The artificial-intelligence industry's appetite for electricity has become so voracious that investors are now writing nine-figure checks for technology that has been "twenty years away" since the Eisenhower administration. Thea Energy, a Princeton University spinout working on stellarator-based fusion, has closed a $100 million round that vaults it into the top tier of fusion startups globally—a cohort that barely existed five years ago and now commands billions in committed capital.
The timing is not coincidental. Every major hyperscaler is scrambling to secure power for GPU clusters that can draw more wattage than small cities. Microsoft has signed a deal to restart Three Mile Island's dormant reactor; Google is buying geothermal; Amazon is snapping up nuclear-adjacent sites. Fusion remains speculative, but the calculus has shifted: if you believe AI training runs will keep doubling in energy demand every year or two, the option value of a genuinely clean, nearly limitless power source starts to look cheap at almost any price.
Why stellarators, why now
Fusion research has long been dominated by tokamaks—doughnut-shaped machines that use powerful magnets to confine superheated plasma. Stellarators twist that doughnut into a more complex geometry, which makes them harder to build but theoretically easier to operate continuously. For decades the engineering challenges kept stellarators on the academic fringe. Thea's pitch is that modern computational tools—including, yes, AI-assisted design—have finally made the geometry tractable at commercial scale.
The company's founders emerged from Princeton's plasma-physics program, and their approach relies on arrays of planar coils that can be manufactured with existing industrial techniques rather than bespoke superconducting magnets. Whether that simplification survives contact with a 100-million-degree plasma remains to be seen, but the thesis is appealing enough that the round reportedly attracted both traditional energy investors and tech-adjacent funds hunting for AI infrastructure plays.
The competitive landscape
Thea is hardly alone. Commonwealth Fusion Systems, TAE Technologies, and Helion have all raised comparable or larger sums, and Helion has a power-purchase agreement with Microsoft for electricity that does not yet exist. The sector's collective funding now exceeds $6 billion, a figure that would have seemed hallucinatory a decade ago. Skeptics note that none of these companies has achieved net energy gain outside a laboratory setting, and that even optimistic timelines put commercial power plants in the mid-2030s at the earliest.
But the AI power crunch is rewriting the risk calculus. Utilities in Nevada, Pennsylvania, and California are already turning away new data-center connections because the grid cannot keep up. If fusion delivers even a few years earlier than expected, the first movers will own infrastructure worth multiples of their R&D spend. If it doesn't, the sunk cost is a rounding error for funds managing hundreds of billions.
Our take
Fusion has been the punchline of energy investing for so long that it is tempting to dismiss Thea's raise as another triumph of narrative over physics. But the narrative has changed. The AI industry's power problem is real, immediate, and worsening, and it has made patient capital suddenly very impatient for alternatives. Stellarators may still be a long shot, but the pot odds have improved—and in a world where a single frontier model training run can cost north of $100 million in electricity alone, betting on moonshots is starting to look like prudent portfolio management.
