China controls about 90% of the world's solar supply chain — panels, cells, wafers and the polysilicon upstream of all three — and Joel Jean said a domestic American industry could be built for less than a single hyperscaler will spend on data centers next year.
The comparison is not rhetorical. Swift Solar, the company he runs, is trying to raise factory money in a market where the same investors are writing far larger checks for computing capacity that the solar would end up powering.
"We could build an entire domestic solar industry on far less than that amount."
Jean did his PhD on solar cells at MIT, co-founded Swift Solar with the researchers who built the first perovskite-on-silicon tandem cells, and has just bought the assets and the leadership team of Meyer Burger, the European silicon manufacturer that failed trying to build American plants.
The full interview is covered here so you can skip it. 90 minutes of audio, 20 minutes of reading.
Here are the 11 insights that matter.
👤 Guest: Joel Jean, CEO and co-founder of Swift Solar, who did his PhD in solar technology at MIT and whose company licensed the academic patents that he says every world-record perovskite cell has used
🎙️ Host: Ashlee Vance, founder of Core Memory, who spent 14 years at Bloomberg Businessweek and wrote the 2015 biography of Elon Musk
📰 Published: 16 September 2026 on YouTube (Core Memory) and the show's own feed
🔴 YouTube | 🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 1 hr 30 min | ✅ Time saved: 70 min
Key Takeaways
China holds about 90% of solar manufacturing, and the further upstream you go the tighter the hold
The scaling was subsidized at the province level with free land, cheap power and cheap capital, not planned centrally
Silicon has run into a physical ceiling of about 30% efficiency, and commercial cells are already at 24% to 26%
Stacking a perovskite cell on a silicon one raises the theoretical ceiling to 45%; the lab record is 34.8%
The unsolved problem is not efficiency but whether the new material survives 30 years outdoors
Jean says Swift has improved stability a thousandfold and has conviction, but not yet the data to back a 30-year warranty
Almost every solar factory outside China runs on Chinese equipment, and China is reportedly restricting the leading-edge machines
The trigger, he said, was a reported $2.9 billion equipment order by Tesla
US-made panels sell at a premium because buyers want a supply chain that will not be seized at the border
Jeff Dean is an investor in Swift Solar, on the view that AI will be powered mostly by solar
Solar costs fall about 20% with every doubling of deployment, which Jean says matches the pace AI needs
The company is selling conventional silicon panels first to fund the new technology
Jean calls it a bankable path: proven product now, perovskite layered on top in three to five years
1. Bell Labs to Beijing
Vance opened by asking Jean to tell the story from the beginning. The short version is that the United States invented the technology, Japan and Germany paid to create the market, and China took it from there.
Jean dated the industry precisely: "It started in 1954 with the Bell Labs, right? The very first silicon solar cell was invented at Bell Labs." He put the first cell at about 6% efficiency, and its first use in satellites — a full circle, he said, now that solar is being discussed for space again.
Japan subsidized rooftop solar and Germany deployed gigawatts of it through the 2000s and 2010s with feed-in tariffs, at a time when the economics did not work. Vance noted that Germany is not a sunny country; Jean put its solar resource near Boston's, and said the push was about green technology and about shutting down nuclear.
Australia supplied the research lineage. Jean named Martin Green at the University of New South Wales as the godfather of silicon solar, whose PhD students and postdocs went on to found Suntech and the other Chinese giants.
China's share is the number that matters: he put it at 90% of silicon panel, cell, wafer and polysilicon manufacturing. "And you know, they built it out in a way that maybe no one else in the world could have."
His account of how is bottom-up rather than master-planned. Successive five-year plans told provinces and local governments to create jobs in solar, and those governments supplied free land, subsidized electricity, cheap labor, tax benefits and cheap capital — a combination he said is not available anywhere else in the world.
The competition inside China was real: "And it's like the early days of Silicon Valley." He described solar engineers at rival Chinese firms sharing trade secrets over WeChat groups, which he said is not free-market behavior as it is usually understood but did produce a fast-moving industry.
Vance asked whether the reports that China now installs more solar than the rest of the world combined are accurate. Jean said they are.
2. Why the US Let Go
Vance asked whether American oil politics explain the handoff. Jean's answer was that the oil companies were the early backers, and that federal money peaked half a century ago.
