The Solar Cell Factory Step SpaceX Could Not Skip

A German company with about 312 employees just became one of the most important suppliers in the AI race. It makes machines that coat silicon wafers. That sounds like boring industrial plumbing. But this plumbing decides whether SpaceX can bolt a power plant onto every AI satellite it wants to build.

Singulus published its SpaceX contract release on November 10, 2026, and SpaceX announced it on X. I'll skip the recap. The question here is why a solar cell equipment order is the factory step a satellite compute network can't do without, and why I think it puts SpaceX on a path to becoming the biggest company in the history of the world.

The smallest company in the deal

Singulus Technologies was founded in 1995 and builds precision vacuum equipment for semiconductors and solar cells. In all of 2025, it brought in about 48.3 million euros in total revenue.

The mandatory European disclosure says the first turnkey production line under the SpaceX agreement has an order volume of over €60 million. So one machine line is bigger than the supplier's entire revenue from the year before. The next day, Handelsblatt reported the stock at 25.90 euros, up 104% from the prior close of 12.70 euros.

Singulus plans to deliver that first heterojunction line in the first half of 2027. More lines are planned under a strategic partnership, and Singulus says production will scale to support more than 100 gigawatts of production equipment as SpaceX ramps satellite manufacturing in Bastrop, Texas.

A gigawatt is roughly the output of one large nuclear reactor. The release gives that 100 gigawatt figure no time unit, but on my reading it means tools sized to make enough solar cells each year to match the peak power of more than 100 nuclear plants. I read that as a clear signal. SpaceX will build whatever it needs so it never ends up supply constrained.

The cells are for a planned constellation that puts AI supercomputers in orbit. It's separate from Starlink, which beams internet down to dishes on the ground.

Waiting on the grid

Why put computers in space when Earth has plenty of land? Because the bottleneck on the ground is power.

Every time someone asks an AI model to write code or run a business workflow, a warehouse of chips crunches the math and returns small pieces of data called tokens. The base of that stack is the chip plus the power source.

A single frontier data center campus on Earth is designed to draw one or two gigawatts around the clock. That's one complex eating the output of two nuclear plants. Utilities across the United States make new campuses wait years to hook up to the high voltage grid. And local residents and county commissions fight new data centers over power bills, transformers, permits and water for cooling.

SpaceX hits the same wall. In November 2026, Reuters covered a public hearing notice showing SpaceX plans to spend roughly $24 billion on a gas-fired power plant and computing facility at Starbase in South Texas. That's on top of ground compute that reached about 1 gigawatt of nameplate capacity in March 2026.

Sunlight with no night shift

SpaceX's program web page defines the mission as capturing solar energy in space to power low-cost, high-performance AI compute for Earth. The satellites are planned for low Earth orbit, a few hundred miles up. Specifically, a sun-synchronous orbit, a polar path that rides the line between day and night in constant sunshine.

On the ground, a solar panel sits dark every night, loses output under clouds and loses about 30% of the sun's energy to the atmosphere. In Elon Musk's January 2026 World Economic Forum framing, as reported by PV magazine, a solar panel in space generates about five times more energy over time than the same panel on the ground.

Cooling flips too. Instead of burning electricity on chillers, the satellites use large radiator panels aimed edge-on to the sun that radiate infrared heat into deep space. SpaceX's page says that cuts cooling power overhead by an order of magnitude. But every satellite becomes a flying power plant, and a power plant needs a lot of solar cells.

A satellite that's mostly wing

For decades, a communications satellite needed a few kilowatts, about what a couple of hair dryers draw. According to Spaceflight Now, an original single-wing Starlink satellite measured about 36 feet end to end.

Nothing from the AI program has launched yet. SpaceX's Federal Communications Commission application for up to 1 million orbital data center satellites was accepted for filing in February 2026 and is still pending. On SpaceX's page, each AI1 would unfold to 98 feet tall and 246 feet across, wider than a Boeing 747's wingspan.

