Space Power

Today, Earth installs the equivalent of all orbital solar power roughly every hour and a half. 

The combined solar capacity of every spacecraft, satellite, and space station in orbit today – roughly 100 MW across 10,000+ Starlink birds and everything else – is smaller than a single mid-size solar farm in West Texas. SpaceX alone accounts for the vast majority – around 90%, we’d say – of the orbital total. 

With the Overton Window having dramatically expanded in the last 6 months, and space-based datacenters having been normalized among the powers that be in Silicon Valley, the space industry, and everywhere in between, we think that online discussion of the challenges (thermal, Kessler, etc.) is fairly well-saturated today – including in Per Aspera’s own pages, with our second-most popular Antimemo ever published (Realities of Space-Based Compute from May 2025).

But one piece that’s been relatively overlooked in the online sparring on “space datacenters” is more thorough consideration of the sun‑harvesting stack: the cell chemistry, precursor materials, manufacturing throughput, and ultimately, the key metric – $/W ¹ – that will be required to turn this dream into a reality. 

 ¹ Plenty of companies are looking at developing power-beaming satellites with concentrated solar energy or lasers. Here we’re talking about the panels themselves, that convert the light to energy, because it is one of the more important points of consideration in building large constellations of orbital datacenters. 

We’ve been closely monitoring the situation here. We’ve seen plenty of developments in the first quarter alone:  

• This January, Elon announced at Davos that SpaceX and Tesla are each targeting 100 GW/yr of U.S. solar cell manufacturing within three years (the entire U.S. installed 43 GW last year), and SpaceX filed with the FCC for a 1M-satellite orbital datacenter constellation.
• Around the same time, Chinese state media and trade press reported that a SpaceX/Tesla delegation was touring China’s solar belt, visiting with JinkoSolar, GCL Group, TCL Zhonghuan, and others, specifically to inspect heterojunction and perovskite production lines.
• Meanwhile, back in Bastrop, TX, SpaceX has 16 open roles tied to a “Solar Cell Factory” — job postings point to an internal SpaceX division that is racing to stand up its own space-solar manufacturing line rather than staying tethered to legacy III-V suppliers.

This begs the question: What do SpaceX and Elon see that others haven’t yet? 

A few off-the-cuff observations:

• This is a disruptable oligopoly: Space-grade solar is a relatively small (~$1.5B/yr), oligopolistic market (Boeing’s Spectrolab, Rocket Lab’s SolAero, a handful of European/Japanese suppliers) built for modest volumes. Great work if you can get it, but highly disruptive if someone comes along looking to fly a million birds.
• GaAs is the gold standard, but it’s also the bottleneck: Gallium‑arsenide multi‑junction III–V cells deliver incredible efficiency (c.30%) and serious radiation hardness. But at +$200/W, global production capacity in the single-digit MW/yr, and the PRC controlling ~98% of the world’s primary gallium supply, the technology simply does not pencil at GW-to-TW ambition. There aren’t many levers left to pull on cost, either – GaAs manufacturing is about as optimized as it’s going to get. 
• The curves must converge: Terrestrial solar has a learning rate of ~20%, meaning that every doubling of cumulative installed capacity drops costs by a fifth. That curve took module prices from $106/W in 1976 to <$0.40/W today, a 99.6% decline across 20+ doublings. Space solar has experienced almost none of this. It has lived in a cost bubble sustained by low volumes, captive government buyers, and less pressure to scale. But if orbital datacenters are to truly be a thing – orbital solar must break free of this bubble and ride the same manufacturing learning curves that have made terrestrial solar the cheapest energy source on Earth – via silicon-class manufacturing, tandems, and eventually perovskites. 

The Vendor Landscape 

Incumbents are investing, but face the classic Innovator’s Dilemma. Their existing III-V business is high-margin, low-volume, and built around a customer base (defense, institutional satellite operators) that values performance over cost. Retooling for silicon or hybrid arrays means potentially cannibalizing the product line that pays the bills, to serve a megaconstellation market that doesn’t fully exist yet (or want external suppliers). 

The largest player is Boeing’s Spectrolab, which has been building space-grade solar cells since the 1960s and generates an estimated $60M+ in annual space solar revenue. Their triple-junction GaAs cells can be found on everything from GPS satellites to Mars rovers. They are the poster child of a scaled incumbent vendor: excellent product, deep government relationships, zero incentive to blow up their own cost structure.

