The real fuel for fusion isn't deuterium-tritium, it's deuterium-deuterium
The D-T route has to breed tritium while it generates power, and both regulation and cost are high; D-D involves no tritium and is the precondition for a scalable commercial power plant, but the triple product has to climb another order of magnitude from 10^21.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
High-temperature superconductors shrink the device to 2%
Low-temperature superconductors have a limited critical magnetic field, so the device must be built enormously large to get enough energy gain, which means long construction cycles and extremely high cost — the levelized cost of electricity could be more than 100 times that of coal power. High-temperature superconductors have a critical field at least twice as high, so at the same energy gain the device volume can shrink by two orders of magnitude, and the build cost also shrinks by roughly two orders of magnitude. The quantitative relation Yang Zhao gives is: the triple product is proportional to the magnetic field to the 3.5 power and to the linear size of the device to the 2.5 power. Double the magnetic field and the linear size can shrink to about 30%, and volume is 30% cubed, about 2%.
— Yang ZhaoHonghuang 70 is the world's first steel ship
The high-temperature superconducting tokamak had only ever been an idea, until Honghuang 70 was completed in 2024 and engineering feasibility was validated at the level of a complete device for the first time. Yang Zhao's analogy: every ship in the past was built of wood, and you say an aircraft carrier must be built of steel, but nobody has ever built a ship out of steel; change the primary material and welding, rust removal and hull design all have to be redone, and before launch nobody can guarantee it won't leak. Honghuang 70 is not a high-parameter device — its triple product is only 10^17, four orders of magnitude short of 10^21, and it has nothing to do with Q value. What it validates is that all systems can run together as designed and light up the plasma.
— Yang ZhaoThe Jingtian magnet pushes a large-aperture field to 21.7 tesla
The hard part of a large-aperture magnet is that it must leave a metre-scale hole in the middle for the blanket and other equipment, whereas most high-field magnets have only millimetre or centimetre apertures. Once the aperture grows, three parameters — force, heating and current density — must be pushed to their limits simultaneously: the ampere force equals magnetic field times current, and adding structural material to bear the force eats into engineering current density, while higher current density means a smaller cross-section and therefore a smaller device. Near 22 tesla a single strand carries over 20 kiloamps, so one nano-ohm of resistance means nearly one watt of heating; total heating must be held to the hundred-watt level, and the resistance of joints and other links must be pushed down to the hundred-nano-ohm level. The previous record was TFMC, built by MIT and CFS at the end of 2021, which just passed 20 tesla; the Jingtian magnet reached 21.7.
— Yang ZhaoBuild the 120-million 70 first, only then dare to touch the 3-billion 170
Honghuang 170 costs about 3 billion RMB. Yang Zhao says that a few years ago a team of four people who had never built a tokamak could hardly have raised that money, nor had the confidence to build the world's highest-parameter device directly. So they built the lower-parameter Honghuang 70 first, with a budget of 150 million and an actual spend of about 120 million, to prove the team had systems engineering capability. The 70's peak field is 3.1 tesla; the 170's peak is 23 tesla, nearly a 10-fold change, and the gulf is that every subsystem's parameters rise. The strategy is to decompose whole-machine risk into subsystem risk: keep the physics design on the conservative method ITER used 30 years ago, so that as long as engineering parameters are met, plasma performance is highly probable; then validate each subsystem with its first prototype — the Jingtian magnet, for instance, is the validation unit for the 170's toroidal field magnet.
— Yang ZhaoThe 380 targets 40,000 to 50,000 RMB per kilowatt
Honghuang 380 is designed to the requirements of a complete demonstration power plant that can run for long periods, with linear dimensions twice the 170's, volume about 10 times larger, and magnetic field raised from 23 tesla to around 29 tesla. Yang Zhao's target is a sale price of 40,000 to 50,000 RMB per kilowatt, which for a 500 MW plant is on the order of 20 to 25 billion RMB, with a pure cost estimate of 100 to 200 billion. The benchmark is China's newly built fourth-generation fission high-temperature gas-cooled reactor, also 40,000 to 50,000 RMB per kilowatt. The logic: fusion fuel is non-radioactive and its products are not long-lived radioactive, so at equal build cost it becomes attractive to nuclear power owners. The money the team really needs to raise internally is mainly to build the 170, about 3 billion.
— Yang ZhaoHelion's route carries very high scientific risk
Helion is also magnetic confinement, but its field configuration is a linear field-reversed configuration (FRC), not a doughnut. Yang Zhao says the highest-parameter FRC device in the public academic literature has a triple product of about 10^17, maybe not even 10^18, four orders of magnitude from 10^21. His analogy: you have flight test data only from 0 to 10 metres, yet you extrapolate to design an aircraft for 10,000 metres altitude — you may not realise the air gets thinner and the temperature drops, and the thing you designed by extrapolating three orders of magnitude may not fly. Whether new physical processes emerge between 10^17 and 10^21 that invalidate the extrapolated design is uncertain, so he thinks problems of this kind are better suited to research institutes and universities.
