capabilities. The grid story is about the nuclear-and-renewables transition that is arriving now, driven by an energy demand curve that nobody projected: the electricity consumption of AI data centres.
Fusion’s hype-vs-reality moment. NIF ignition was one of a series of developments since 2018 that have made fusion feel close. Private fusion startups have raised more than $7 billion in total. Commonwealth Fusion Systems (the MIT-spinout using high-temperature superconducting magnets) has raised nearly $3B across ten rounds and is building its SPARC demonstration reactor. Helion Energy has an agreement with Microsoft to deliver fusion electricity to the grid by 2028 — a date most fusion physicists consider aspirational. TAE Technologies and Trump Media & Technology Group announced an all-stock merger agreement in December 2025 valued over $6B, with closing expected mid-2026 — making TAE one of the first publicly traded fusion-focused companies if the deal completes. These are real commitments of real capital. They are not, yet, grid electricity.
The SMR story is more mixed than the press cycle suggests. Small Modular Reactors were, in 2018, the most-hyped near-term nuclear solution. NuScale Power’s Idaho project — meant to be the first US SMR plant — was terminated in 2023 after costs escalated from $4.2B to $9.3B and the anchor customer withdrew. The collapse was widely reported as an SMR death knell. The picture is messier than that: NuScale has subsequently signed a 6 GW deployment agreement with TVA and Entra1 (September 2025), and other SMR vendors are progressing. TerraPower’s Natrium project (Bill Gates’s sodium-cooled fast reactor) has been delayed by HALEU fuel supply disruptions but is now targeting grid connection by 2031.
AI data centres are reshaping grid demand. Data centres accounted for roughly 4% of US electricity use in 2024; projections suggest this could double by 2030, driven primarily by AI training and inference workloads. This has produced a suddenly-serious set of private-sector commitments to new nuclear capacity. In September 2024, Microsoft and Constellation Energy agreed to restart Three Mile Island Unit 1 — the less-famous twin of the reactor that partially melted down in 1979 — under a 20-year power purchase agreement. Amazon, Google, Meta, and Oracle have announced nuclear procurement arrangements of their own. This is the largest private-sector pro-nuclear development in a generation, and its driver is not climate policy. Its driver is AI.
Geoengineering is now an energy story too. The book’s treatment of climate runs through The Day After Tomorrow and the Geoengineering page, which focus on climate intervention. The missing half of the climate conversation is the energy transition itself — the question of how rapidly and in what form we move off fossil fuels. The P18 pages on Active Geoengineering Proposals and Carbon Removal and Climate Tech sit on the intervention side of this story; this page sits on the decarbonisation side.
Hype vs. Reality is the book’s most directly applicable framework. Fusion is the cleanest current case for the book’s discipline of counting assumptions. The chain from “NIF ignition” to “commercial fusion electricity on the grid” contains: reactor designs that achieve net gain at the plant-wall level; materials that survive decades of neutron flux; tritium breeding and fuel-cycle closure; plant-scale engineering that has not yet been attempted at fusion relevant scales; regulatory frameworks that do not yet exist. Every one of these steps is genuinely hard. None of them is the step NIF completed. The book’s framework does not say fusion will not arrive. It says that “arrived” should be applied to the specific thing that has happened, and that the gap between “scientific breakeven at one facility” and “fusion on the grid” is filled with unreduced engineering and policy uncertainty.
Too Valuable to Fail applies to the current grid in its most entrenched form. The combustion-electricity infrastructure has been built, subsidised, and regulated into the conditions of modern life over more than a century. Replacing it is not a technology problem primarily. It is a coordination, financing, siting, and political problem — all four of which are where energy transitions actually succeed or fail.
Intergenerational Responsibility runs in two directions. Nuclear waste that will remain hazardous for 100,000 years is the conventional case; the book’s treatment of climate intergenerational obligations also applies. There is a symmetry here that is rarely acknowledged: the intergenerational cost of nuclear waste is real, and the intergenerational cost of not decarbonising fast enough is also real. Honest policy engages both.
The AI-driven nuclear revival raises a specific question the book’s frameworks engage well: who decides what the electricity is for? A hyperscaler’s willingness to bankroll new nuclear capacity to power AI inference is, in some sense, a positive climate story — it is new firm clean power that would not otherwise have been built. It is also a distributional story: the fossil electricity that the grid would otherwise need to serve AI demand is not, automatically, being displaced. The counterfactual determines whether this is climate progress or simply private-sector demand that happens to prefer nuclear.
What the book directly addresses. The Day After Tomorrow chapter treats climate at the system level; the Geoengineering page treats planetary intervention; Complexity and Unintended Consequences, Risk Innovation, Too Valuable to Fail, and Hype vs. Reality all apply directly.
What the frameworks suggest when extrapolated. The book’s general pattern of “who decides, who benefits, who is harmed” applies cleanly to the grid transition, including the AI-demand dimension. The specific question of how to reckon with hyperscaler electricity demand that is neither purely private nor purely public in its externalities is a good candidate for the book’s frameworks, though the book does not take it up.
Where the frameworks reach their limits. The specific engineering questions — what fusion concepts are likely to succeed, what SMR designs scale, what grid architecture optimally integrates intermittent renewables with firm clean power — are technical policy questions the book’s frameworks are not designed to answer. The contribution here is diagnostic, not prescriptive.
Films outside the book’s twelve: The China Syndrome (1979, James Bridges) is the cultural reference for nuclear anxiety, and worth revisiting as the political valence shifts. Chernobyl (HBO series, 2019) is television rather than film, but its treatment of institutional failure is the most serious recent cinematic engagement with nuclear. Pandora’s Promise (2013, documentary) presents the pro-nuclear environmentalist case; it is worth pairing with more critical treatments.