Then there’s Rocket Lab, which acquired SolAero Technologies in 2022 for $80M and now operates the world’s largest installed production capacity for gallium arsenide- and germanium-based solar arrays, generating ~$50M+ in annual space power revenue. Rocket Lab is the most unique case in this market because they’re the incumbent that sees the new wave coming, and they’re doing something about it. In February, Rocket Lab introduced silicon solar arrays “designed to power gigawatt-scale space-based data centers spanning kilometers in orbit,” alongside a hybrid III-V/silicon solution that lets them bridge both worlds. They secured $23.9M in CHIPS Act funding to expand their Albuquerque semiconductor production by 50%. As the company itself admits, their existing III-V solutions won’t provide the scale required by future constellations — especially orbital compute constellations.

The stark reality is that the satellite industry is projected to grow seven times by 2035, space-based data centers are on the horizon, and solar power supply chains are at risk of failing to keep up. A new solution is needed now, not years or decades into the future, and silicon is the answer.

-Rocket Lab 

Credit where it’s due: Rocket Lab is investing and acting quickly to avoid the innovator’s dilemma rather than ignore it. Whether they can move fast enough while still serving their defense and institutional book is the $B question. 

Germany’s Azur Space is quietly building capacity in the U.S. as well. Owned by 5N Plus, Azur sits on the premium end. They’re developing beyond triple junction, developing four-junction cells and researching metamorphic five-junction designs to squeeze out even more power. 

• Azur has been ramping aggressively — 35% capacity increase in 2024, 30% in 2025, another 25% planned for 2026. In mid-2025, Sierra Space, using Azur’s cells, opened a new $45M “Power Station” manufacturing facility in Broomfield, CO, touting an automated near-zero-touch process. 
• Nevertheless, this facility is only capable of producing a solar panel a day and a full satellite wing a week, with the goal of scaling to 100 wings per year.

A few other players worth mentioning: 

• Italy’s CESI produces cells with a similar chemistry and cost to Azur. Their capacity is fairly small, able to address the demand of about 8 small satellites per month, though they recently broke ground on a new €20M automated production line in Milan to expand manufacturing capacity. 
• Japan’s Sharp is also in the market as the exclusive supplier to JAXA with their very thin and light IMM3J thin-film cells. They are more of a boutique with relatively small manufacturing capacity, as their panels are often used for planetary and deep space exploration (their cells powered the SLIM lunar lander that made a pinpoint moon landing in January 2024). 
• US-based MicroLink Devices produces panels akin to those of CESI, and at their Illinois facility can currently produce around 100 kW/year with plans to expand capacity in the UK by about 1 MW.

All of these companies make excellent products. None of them have a path to the volumes that megaconstellations will demand. 1M Starlink satellites at even modest power per bird would require hundreds of gigawatts of space-qualified cells, which are orders of magnitude (OOMs) beyond what the entire incumbent vendor base can produce combined.

Then there are the startups. As disruptors, they don’t face the Innovator’s Dilemma problems that the incumbents have, but they have the other problems to deal with (space is hard, startups are hard, production is hard, etc.) 

Starpath recently unveiled Starlight Air, which it claims is the world’s lightest solar panel — 73 g/m², ~$15/W, using a crystalline structure measured in 100s nm. This is still expensive relative to terrestrial panels (~$0.10/W) but it’s a 10x reduction over legacy space solar. And if Earth is any guide, cost leaders win bigger in the solar industry. But at $15/W, when sized up against an industry where incumbents are at $200+/W, we’ll need to see about the economics at production scale or whether this is a loss leader to grab market share. 

Even if new suppliers can introduce new techniques, miniaturize panels, and bring cost down, Starship’s (or New Glenn’s) raw lift capacity makes weight optimization less decisive, while the fragile precursor supply chain at the center of an unprecedented historic trade war remains the real, seriously underdiscussed constraint. But more on that in a moment. 

The SpaceX Calculus 

Someone tweeted recently (we can’t find it, otherwise we’d link to it) that it’s more expensive to buy the solar panels for their satellite than it is to build the rest of the satellite. Whether we’re perfectly recollecting the X post, or regardless of whether the math checks out exactly, it captures the absurdity of the current cost structure. 

When faced with a similar dilemma, the high priest of vertical integration had two options: keep paying the premium for scaling-resistant, space-grade solutions, or figure out how to engineer OOM-more-cost-effective space power components itself. 

And by all accounts SpaceX seems to have opted for the latter. 