"The oil companies were kind of the innovators in solar back in the 80s and 90s. BP and Exxon, all these players had huge solar efforts. They kind of pushed the technology in many ways. They sort of gave up on it."
The high-water mark for federal solar research was the response to the 1973 and 1979 oil shocks — about $3 billion equivalent, which he said was "the highest in terms of like total dollars going into solar research."
Vance connected it to the rest of that era's energy policy, recalling the Nixon-era plan to build a thousand nuclear power plants by the year 2000 and make the country energy independent.
Jean said government money is what kept solar alive through the 1970s, 1980s and 1990s, when it was not a real industry, because it was treated as strategic — and that the same treatment is coming back.
The contrast he drew is consistency. American support fluctuates with administrations. China's five-year plans say the same thing every time.
He also noted a live comparison: the International Energy Agency has called the present situation the biggest energy crisis in history, worse than the 1970s shocks.
3. Efficiency Isn't the Point
Vance asked how solar keeps getting better, and Jean redirected the question to cost.
The distinction he drew first is between the cell and the panel. The cell is the power-generating engine — a wafer of single-crystal silicon he described as a giant chip, roughly two to three times thinner than a semiconductor wafer, produced by the million a day in a gigawatt-scale plant.
Panel efficiency is the number the industry quotes, and it has been rising about half a percentage point a year for decades through better glass texturing, anti-reflection coatings and more transparent plastic layers — and, more fundamentally, better passivation of the silicon, the process of plugging up defects where energy leaks away as heat. PERC, TOPCon and heterojunction are each a better way of doing that.
That improvement is now flattening, and the reason is physics. Any single material has a threshold wavelength beyond which it stops converting light. Silicon's limit is 29.5%; about 30% is the ceiling for any single material.
The record cell is in the high 27s and commercial cells run 24% to 26%.
"I guess what's important to keep in mind is that it's not efficiency that we really care about." What matters is the levelized cost of electricity, which includes the labor, racking, wiring and transformers as well as the panel.
On that measure the argument is settled: "In many places it is the cheapest. It's actually dependent on how much sun you get." Germany's higher power prices make solar competitive there, he said, but it will not be as cheap as Arizona.
4. What a Perovskite Is
A perovskite is a crystal structure, found in nature and reproducible synthetically, that turns out to make an unusually good light absorber. Vance asked for the plain-language version.
The property that makes it work is tolerance for imperfection. A semiconductor wafer normally has to be pure to five or six nines, with every millionth atom in place. Perovskites keep their performance with missing atoms and ions moving around.
The temperature comparison is the one that carries the point. The material forms at roughly 100 degrees Celsius and performs like a crystal of silicon grown at 1,400.
The thin-film comparison explains why this one is different. After the 2008 polysilicon shortage, when prices hit about $100 per kilogram, the industry chased thin films that used a hundredth of the material — but none of them beat silicon on performance, so they could only compete on cost. A perovskite is thinner still, under a micron, and can beat silicon.
The turning point was a 2012 paper from Oxford and a Japanese team showing the material could both absorb light and carry charge, like a conventional semiconductor.
Jean's measure of how good that first result was is his own career: "And they showed a 10% efficient solar cell like from day one. And just to put that in perspective like I spent my PhD making solar cells out of various new materials like quantum dots and organic materials and like I had never made a solar cell in 5 years that was more than 5% efficient."
5. The 30-Year Problem
Vance put the skeptics' case to him: the material has been promising for 15 years and is still not in mass production, and solar people he called for his story said it never will be.
Jean accepted the criticism and blamed the hype cycle rather than the science. University press offices promote efficiency records that the industry already knows are a decade from production, so the public hears that the technology has arrived, then hears nothing for ten years. "The reality is it's like the technology has been developing steadily in the background."
The binding constraint is a 20- to 30-year outdoor life, which is what the economics of a solar panel are built on.
He said the material is genuinely more fragile than silicon — it forms at lower temperatures and can degrade at lower temperatures, and it is an ionic rather than covalent semiconductor, so ions move under electric fields and heat.
His counter is that every commercialized solar technology looked like this early on. "And that's true of every solar technology that's ever been commercialized, right?" He cited the JPL-supported program that took silicon modules from roughly a one-year life to more than ten years between 1975 and 1985.
He walked back his own strongest claim when Vance pushed on it: "maybe I said that too strongly." Nobody has proven a perovskite cell matches silicon on durability, he said, but the difference is a spectrum rather than a wall.
