Helion just won $175 million to build semiconductors, not reactors. Its CEO says halving the electricity lost in power electronics would double the output of a fusion plant. The star may be the easy part; the toll booth is the hard one.
When we picture a fusion reactor we picture the star: a plasma hotter than the core of the Sun, held in place by magnets. That is the part with the glamour and the physics prizes. This week's cover story is about the part nobody photographs, the electronics that move energy in and out of that plasma, and the claim that it may decide who gets to sell fusion power first.
Here is the lens for the issue: find the component that every path must pass through, and ask what it charges. Helion's chief executive put it plainly this week. Today's power electronics lose roughly 5 to 10 percent of the electricity flowing through them. Cut the loss in half, he says, and you have twice the output. A tax that small should not be able to do that. Page 3 shows the arithmetic of how it can.
You will meet the same shape in the Signals: a cancer vaccine whose value lies in teaching one existing drug to work harder, a country counting robots by the tens of thousands as if they were cars, and a ten-dollar chip that makes a laboratory instrument cheap enough to carry. Read the issue asking one question: where is the toll, and who just found a way to lower it?
— The Editors
Helion, the second-best-funded fusion startup, is spending a federal award on power semiconductors. The reason is a piece of arithmetic that makes a small leak enormous.
Imagine you run a water wheel that is supposed to pump more water uphill than it uses. Every bucket has a tiny hole. If the wheel is wildly efficient, you never notice. If the wheel barely beats breakeven, the holes are the whole story. That is the position of a pulsed fusion machine, and it is why Helion announced this week that it will spend a $175 million Department of Commerce award, made under the CHIPS Act, on semiconductors. The money is for research, development and manufacturing of advanced chips and power electronics. Not for plasma.
Helion's design uses a field-reversed configuration. Two plasmas are accelerated toward each other, compressed, and heated to more than 100 million degrees Celsius. When fusion occurs the plasma pushes outward against the magnetic field, and the magnets and capacitors capture that push as electricity. Helion calls this direct energy conversion, and it skips the steam-turbine step that most power plants share.
The design is a pulse. Electricity goes in as a hard, fast kick, and some comes back. Between them sits a bank of capacitors and the switches that decide when current flows. Every joule has to pass through those switches twice, once on the way in and once on the way out.
Helion's chief executive, David Kirtley, says current power electronics lose about 5 to 10 percent of the electricity that flows through them. That sounds like a rounding error. On a pulsed machine it is a tax collected at both doors. If the net gain you hoped for is a few tens of percent, a double toll of that size can eat most of it.
His summary of the stakes: "If we can cut that loss in half, that's actually twice the power output of the fusion system." Page 4 puts toy numbers on why that is plausible. The unusual thing is where he proposes to look for the fix: not in the reactor hall, but in the chip fab.
The award is a three-year, $175 million grant from the Department of Commerce. Helion also announced a $50 million private investment from an undisclosed investor, for a total of $225 million. In exchange for the federal money the government receives a $175 million non-voting interest in the company, priced at its Series G valuation. Helion has raised more than $3.2 billion in all.
More than $100 million funds semiconductor and advanced-packaging research in Washington State, where a new production line will make a class of high-voltage, pulsed semiconductors. Kirtley says it will be the first such line in the U.S. About $80 million funds work in Vermont with GlobalFoundries on tooling, design and process development.
These figures are ours, not Helion's, and exist only to show the shape. Suppose each pulse delivers 100 units of energy into the plasma and the plasma gives back 120. With a 10 percent toll on each pass, you must draw about 111 units to deliver 100, and you recover only 108 of the 120. Net: roughly minus 3. The machine loses.
Halve the toll to 5 percent. You draw about 105 and recover 114. Net: roughly plus 9. The same plasma, the same physics, and the sign of the answer flipped. When the margin is thin, efficiency of the plumbing is a cliff, not a slope.
Helion is racing to finish its first plant, Orion, to meet a 2028 deadline to supply grid electricity to Microsoft. Reporting by GeekWire describes Orion as a 50-megawatt facility in Malaga, Washington, and says Helion's factory will supply roughly 2,500 capacitor units for it. The terms of the Microsoft contract, such as price or capacity, are not disclosed.
Kirtley said the funded research will not reach Orion in time, because federal contracts take months to negotiate. It is intended to improve later plants. So the bet is on plant number two.
Look at who is being paid here. A fusion firm is directing public money toward a factory that makes high-voltage pulsed semiconductors. Kirtley notes that such devices are also used in manufacturing, wastewater treatment, medicine and defense. Electric vehicles are the broadest current use of advanced power electronics. Helion's own focus remains grid power, but the same chip serves several customers.
That makes the investment two-sided. If Orion's successors work, the chip is a captive component that competitors will struggle to match. If fusion takes longer than hoped, the production line is still a business that sells to industries that exist today. A bet with a floor under it is rarer than it sounds in this field.
The Department of Commerce's Bill Frauenhofer framed it as helping to convert and deliver power to critical applications in defense, grid transmission and advanced energy generation. Read that as industrial policy: the part of the supply chain the U.S. wants at home is the part that is hardest to import.
1 · Squeeze. Two plasmas are accelerated toward each other, compressed and heated past 100 million °C.
2 · Push. Fusion makes the plasma expand outward against the magnetic field, like a piston pushing back.
3 · Harvest. The magnets and capacitors capture that push directly as electricity. Every step crosses a power switch.