SpaceX broke from III-V convention years ago for Starlink, which ditched GaAs for silicon-based solar cells for its newer satellites. Starlink’s key supplier has been TSEC (Taiwan Solar Energy Corp), Taiwan’s largest PV manufacturer, which also happens to supply Tesla’s solar roof tiles. SpaceX chose silicon because GaAs was prohibitively expensive at Starlink volumes, and compensated for silicon’s lower efficiency by using significantly larger panels.

But even this arrangement has its vulnerabilities. In late 2024, SpaceX reportedly asked its Taiwanese suppliers to relocate manufacturing off the island, citing geopolitical risk. Taiwan is not China, but it sits squarely in the crosshairs of the same geopolitical tension that makes gallium supply fragile.

Which brings us back to Bastrop. The 16 open roles tied to a “Solar Cell Factory” – metallization engineers, wet-process engineers for cell texturing, process controls specialists — don’t exactly mean that SpaceX is entering the solar business from scratch. It is likely drawing on a combination of A) inspiration from Tesla, and B) vertically integrating what it already sources from overseas. This is the same thinking that led to Raptor engines, Merlin turbopumps, and Starlink user terminals being built in-house. If a component is critical, expensive, or supply-constrained, bring it inside the tent. 

For defense programs (Starshield, SDA Tranche 1, etc.) SpaceX reportedly still buys space-grade GaAs cells from Azur and SpectroLab. It is, in fact, “the largest commercial purchaser by far” of space-qualified solar cells. 

But, remember, we’re talking space datacenters here! So, where does an AI-pilled SpaceX go from here? The thought is to adapt from terrestrial solar — which has that enormous learning curve and massive install base, especially in China — rather than trying to wring more cost efficiencies out of legacy space solutions. And that is why the China solar belt trip sticks out so much to anybody closely monitoring the situation here, AKA us.

All Eyes on Perovskite 

Most of the conversation about next-gen space solar stops at silicon. Eventually it won’t, and behind closed doors today, it likely isn’t. 

Because, remember: Gallium arsenide is more or less maxed out on cost efficiencies. Silicon is proven at terrestrial scale but has hard limits on power-to-weight for space. 

Perovskite is the substrate that can flip the script. Perovskite solar cells achieve specific power densities of 23–30 W/g — 10–15x better than conventional silicon arrays (0.5–2 W/g) and 4–6x better than III-V multi-junction cells (~5.5 W/g), Lab-scale efficiencies have hit 27% for single-junction and 34.85% for perovskite-silicon tandems — comparable to SOTA III-V cells, at a fraction of the mass and material cost.

And energy density is massively consequential in space because the binding factor is still $/kg to orbit. The more power you can extract per gram of panel, the less mass you’re paying to launch. Perovskite’s power-to-weight advantage is so large it could enable an OOM reduction in cost for equivalent power generation.

And wait, we’re not done. Perovskites are actually better in space than on Earth in some respects. They demonstrate remarkable radiation hardness, maintaining performance under proton fluences that would destroy silicon cells. Under certain radiation conditions, perovskite cells exhibit a self-healing response where radiation-induced defects actually passivate pre-existing trap sites in the material — a phenomenon that does not occur in conventional semiconductors. Lab tests show perovskites retaining performance equivalent to ~20 years of cumulative LEO exposure.

The catch: perovskites are still at TRL 4-5 for space, vs. TRL 9 for silicon and III-V. Long-term reliability over multi-year missions is an open question. Thermal cycling, UV degradation, and encapsulation challenges haven’t been solved for production. This is likely why SpaceX has its eyes on Chinese HJT and perovskite lines — they’re scouting for the next generation while buying/building the current one.

If the endgame is truly TW-scale orbital power, neither GaAs nor silicon alone will get there. Perovskite — or more likely, perovskite-silicon tandems — is where the cost and density curves converge. 

Supply Shocks & Export Controls

Ready to come back crashing down to Earth for the hard truth? 

The U.S. woefully gave away the solar industry 20 years ago and remains heavily dependent on Chinese precursors to produce solar panels. 

A brief history for the uninitiated: Bell Labs invented the silicon solar cell in 1954. American companies took it from satellites to the White House roof to MW-scale terrestrial farms. Then China subsidized the entire supply chain, dumped cheap panels into the U.S. market, retaliated against American tariffs by locking out U.S. polysilicon, and today controls 90%+ of total solar PV manufacturing. The American companies that once led — SunPower, Solyndra, Abound Solar — are long gone.