He put the field at roughly five perovskite startups in the United States, a couple in Europe, a few elsewhere, and a large number in China — all competing on the same stability threshold.
On the Chinese firms already building factories, his read was that they know it too: "I think they would tell you the same thing." They are deploying, watching panels degrade and iterating, which he said can be done at gigawatt scale if you are willing to set money on fire, and which the Chinese government is willing to fund strategically.
His thesis is that iteration does not substitute for invention: "But I guess my thesis is that, scaling something up and improving it with like continuous improvement in manufacturing is not the same as making the sort of binary step changes that sometimes are needed to solve these technical challenges."
Asked whether the problems are solved, he separated two standards: "I think there's like two levels of solving things." Enough understanding to be convinced it is solvable, which he has; and enough field data to underwrite a 30-year warranty, which he does not. "I would say like we have some of the best data in the world on this, right? But like it's not the definitive proof 30-year lifetime yet."
6. Stacking to 45%
The reason to care about perovskites is not that they replace silicon but that they sit on top of it.
A tandem cell stacks two absorbers: the perovskite takes the high-energy visible light, the silicon takes the infrared, and both convert their share more efficiently than one material covering the whole spectrum.
That raises the theoretical ceiling from about 30% to 45%. Jean called it the step change. The record today is about 34.8%, in a research lab.
Swift set the first world record with the tandem and then deliberately stopped working on efficiency, on the reasoning that efficiency is the solvable part — many labs can make an efficient perovskite, and almost nobody can make a stable one.
Swift's own first tandem demonstration reached about 24%, already competitive, but with no stability behind it.
Jean said his founding team made the first perovskite tandem cells on silicon, and that every company in the world that has set a perovskite world record has used Swift's licensed intellectual property to do it. The work was done in academic labs before the company existed and was licensed in afterward.
The second body of work is manufacturing. His co-founders built the deposition tools themselves in the early days, and Jean said controlling the tools and the software that runs them is what let the company change recipes and designs quickly. He was explicit that Swift does not want to be a tool manufacturer; it did it because it had to.
7. Buying Meyer Burger
Swift recently bought the core assets of Meyer Burger, the European silicon manufacturer that had a factory in Arizona and was building one in Colorado before it ran into trouble.
The point of the deal is the silicon technology, not the distressed price. Meyer Burger was the Western leader in heterojunction, one of the two leading silicon cell technologies — and, Jean said, the best bottom cell to build a tandem on.
Swift took the assets, the equipment, the intellectual property and the team, including the former chief executive and the technical leadership.
The strategy this produces is sequential rather than simultaneous: "So what's cool about this is like it gives us a real like what we consider a bankable path to scale." Build proven silicon at gigawatt scale in the United States now, sell to customers who want domestic product, and mature the perovskite layer in parallel until it can go on top.
Jean's reason for doing it now is a set of conditions arriving together: a decade of stability work coming good — "It's like we feel like after this decade of development, like we have shown, a thousandx improvement in stability over the last few years" — plus demand from AI data centers, reshoring policy, domestic-content rules and tariffs. "Tariffs as well, all kind of like moving in the same direction of like the US needs a crap ton of solar."
Asked whether he was being modest or was genuinely confident, he distinguished the two: "I've been doing this for a long time, right? And like I'm not the most let's say boastful person around." He added that these projects always take more time and money than anyone projects, which is why startups fail, and that he does not want to mislead anyone.
"But that said, like I am genuinely like the most optimistic by far as I've ever been about this company and the trajectory we're on and about the technology and the state that it's in."
On timing, he was specific: "It's going to take 3 to 5 years to get to that kind of gigawatt scale where it's actually competitive." Swift already has product in the field — a pilot on a utility-scale solar farm in Texas with a major developer, and a project with the Department of Defense announced last year.
8. Competing at 30c a Watt
Vance asked the obvious question: even with proven silicon, can an American factory match China on price?
Jean's answer was no, not against most of the world. China has done an extraordinary job scaling silicon and driving cost down, and he did not dispute it.
The US market is a different market. Imported panels from China or Southeast Asia land in the United States at around 30 cents a watt, well above the price inside China, and American-made product sells at a premium above that.
The premium is not only subsidy. Buyers want a supply chain that will not be stopped at the border over forced-labor findings or tariffs.