In June 2026, in a pre-IPO video covered by Reuters, Musk described a draft version of AI1 at about 150 kilowatts of peak power and 120 kilowatts sustained. SpaceX's program page now lists up to 250 kilowatts peak and 175 kilowatts average, with 75 kilowatts of compute per ton.

A clip in the video, from SpaceX's June presentation, explains why this is buildable. Musk says an AI satellite is much simpler than a Starlink satellite, which carries gigantic antennas and a lot of laser links. An AI satellite is essentially a lot of solar cells and a radiator, plus some laser links. Take away the hard antennas and what's left is mostly solar array and radiator.

Do the cell math

SpaceX's June 2026 IPO prospectus says early AI satellites targeting 100 kilowatts per ton will need roughly five times the solar array output of V3 satellite designs.

In the June video, Musk assumed solar arrays making about 250 watts per square meter, or about 23 watts per square foot. At that rate, one 250 kilowatt AI1 needs over 10,700 square feet of solar array. That's more than two basketball courts of cells on one spacecraft.

The prospectus says SpaceX aims to launch 100 gigawatts of AI compute a year on solar-powered satellites. At 250 kilowatts each, 100 gigawatts a year means building 400,000 satellites a year. That's roughly 4.3 billion square feet of solar panels annually. The prospectus says Starship V3 is designed to carry 100 metric tons to orbit fully reusable, and that 100 gigawatts a year means about 1 million metric tons to orbit annually. At 100 tons a flight, that works out to 10,000 flights a year.

Everyone watches the rockets. The cells are the quieter bottleneck.

Why boutique cells lose

Classic satellites in high-radiation orbits use multi-junction space cells. Singulus notes on its website that those cells reach over 35% efficiency. The catch is cost and scale. They're grown in tiny batches and cost tens to hundreds of dollars per watt. Covering 400,000 jumbo-jet-sized satellites a year with them would bankrupt any buyer on Earth.

Low Earth orbit changes the math. The radiation is milder, each satellite is built for a refresh cycle of a few years, and Starship cuts the cost of lifting mass. So silicon cells, which cost pennies or dimes per watt on Earth, beat exotic space cells by a mile. The condition is that the silicon has to be light enough and efficient enough for orbit. That condition is what SpaceX is buying from Singulus.

Plugging the manholes

Think of a paper-thin slice of crystalline silicon as a packed stadium. Sunlight knocks electrons loose from their seats, and a one-way gate called a junction pushes them into a metal wire. That's current.

The trouble is the cut surface. When a factory saws silicon into wafers, both faces are left with millions of broken atomic bonds. They act like open manholes on the stadium floor, trapping electrons and turning their energy into wasted heat.

Hetero means different. A heterojunction cell keeps crystalline silicon in the center and coats both faces with an ultra-thin film of amorphous silicon, where the atoms sit smoothly like glass. That film plugs the manholes. Then a vacuum machine coats both sides with a transparent conductive layer that funnels electricity into fine metal fingers.

Why does that matter in orbit? Older silicon cell lines bake wafers in furnaces at roughly 1,300 to 1,500 degrees Fahrenheit, which can warp a razor-thin slice like a potato chip. Singulus machines deposit the heterojunction layers in a vacuum chamber at around 400 degrees. So SpaceX can slice the silicon much thinner without warping it, and every wing weighs less.

They also hold up better in space sunlight. Based on November 2026 space cell figures published by Swift Solar, heterojunction cells convert about 25% of sunlight, compared with about 23% for older silicon cells. They lose less power in direct sun and deliver roughly 10% more power at the start of a mission.

What Singulus actually sells

Singulus sells the whole factory line as a turnkey package. In the November 2026 release, Lars Lieberwirth, a member of the Singulus executive board, said, "Our unique platform approach, which combines wet chemistry, vacuum coating and thermal processing, is a strong differentiator in the market." In a December 2024 presentation covered by TaiyangNews, Singulus said its wet cleaning tools process over 14,000 half-cut wafers an hour with breakage under a tenth of 1%. Its sputtering machine coats up to 1 gigawatt of cells a year in a single tool.