For space solar, the chokepoint is even tighter. Recall the primary importance of gallium in legacy solutions (it’s in the name…gallium-arsenide). The USGS reports that no domestic primary gallium has been recovered in the U.S. since 1987. All U.S. gallium consumption runs through imports, and even when those imports come from Japan or Germany, the feedstock traces back to China, which produced 98% of the world’s low-purity gallium in 2024.  

The total global market is under 700 metric tons per year — small enough that Beijing can move the price with minor supply adjustments. Rotterdam gallium prices have spiked 150% above pre-control levels. And we can observe a deliberate and escalatory tightening of controls: 

• July 2023: China instituted a rule requiring exporters of gallium, germanium, and related chemical compounds to apply for special licenses and disclose detailed “end-user” tracing information to China’s Ministry of Commerce (MOFCOM). 
• December 2024: in response to expanded U.S. semiconductor restrictions, China functionally banned the export of gallium, germanium, antimony, and super-hard materials specifically to the United States, with extraterritorial jurisdiction blocking re-export via third countries.  
• January 2025: China added controls on gallium extraction technologies themselves — the specialized resins and ion-exchange methods that make recovery from bauxite cost-effective. One Chinese firm, Sunresin, dominates global supply of these chelating resins.
• November 2025: China temporarily lifted its outright ban on gallium exports to the US for one year while keeping strict reporting requirements and the military end-use prohibition in full effect.

This places a company like SpaceX squarely in between a rock and a hard place. They’re involved in defense, as 99.9% of space companies are, and Chinese suppliers trying to sell space-grade gallium components must still pass MOFCOM’s end-user checks. And because SpaceX is an American defense contractor, Beijing would likely deny the export license on national security grounds. The DoD has identified 11,000+ parts that require gallium, and ~85% of defense supply chains containing the metal include at least one Chinese supplier. Sourcing critical spacecraft hardware from a geopolitical rival is viewed as an unacceptable vulnerability. Economics and control of one’s destiny aside, Elon’s endless push for vertical integration is a lot easier to understand in this light – in the context of national security. 

Space Power: The Gating Factor, and Leapfrog Opportunity

The dream of orbital datacenters is still tethered to Earth by a fragile, geopolitical bottleneck. With China historically controlling the vast majority of the precursor materials and weaponizing export controls against the U.S. defense contractors, as virtually all dual-use “orbital datacenter” players are, the status quo will not work. And space-grade solar manufacturing capacity isn’t remotely close to what orbital datacenters will demand. 

It’s early days, but the industry seems to be splitting into three bets:

1. Silicon now, because the terrestrial supply chain already manufactures at 100s GW/yr scale. The world added 380 GW in the H1 2025 alone. SpaceX made this bet years ago with Starlink and is now bringing it in-house. Rocket Lab is making the same bet from the other direction, expanding its incumbent III-V operation toward silicon. 
2. Perovskite next, because the power-to-weight economics will be irresistible once the technology matures from lab to production. The China solar belt tours suggest SpaceX is already looking into this. The specific power numbers (23–30 W/g) represent a step change that will eventually make silicon look like what GaAs looks like now: too expensive per watt at the scale the market demands.
3. And hybrid III-V/silicon as the bridge, for defense and deep-space missions where considerations like efficiency per square meter still (and will) trump cost per watt, and where the legacy vendors will hold a premium niche for as long as mission requirements call for it (and upmarket/government buyers are willing to pay for it).

There is an irony here worth sitting with. The U.S. invented the solar cell, built an industry around it, and then handed it off to Shenzhen and China’s solar hinterland. But the learning curves that Beijing rode to dominance — 20+ doublings, 99.6% cost decline, 100s of GWs’ worth of annual manufacturing knowledge — are not confined to China. They created a global technical foundation of process engineering, materials science, and operational data on manufacturing high-efficiency cells at scale. And if perovskite clears its remaining durability hurdles, we’ll have a technology with no entrenched terrestrial incumbent, where the U.S. and Europe are still competitive in fundamental R&D, which is manufactured on flexible substrates that borrow more from printing than from semiconductor fabrication.

This foundation is an on-ramp for space-grade silicon, with perovskite in the fast lane behind it, and it represents an opportunity for the U.S. to ride the learning curves it lost on Earth and leapfrog into the lead for a new class of power manufacturing: this time, for orbit. 

In sum, don’t sleep on space power!