He listed the policy support as bipartisan and durable: 45X manufacturing tax credits, domestic-content requirements, and tariffs on below-cost imports.
Vance raised the contradiction facing any clean-energy startup — an administration that wants manufacturing jobs and is hostile to green technology. Jean said he does not feel it directly: "I think the way I think about it is like solar is actually the one energy source that everyone can agree on." He put public support at about 70% of American adults, higher still for domestically made panels.
His own position on where this has to land: "I think the US needs to make some of its own solar." He was careful not to claim Swift is the only company that can do it — "I think there are multiple players in the world who can develop this and commercialize it and many people are making great progress" — and added that "That doesn't mean that China can't also succeed in parallel."
His objection to the free-trade answer is political rather than economic. Letting China make cheap panels while America deploys them is, he said, the same class of idea as a carbon tax: economically efficient and not politically survivable, because the voting public will not choose a transition whose jobs and economic benefits all go somewhere else.
9. Solar Is AI's Power Source
Vance described visiting JB Straubel's Redwood Materials in Nevada, where used electric-vehicle batteries with 50% to 70% of their life left are lashed together, tied to a solar field and used to run an off-grid data center in a container. He asked whether solar can do that at real scale.
Jean said the hyperscalers have already answered it. Google, Amazon Web Services and Microsoft are all deploying large amounts of solar as their main source of electrons, alongside firm generation and batteries.
Most of it is bought through power purchase agreements rather than built next door, which is why a visitor to a data center does not see it. He noted Google's acquisition of the renewable developer Intersect.
The case for putting it next door is the interconnection queue. Modular solar beside modular batteries beside modular compute skips a wait for grid connection that can run years, and Jean said the opportunity cost of that wait is millions of dollars a day.
On transmission, he took both sides of the familiar argument. Elon Musk's point that a small patch of land could power the country and Bill Gates's point that transmission is the constraint are both right; the United States has a great deal of existing wire, old as it is.
Vance asked why American deserts are not simply covered in panels. Jean said the country deploys 40 gigawatts of solar a year and that the easiest sites — easy to permit, easy to connect — are taken, but that hundreds of gigawatts to terawatts of additional capacity remain buildable. The whole country could be powered, he said, on the land currently used for coal mining.
The modularity is what he thinks is underrated. Solar scales down without losing efficiency: ten panels on a roof work as well per panel as a ten-square-mile array, which is not true of a nuclear plant.
The disclosure at the end of the conversation is the news in it: "I hope he's not mad at me if I say he's now an investor at Swift, but he is backing Swift." The investor is Jeff Dean, whom Vance introduced for listeners as Google's longtime resident genius and its AI chief.
Jean's reason for the shared view: "And I think part of the reason is that like we share this belief that the future of AI is going to be heavily powered by solar. It's the fastest thing you can deploy. It can match the speed of AI. It's on this learning curve that's getting cheaper and cheaper, right? 20% decline in cost, every doubling of deployment."
10. China's Equipment Lever
The chokepoint Jean is most worried about is not panels. It is the machines that make them.
Almost every solar factory outside China, including the ones being built to reshore production, runs on Chinese manufacturing equipment. Swift's does not, he said; the generic factory does.
"But yeah, there's this new announcement like this week this news that The Chinese government was like influencing Chinese manufacturers to say like hey like maybe you shouldn't sell that leading edge manufacturing equipment for solar cells to the US." He described the reporting as rumor-level and could not say whether it covered the United States specifically or the West broadly.
The trigger named in the article, he said, was Tesla — a reported $2.9 billion purchase of solar equipment to build domestic cell and panel manufacturing, which would use Chinese machines to create capacity that competes with China.
On whether Tesla is still serious, he would not go further than the public record: "I can't speak to like where they stand today, but I think they're definitely exploring that for sure. And that's been widely reported." He added that "they're definitely making some serious bets."
Vance's framing of the underlying problem, from the close of the interview: "It's just so what you just said, it's like if you pair solar up with batteries, well, it happens to like China's dominant and completely dominant in both of those." Jean's answer was that the country had better get building.
11. Why They Don't Patent
Vance asked why a company in this position publishes patents at all, given how the last transfer went.
His own case study: "I've done stories where the US was like leading in windmills, too. But then, China got the source code to the controller in this windmill. And the company I was writing about in Massachusetts was like, went from the world leader to bankrupt in like 18 months."