Cells are not modules

Why not buy cells in America? Because there aren't many. China controls roughly 80% or more of the world's solar wafer and cell manufacturing. Assembling a module means framing finished cells under glass. Making the cell is the hard part. Solar Energy Industries Association figures reported in early 2026 put U.S. module assembly capacity at about 65 gigawatts and domestic cell capacity at only about 3.2 gigawatts.

Even the biggest vertically integrated solar plant in America, a $2.5 billion facility in Georgia that began cell production in June 2026 according to PVTech, has a nameplate capacity of 3.3 gigawatts of cells a year. SpaceX's target is roughly 30 times that. And the main Western heterojunction panel maker that built in America went under. Reuters reported in May 2025 that Swiss manufacturer Meyer Burger shut its module plant in Goodyear, Arizona and laid off all 282 workers there. Its US units filed for Chapter 11 in June 2025.

The loan before the order

Because nearly every new cell factory in the past decade was built in China, one of Singulus' most important customers for many years, and its largest shareholder, was Triumph Science and Technology Group, a subsidiary of state-owned China National Building Material, holding about 16.75% of the stock. Going into 2026, it also owed a 9 million euro Chinese bank loan and project advances to Triumph.

On March 23, 2026, Singulus announced a solar partnership with an unnamed leading global technology company. Filings showed that partner gave a 29 million euro five-year loan and bought a convertible bond that could become up to 20% of Singulus' stock. The money paid off the Chinese bank loan. On October 21, 2026, the same partner provided a 22.75 million euro subordinated secured loan to repay what was still owed to Triumph. That same day, Triumph sold its entire stake to a German investor.

Then the November 10 release said SpaceX provided significant funding to Singulus in March 2026 and again in October 2026. SpaceX cleared a supplier's Chinese state-backed obligations before placing the big order. That's what a durable option looks like.

The factory down the street

Bastrop County sits about 30 miles east of Austin. Citing the Austin Business Journal, The Real Deal reported in March 2026 that SpaceX's dish factory there spans about 1.1 million square feet and could grow to about 1.6 million. In May 2026, it reported that Bastrop County officials say SpaceX filed permits for a 1.1 million square foot solar cell facility. SpaceX's page says its Bastrop satellite factory is planned to enable production of thousands of AI satellites starting as soon as late 2027. The prospectus says as early as 2028.

In the Singulus release, Mike Nicolls, SpaceX's senior vice president of Starlink, said partnering with Singulus gives SpaceX "a high-performance manufacturing path for solar cells for our Starmind satellites at our production facility in Texas."

Owning the factory clock

Why does it matter that cells get made next to satellite assembly? Look at the old flywheel. When Falcon 9 needed more flights to drive down cost per launch, Starlink was the captive customer ordering launches every week. SpaceX's June 2026 IPO prospectus says the move from Starlink V1 to V2 Mini cut satellite manufacturing cost per gigabit per second of downlink capacity to about a third of the V1 level.

Now the same loop runs at 10 times the scale. The cell line feeds the satellite line, and the satellites keep Starship full. With the Singulus lines next door, SpaceX controls the factory clock. I count six layers that matter for AI infrastructure once data centers go to space: launches, chips, gigawatts, satellites, models and developer surface. SpaceX has all six, and this contract fills in the gigawatts layer.

The skeptic and the token

A fair skeptic says this is one line arriving in the first half of 2027. Getting to 100 gigawatts takes dozens more lines, a massive wafer supply chain and thousands of Starship flights, for a constellation that hasn't launched. I still wouldn't bet against this team on a manufacturing ramp. That's been a losing bet for 20 years.

Once frontier models are smart enough for almost any digital job, the company that makes and delivers tokens at the lowest cost wins most of the demand. A competitor on Earth pays markups on chips, grid power, cooling water and land. SpaceX's plan is to build the cells in Bastrop, build the satellite down the hall, launch it on Starship, power it with sunlight and beam tokens down over Starlink lasers.

So here's my bold projection. SpaceX starts capturing the majority of profits from generating AI tokens by the end of 2028. If that happens, the foundation will be a solar cell machine contract in a small Texas town, signed with a company of about 312 employees.