Commonwealth Fusion Systems: the best-capitalized bet on a thing that has never worked

The original problem. Fusion power has been "thirty years away" since the 1950s. The specific bottleneck for tokamaks is magnetic field strength: confinement scales steeply with field, and conventional superconducting magnets cap out.

Founder background and the actual insight. CFS spun out of MIT's Plasma Science and Fusion Center in 2018, founded by Bob Mumgaard, Dennis Whyte and colleagues. The insight was not about fusion physics; it was about materials. High-temperature superconducting (REBCO) tape had become commercially available, enabling magnets roughly twice as strong as ITER's. Because fusion power density scales with the fourth power of magnetic field, a doubling of field allows a machine roughly 1/40th the volume of ITER. The company's bet was that a decades-old physics program could be shrunk by a new component.

Validation — and this is the model for deep tech. In September 2021, CFS and MIT demonstrated a 20-tesla large-bore HTS magnet, the key enabling component, before building the reactor. That is validation done correctly: isolate the single riskiest technical assumption and test it as cheaply and as early as possible.

Funding history. Over $2 billion raised in total, including a $1.8B Series B (2021), an $863M Series B2 in September 2025 — which the company described as "the last funding we'll raise before SPARC reaches its first key milestone" (CFS blog, 2025) — and a further $1B raised in 2026, notably from pension funds, a first for the fusion sector (TechTimes, July 2026). Investors include Breakthrough Energy Ventures (Gates), Eni, Google, Temasek and Nvidia's venture arm.

Business model — and the thing that changed everything. CFS signed a power purchase agreement with Google for output from ARC, its first commercial plant: a 400MW facility in Chesterfield County, Virginia, developed with Dominion Energy. A PPA is a contract to buy electricity that does not yet exist from a technology that has never worked. It is the single strongest available evidence of demand, and it exists because AI data centers have made large, firm, carbon-free baseload power scarce enough that hyperscalers will underwrite unproven supply.

Current status (September 2026). SPARC, the demonstration tokamak in Devens, Massachusetts, is in assembly and targeting first plasma and then Q>1 (net fusion energy gain) — the milestone no magnetic-confinement device has achieved. ARC is under parallel development, targeting the early 2030s. No fusion company has yet produced net electricity for the grid, and CFS has no revenue from power.

Lessons that generalize.

1. De-risk the single hardest component first, in isolation. The 20-tesla magnet test is the deep-tech equivalent of a landing page test, and it unlocked $1.8B.
2. A creditworthy offtake contract is the strongest validation a pre-revenue hard-tech company can obtain — stronger than any amount of investor enthusiasm, because the counterparty is buying the output, not the story.

---