Field-reversed configuration. A compact, doughnut-less plasma shape held together by its own magnetic field.
Direct energy conversion. Turning the plasma's motion straight into current, without making steam.
Power electronics. Switches and circuits that control large flows of electrical energy.
CHIPS Act. U.S. law that funds domestic semiconductor research and manufacturing.
The market read. The customers for a better switch are not only fusion firms. They are anyone whose product moves large amounts of power in bursts. If a pulsed-power chip really does lose half as much, its first sales could be somewhere dull and profitable long before a single fusion kilowatt-hour is billed.
The consensus read: fusion is a physics problem, so progress is measured in temperatures, confinement times and gain. The contrarian read: once the plasma works well enough, the winner is whoever wastes the least electricity moving it around. Fusion becomes a power-electronics business with a star attached.
Pulsed fusion recirculates a huge amount of electricity to produce a modest net surplus. In that regime the toll matters more than the reaction. A machine that loses 5 to 10 percent at every crossing is competing with its own plumbing. Whoever lowers that cost improves every plant they build, and their advantage sits in a factory, which is far harder to copy than a paper design.
The Move 37 corollary: in an immature industry, the durable moat is often not the headline technology but the unglamorous part that all paths share. Gold-rush logic says sell the shovels. Fusion logic may say make the switch.
It is one company's claim. The 5 to 10 percent figure and the doubling come from Helion's chief executive. The reporting we found offers no independent test of either, and "twice the power output" is conditional on a margin we cannot see.
It will not help Orion. Kirtley says the funded work will not arrive in time for the 2028 plant. The nearest deadline is untouched.
Other designs may not care. A machine that recirculates less power feels the toll less. The thesis is strongest for pulsed, direct-conversion designs.
Chips are hard too. A first-of-its-kind line for high-voltage pulsed devices has its own yield, qualification and supply risks. The terms of the Microsoft contract are undisclosed. Where we land: the physics may be necessary, but the switch is the lever.
Merck and Moderna's intismeran, given with Keytruda to high-risk Stage IIB to IV melanoma patients after surgery, cut the risk of recurrence or death by 49 percent versus Keytruda alone (hazard ratio 0.51 at five years). The trial met both its primary endpoint of recurrence-free survival and the key secondary endpoint of distant metastasis-free survival. Full data were scheduled for ESMO on October 24, and the companies plan nine further Phase 2 and 3 studies in other tumors. Enrollment and safety figures were not in the report we read.
At an October 19 Beijing press conference, Wei Wei of the Ministry of Industry and Information Technology said China expects humanoid-robot production to surpass 100,000 units in 2026. By the end of June the country had developed over 400 humanoid models and produced more than 40,000 units, with embodied-AI products being tested in more than 300 priority scenarios such as warehousing, logistics, retail and household services. The report gives no prices.
Cambridge researchers and the start-up GlitterinTech built a centimeter-scale near-infrared spectrometer on a silicon-nitride photonic platform. In reported tests it identified plastics, pharmaceuticals, coffee, flour and tea with 100 percent success, measured concentrations to about 0.01 percent, and ran from minus 20 to 80 degrees Celsius. Blood-glucose tracking was shown in a single participant and the biomedical tests used simulated conditions, so it is a long way from a medical product.
Three of this week's stories are about what happens between the idea and the outcome. A fusion plasma must be fed and drained through switches. A cancer vaccine does not replace an immunotherapy; it makes the immunotherapy work harder on a patient's own tumor. A humanoid robot is only as useful as the factories, software and prices behind the count. And a spectrometer's lab-grade precision matters only once it is cheap enough to leave the lab.
The common thread is that value migrates to whoever controls the connective tissue. The star gets the headlines; the toll booth gets the revenue. For a technology leader the practical habit is to draw the whole path, from input to customer, and mark every place where something is lost. The biggest loss on that map is often the best product.
None of it is finished. Helion's chief executive says the funded research will not reach Orion, the vaccine's full dataset was just presented, and China's robot count is a production forecast. Treat each as a direction. Then ask where the toll is, and who is working to lower it.
1. TechCrunch, "Fusion startup Helion is developing new semiconductors that could double output of its power plants," Oct. 22, 2026. https://techcrunch.com/2026/10/22/fusion-startup-helion-is-developing-new-semiconductors-that-could-double-output-of-its-power-plants/
2. GeekWire, "Report: Helion is working on a massive fusion power deal with OpenAI," 2026 (Orion plant, Microsoft, capacitors). https://www.geekwire.com/2026/report-helion-is-working-on-a-massive-fusion-power-deal-with-openai/
3. MedPath, "Moderna and Merck's intismeran cuts melanoma recurrence 49% in Phase 3 INTerpath-001." https://trial.medpath.com/news/moderna-and-merck-s-intismeran-cuts-melanoma-recurrence-49-in-phase-3-interpath-001
4. China.org.cn, "China expects humanoid robot production to exceed 100,000 units this year," Oct. 20, 2026. http://www.china.org.cn/2026-10/20/content_118741588.shtml
5. ScienceDaily, "Cambridge Scientists Create a $10 Chip That Could Put a Medical Lab on Your Wrist," Oct. 22, 2026. https://www.sciencedaily.com/releases/2026/10/261022081312.htm
6. Yao, Cheng, Penty et al., "Optical convolutional spectrometer," Nature Photonics 20(6), 664. https://doi.org/10.1038/s41566-026-01891-6
Where is the toll in your own system, and what would you build if you could halve it?