Inside Starcloud’s $2B Space Factory

Summary

• Starcloud’s claim to cheaper orbital compute depends on launch prices below around $500/kg. Philip Johnston compares that threshold with about $5,000/kg on Falcon 9 and Starship’s designed internal costs of $20–$40/kg. Continuous sunlight removes battery and backup-power requirements; he also cites an eightfold per-square-meter energy advantage for space solar panels. “The whole thing is actually contingent on the launch cost.”
• The $2 billion company established its operating model with a roughly $2.5 million satellite built and launched within 21 months of founding. Johnston says a prime contractor quoted $75 million–$100 million for Starcloud 1; the company now has 30 employees, only two non-engineers. A broken $200,000 vibration table repaired over a weekend captures the approach: “not take no for an answer.”
• Chip reliability could break the economics even if cooling works. Johnston says a 10% or 20% higher chip failure rate in space would make “the whole economics of the model fall apart,” because chips dominate costs. Starcloud 2 underwent three months of chip burn-in; its planned January launch is the next proof point for radiators Johnston claims deliver at least tenfold lower mass per watt than the ISS.
• Government satellite inference is an initial use case before orbital compute reaches terrestrial cost parity. Johnston describes test-phase workloads processing radar and hyperspectral imagery, replacing large image downlinks with actionable coordinates in “minutes rather than hours and days.” Longer term, he expects coding agents, business-processing agents and ChatGPT queries, while tightly networked frontier-model training remains harder to accommodate.
• The scaling plan combines small independent compute platforms with mass manufacturing. Starcloud 3 is designed around roughly two tons and 235 kW, with up to 50 vehicles per Starship launch; Johnston describes an initial 88,000-satellite constellation supporting about 20 GW. His larger estimate is about 10 TW in dawn-dusk sun-synchronous orbit, approximately 20 times the entire US power grid.
• The $450 million raised is funding launch reservations and a factory before the launch market fully opens. Johnston hopes commercial Starship capacity becomes available from 2028 onward, while Starcloud has also booked alternatives including Stoke and Firefly. Approximately 100,000 square feet could support production of 100 satellites weekly, with an eventual line “like an auto manufacturing line.”
• Johnston’s largest investment prediction is that SpaceX might exceed $100 trillion in value within ten years. Molly O’Shea frames that outcome as contingent on execution and no world war; Johnston’s mechanism is physical recursive manufacturing once “Optimus is producing Optimuses.” He also categorically predicts SpaceX will buy Tesla, calling it “just a fact”; the interview supplies his assertion and timing theory, rather than evidence of a transaction.

Deep dive

1. A $2.5 million satellite established the operating model

• Johnston says Starcloud went from founding to orbit in 21 months, a record against a previous four-year benchmark. He says Adi and Ezra built the earlier fastest satellite by hand. Starcloud 1 launched in November “last year” carrying NVIDIA H100 hardware. A prime contractor quoted $75 million–$100 million for the mission; Starcloud did it for about $2.5 million. His lesson: “you can do things very cheaply.”
• Seattle supplies the relevant combination of skills. Johnston says 90% of satellite mass launched in the preceding few years was designed, built and tested in Redmond. About half the team previously worked at SpaceX; the other half comes from hyperscalers, with AWS and Azure nearby. The company has 30 employees, only two non-engineers, and deliberately keeps staffing lean.
• Procurement illustrates the approach: a new vibration table cost about $1.5 million with a roughly two-year lead time, so Starcloud bought a broken one for $200,000. The 16-ton machine arrived waterlogged, with a major internal leak. Despite an expert declaring repair impossible, Feilden and the team rebuilt it over a weekend sufficiently to complete spacecraft testing: “not take no for an answer.”

2. Cooling is a manufacturing problem with an orbital test ahead

• Starcloud representatives identify cooling and radiators as the largest technical challenge, while framing the remaining problem as engineering cost rather than feasibility: “We know it’s possible.” Johnston supplies the commercial question: “how do you make it super cheap and how you make it manufacturable at scale.” The team adapts liquid-cooling technology used in terrestrial data centers.
• Starcloud 2 has a roughly 120-foot deployed wingspan. Johnston says its January launch will carry “by far the largest commercial deployable radiator in space.” Against the International Space Station, he claims at least tenfold lower radiator mass per watt dissipated and about 500-fold lower cost. He concedes the ISS is an easy cost benchmark; the mass improvement is the achievement.
• A Starcloud representative explains the accumulator being vibration-tested: heated coolant expands, and the component accommodates that expansion without bursting the cooling system. Launch qualification also checks resonance. Johnston thinks SpaceX’s minimum first resonant frequency is around 40 Hz; a structure oscillating slowly could move neighboring payloads and destabilize the rideshare stack.
• Engineering manager Bailey Montaño designed and fabricated the radiators and recalls the first fully integrated cooling-system test in Colorado. Final satellite assembly then had only three weeks available, with Montaño building the bus at the provider’s office. Testing is Montaño’s favorite stage: “That’s where the rubber really hits the road.”