Jean said most of Swift's work is not patented for exactly that reason, and is held as trade secrets instead. The manufacturing recipes in particular are not filed and are not obvious from opening up a finished panel, though he allowed that some of the cell stack would be visible in a cross-sectional electron-microscope image.
What does get patented is patented defensively, to preserve freedom to operate and avoid being attacked by someone who filed on the same work.
He said intellectual-property enforcement in solar is rising, including inside China, with Chinese companies suing each other in the United States and a new investigation into TOPCon technology.
Asked whether other materials could leapfrog perovskites, he said no. Cadmium telluride, CIGS, quantum dots and the rest have all failed to outperform silicon, and silicon has had seven decades of refinement behind it. Perovskites, he said, are the only material to come along since then that can match silicon and do it economically.
On whether AI-driven materials discovery changes that, his answer was that the search has already been run. If you had asked twenty years ago for the perfect solar material, what came out of the hat would look very much like a perovskite; the likeliest outcome is a refined version of one, not something new.
"Yeah, I think the whole industry will shift." To him it is a question of when rather than if.
Bonus Insights
Jean's argument for why private capital should care is that a technology transition is when incumbents are vulnerable. Momentum is building behind perovskite tandems, he said, because the alternative use of the money is the next silicon factory, and this is a technology that could upend the industry instead. He was careful to add that China's silicon advantages are nowhere near gone.
His frustration with American capital allocation: "It's kind of weird because we're actually fantastic at building data centers. We're like amazing at software. We're fantastic. We have more data centers than the rest of the world combined." Solar, batteries and robotics actuators, he said, are the pieces nobody takes as seriously.
On what the government should do: "Yes. The government needs to do stuff like we can take a page out of the Chinese playbook and how they dominated solar and storage and all that, right?" He named cheap capital, consistent research funding, loan guarantees and tariffs.
Solar roofs are a niche, in his view: "Yeah. I mean I think it's happening in small ways here and there. It's like it's more of a niche product, right?" Every roof is different, so the product has to be customized, and it is less efficient than putting panels in a desert.
On Germany, Jean declined to pick a side between the solar and nuclear readings of its energy policy. Both are true, he said: it should have deployed the solar and it should not have shut the nuclear plants.
Vance raised his own scar tissue on materials stories — carbon nanotubes, which he said he wrote about repeatedly and now refuses to write about again — as the reason for his skepticism.
Vance's compliment at the close doubles as a description of the guest: "I always appreciate when somebody comes from academia who can also explain what's going on in like human terms that I can understand."
The two men met a couple of months before the recording, when Vance reported a story on Swift Solar. Jean noted that Core Memory's office sits near Tesla's original headquarters on San Carlos, and that a Tesla co-founder once visited and thought it had been their building.
Jean's bottom line is that the window for the United States is a technology transition rather than a cost war: silicon is at its physical limit and China owns it, perovskite tandems raise the ceiling, and whoever proves a 30-year lifetime first gets to rebuild the supply chain somewhere it does not currently exist.
Products, Companies & Tools Mentioned
Swift Solar (Jean's company, building perovskite-on-silicon tandem cells and now the acquirer of Meyer Burger's core assets)
Meyer Burger (The European silicon manufacturer whose heterojunction technology, equipment, intellectual property and leadership team Swift bought)
Tesla (Reported to have ordered $2.9 billion of solar manufacturing equipment, which Jean said prompted the Chinese export restrictions)
Redwood Materials (JB Straubel's battery recycling company, where Vance saw used electric-vehicle batteries paired with a solar field to run an off-grid data center)
Google, Amazon Web Services and Microsoft (The hyperscalers Jean said are buying large amounts of solar through power purchase agreements as their main source of electrons)
Intersect Power (The renewable developer Google acquired, which Jean cited as evidence the hyperscalers are building solar rather than only contracting for it)
BP and ExxonMobil (The oil companies that ran large solar programs in the 1980s and 1990s and then abandoned them)
Suntech (Founded by researchers out of the University of New South Wales, and Jean's example of how Australian research became Chinese manufacturing)
Oxford PV (The European perovskite company Jean named as the early leader and the first out of the labs)
Books & Resources Mentioned
Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future – Ashlee Vance (The host's 2015 biography, which is why he had visited Tesla's original headquarters)
Watch the full episode:
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: