An interesting article was published this week on LiveScience discussing a possible synergy between quantum computing and nuclear fusion: two topics we’ve frequently covered on our website’s in-depth analysis page. Apparently, both topics share a single element: their risk profile. In fact, in both cases, it’s a gamble that can be summed up in one sentence: “moving out of the labs and into the world of industry.” In reality, the two fields have something more in common: the potential for one to act as a catalyst for the other in a virtuous cycle of interaction. To understand this interesting development, we invite readers to first read the report on fusion available for download on our website and then continue reading the text below, which may be supplemented by a new in-depth report also available for download from the site.
Quantum computing does not yet accelerate fusion, but it has identified a concrete problem to tackle: simulating the chemistry of the salt needed to produce tritium. To be precise, the simulations aim to make the most of the best advanced computing solutions available to us: in addition to quantum computing, these include supercomputer technology and artificial intelligence, which—compared to quantum computing—have the major advantage of already being operational realities.
However, the synergy between quantum computing and nuclear fusion represents a far more promising challenge for a long-term investor because it emerges as an “option on an option”—one that is only valuable if fusion becomes commercial and if quantum computing outperforms classical methods precisely where it matters most.
Why Fusion Has a Computational Problem
No facility has ever been continuously exposed to the heat and radiation of a fusion reaction, so much data is missing and can only be obtained by building expensive machines or through simulation. There are three bottlenecks:
• Tritium: The reactor must produce more than it consumes (a tritium breeding ratio greater than 1), but tritium breeding has only been demonstrated in the laboratory.
• Materials: Mohamed Abdou (UCLA) estimates that first-generation components could fail on average every 20 minutes, with repairs taking “many months.”
• Plasma: It is inherently unstable, and keeping it stable over the long term remains one of the most difficult challenges.
The IBM–ORNL Case
On June 29, 2026, Oak Ridge, the Cleveland Clinic, and IBM used quantum hardware to calculate nine configurations of FLiBe, a molten salt of fluorine, lithium, and beryllium for the blanket (the wall surrounding the plasma). Each configuration is a cluster of 21 ions. The crux of the matter is what happens to the tritium: if it binds to fluorine, it forms a corrosive compound; otherwise, it escapes as a gas. DFT, the standard classical method, miscalculated the salt’s free energy by up to 10%, and that’s where quantum computing comes in. The result is a proof of concept, not a demonstrable advantage: the nine calculations match the classical results, and the preprint has not yet been peer-reviewed. A real blanket has about 10²⁴ particles, and the study tests only one of the three stages of the workflow (AI agents, supercomputers, quantum computers).
Tritium by the Numbers
The world produces about 20 kg of it per year: enough for about 44 days of operation at a 1 GW plant, which would consume about 165 kg per year—eight times global production. With a half-life of 12 years, it decays by 5.6% per year; therefore, it cannot be stockpiled and must be produced within the reactor itself.
Where quantum computing helps and where it doesn’t
• Blanket chemistry: the most credible contribution, currently being tested.
• Materials under irradiation: no evidence found in the sources consulted.
• Plasma: theoretical; in the nonlinear case studied on Quantum, the algorithm is more computationally expensive than the classical one.
• Inertial fusion: requires approximately 1,000 logical qubits and 10¹⁵–10¹⁷ gates, 6 to 9 orders of magnitude beyond IBM’s targets.
• Real-time plasma control: no plausible role.
Two Pathways
The IBM roadmap calls for Starling, a 200-qubit fault-tolerant system, in 2029: 8–10 years before fusion pilot plants, expected in the late 2030s. Quantum computing could therefore arrive in time for the final iterations on materials, though not necessarily for the initial selection. However, both are planned dates, and in fusion, delays are the norm: General Fusion has shifted its target from 2022—promised in 2014—to 2035.
A Shared Supply Chain
The two sectors do not share a product but rather industrial expertise: cryogenics, superconductors, and vacuum technology. The most tangible link is helium-3, the theoretical fuel for D-He3 fusion and a component of quantum refrigerators: Bluefors has agreed to purchase up to 10,000 liters per year from Interlune, with deliveries from 2028 to 2037. The connection to AI is also financial rather than physical: Chinese data centers will reach 774 TWh in 2030 (approximately 88 GW), while the 50 MW Microsoft–Helion contract would cover 0.06% of that.
For the investor
The barbell structure of the fusion report remains unchanged. If anything, it reinforces the “picks and shovels” common to both sectors. Key points to monitor: peer review and replication of the IBM–ORNL results, the transition to larger clusters, adherence to IBM’s milestones (Kookaburra 2026, Cockatoo 2027, Starling 2029), and a genuine demonstration of an advantage in addressing a materials chemistry problem.
Scenarios for 2026–2040
These are the author’s qualitative assessments: “Classical is enough” has a medium-to-high probability; “Useful quantum, fusion delayed” and “Both delayed” have a medium probability; “Convergence” has a low probability.
Risks
The most solid evidence is a preprint on nine configurations. There is a risk of overlapping hype between two sectors that hold great long-term promise, and a model is no substitute for experimentation.
Conclusion
The decisive test will not be another qubit record, but a computer-designed salt that produces and releases tritium in an actual reactor. For a more in-depth technical analysis (perhaps too technical), please download the report.
Disclaimer
This post reflects the personal opinions of the Custodia Wealth Management staff who authored it. It does not constitute investment advice or recommendations, nor does it constitute personalized advice, and should not be considered an invitation to trade in financial instruments.