— Yang ZhaoD-T needs a tritium plant; D-D is the endpoint
Yang Zhao argues the real fuel for fusion commercialisation is deuterium-deuterium, not deuterium-tritium. Tritium is tightly controlled, can be used to make hydrogen bombs, and carries high regulatory and cost burdens; it also cannot exist stably in nature, so it must be bred while power is generated: fusion neutrons hit lithium-6 to produce tritium, requiring a tritium breeding ratio greater than 1 — consume one, produce more than one — and then it must be collected and sent back to the device. That is the tritium plant. Deuterium in seawater is enough for humanity for tens of billions of years, and without tritium there is no need for a tritium plant, so regulation, pressure containment and safety costs are all lower. The price is that D-D is an order of magnitude harder than D-T: reaching 10^21 today is a D-T achievement, while D-D would need roughly 10^22.
— Yang ZhaoThe first device with Q above 10 is fusion's ChatGPT moment
Yang Zhao says that when the first demonstration plant reaches four or five mao per kilowatt-hour it is still more expensive than coal power, but it is already near the inflection point, because everyone can predict with the naked eye that costs at scale will be far lower and demand will surge. He thinks the earlier inflection point is the world's first device with Q greater than or equal to 10 whose cost is comparable to coal power, similar to ChatGPT's impact on AI: before ChatGPT people treated AI as a tool for image recognition and speech recognition; after it appeared, although artificial general intelligence had not been achieved, it created a consensus among many people that AGI would arrive. In fusion, he thinks that consensus will roughly form when Spark or Honghuang 170 is completed.
— Yang ZhaoIn their own words · checked verbatim
At the same performance, say at the same energy gain, I can shrink the device volume by two orders of magnitude — which really means your build cost shrinks by about two orders of magnitude.
在同样的性能的情况下 比如说同样的营量增益的条件下 我可以将装置的体积缩小两个数量级 其实也就意味着 其实你的建造的成本大约缩小了两个数量级
Yang Zhao14:15
Every single link keeps going wrong, and the closer you get to the food state, the bigger your problems, the more problems you have.
就是你每一个环节 都在不断的出问题 而且你越接近食物状态 你的问题越大 问题越多
Yang Zhao1:07:33
Say I want to build an aircraft now. I have flight test data from 0 to 10 metres, and I use that data to extrapolate to 10,000 metres altitude and design an aircraft.
就比如说我现在要造一架飞机 我现在有0到10米的飞行的实验数据 然后我用这个实验数据 我外推到万米高空 我去设计一架飞机
Yang Zhao1:42:17
It's an engineering problem where a solution definitely exists — yes, that's one of our basic judgments.
它是一个解一定存在的工程问题 对 这是我们的一个基本判断
Yang Zhao2:28:04
Figures
| Honghuang 70 triple product | about 10^17 | 1:14:57 |
| Honghuang 70 peak magnetic field | 3.1 tesla | 1:34:06 |
| Honghuang 170 peak magnetic field | 23 tesla | 1:34:06 |
| Jingtian magnet magnetic field | 21.7 tesla | 1:21:10 |
| Honghuang 70 actual cost | about 120 million RMB | 1:33:06 |
| Honghuang 170 cost | about 3 billion RMB | 1:33:06 |
| Honghuang 380 target sale price | 40,000 to 50,000 RMB per kilowatt | 1:39:11 |
| Total fusion sector funding | close to 6 billion USD | 1:52:32 |
| Number of domestic fusion startups | fewer than 10 | 1:52:32 |
Glossary
- Q value / energy gain
- Output power divided by input power; 1 is break-even, and power plant designs generally need 5 to 30.
- Triple product / density × temperature × confinement time
- The physics parameter that determines energy gain; at the 10^21 level, Q is about 1.
- Tokamak
- A magnetic confinement route that confines plasma with a doughnut-shaped field configuration; more than 100 have been built worldwide.
- HTS / high-temperature superconductor
- Second-generation superconducting material whose critical field is more than twice that of low-temperature superconductors; only industrially mass-produced after 2015.
- FRC / field-reversed configuration
- A magnetic confinement route with a linear field configuration, used by Helion; the highest publicly reported triple product is about 10^17.
How to listen
Investors watching hard tech and energy, fusion and superconductor engineers, and founders who want to see how a long-cycle, capital-heavy startup decomposes risk.
The first 20 minutes of basic fission/fusion popular science, and the rapid-fire Q&A after 2:30.