3. Launch below $500/kg is the terrestrial crossover

• Johnston’s comparison starts with a new solar project powering a terrestrial data center. Its largest cost is permitted land, depending on location; next come batteries and backup power, then solar cells. In the intended orbit, he says Starcloud needs neither permitted land nor batteries and backup, because it remains in sunlight continuously.
• Johnston says Starcloud needs eight times fewer solar cells, citing an eightfold per-square-meter energy advantage for space solar panels. Launch is the principal added expense. Combining these savings and costs produces his crossover of “around $500 a kilo,” versus about $5,000/kg today on Falcon 9.
• Starship’s designed internal launch cost, in Johnston’s account, is $20–$40/kg. The model therefore works if SpaceX sells capacity at anything below roughly ten times marginal cost, or potentially more. He considers that plausible because Falcon 9 currently charges two or three times marginal launch cost. Those are his expectations for future pricing, rather than an available commercial Starship quote.

4. Radiation and chip failures can erase the savings

• Johnston considers chip survival more pertinent than cooling, which he says the team has “basically solved.” Radiation shielding and particle-accelerator testing matter because chips are the most expensive component. If their failure rate is 10% or 20% higher in space, “the whole economics of the model fall apart.” Lower infrastructure costs cannot absorb that loss.
• The satellites will be inaccessible after launch, so the operating model resembles Starlink: duplicate critical systems such as pumps and over-provision components that degrade. Solar panels, for example, lose a few percent annually. Shielding and testing address radiation exposure; redundancy and excess capacity address failures the company still expects during service.
• Starcloud 2’s chips underwent three months of hot, full-power burn-in to expose early failures before launch. Johnston calls this chip “infant mortality” and says subsequent failure probability diminishes dramatically after surviving that period. Asked why early failures happen, his answer is simply “I don’t know”; the mitigation rests on the observed failure pattern.

5. Government inference supplies the first workloads

• Starcloud 1’s payload includes three NVIDIA chips, two Arm chips and AMD CPUs. Johnston says it trained the first model in space, nanoGPT from Andrej Karpathy, and “ran the first version of Gemini.” The satellite has since moved into useful, mainly test-phase workloads for government customers, including inference on synthetic aperture radar and hyperspectral data.
• His concrete example is monitoring whether a vessel crosses a boundary. Conventional satellites collect enormous imagery volumes, may wait at least an hour and a half for a ground station, and cannot necessarily downlink much. Optical links can instead send imagery to Starcloud for inference; it then transmits vessel coordinates in “minutes rather than hours and days.”
• On security, Johnston argues that striking a satellite moving about 17,000 miles an hour is harder than attacking a Virginia data center. He points to Starlink and Starshield usage in Ukraine as existing cases of adversarial pressure. Starcloud is using the same kind of cryptography, he says, to protect its satellites.
• Johnston also argues that approaching objects would be visible, claiming objects larger than a one-inch cube are trackable. He describes interference with the satellites as an act of war against the US and invokes the Outer Space Treaty and Space Force’s deterrence role. These are his security and legal characterizations.

6. Launch availability constrains the transition

• Feilden says booking Starcloud 1’s Falcon 9 rideshare roughly 21 months ahead was essentially the company’s first action; the initial reservation could be made by credit card. That simplicity is changing. He describes Falcon 9 rideshare as going away while Starship has yet to ramp: “we’re definitely in a crunch time at the moment.”
• Launch insurance was relatively inexpensive given Falcon 9’s record, which Feilden describes as 300 consecutive launches without a failure. Insuring a first-of-its-kind spacecraft’s subsequent operation is harder. He remains “slightly on the fence” about payload insurance: financial compensation cannot recover lost time, the startup’s most valuable resource. Future coverage will be decided case by case.
• Johnston expects Starship reusability probably within a year, or at least upper-stage capture; reflight might take longer. From 2028 onward, Starcloud hopes SpaceX will sell commercial capacity. It has also booked other vehicles, with Stoke and Firefly already announced and further announcements expected, to provide options if that timing slips.
• Asked about Rocket Lab’s delayed Neutron, Feilden says the company announced a 2027 first flight the previous day. “Composite manufacturing is really hard,” especially large structures and first-time tooling. Nevertheless, he has substantial confidence in Rocket Lab’s ability to operate regularly, citing its 100 Electron launches: “we’re definitely rooting for them as well.”

7. Independent modules are the route to gigawatts

• The Starcloud 3 manufacturing demonstrator represents a roughly two-ton fully loaded vehicle rated at 235 kW. It is designed to pass through Starship’s “PEZ dispenser door,” with up to 50 per launch: two across and 25 layers tall. Upcoming individual missions still use Falcon; “the future is definitely Starship,” Feilden says.
• Johnston distinguishes this architecture from terrestrial frontier-model training, which requires a highly interconnected cluster. For a long time, he expects Starcloud will probably avoid docking spacecraft together. Most orbital workloads would therefore be inference: coding agents, back-office business-processing agents and ChatGPT queries, distributed across independent platforms.
• Using 200 kW units in his deployment arithmetic, Johnston estimates close to 10 MW of additional compute per Starship launch. He describes a filing for 88,000 satellites supporting the first 20 GW. “That’s just the first 88,000,” he adds; the planned constellation is an initial deployment rather than his eventual limit.
• Johnston thinks dawn-dusk sun-synchronous orbit can accommodate about 10 TW, roughly 20 times the entire US power grid. Beyond that, he names cislunar space, MEO, beyond GEO and Lagrange points. Answering Molly’s concern about blocking sunlight, he says the polar geometry keeps satellites out of Earth’s shadow and prevents them casting shadows on Earth.

8. The capital funds launch slots and a production line

• Starcloud is expanding into adjacent buildings, with one section coming online in November and another available in February next year. About 100,000 square feet could support 100 satellites weekly. Space is needed both for manufacturing and for testing Starcloud 3’s 400-foot deployables. The intended result looks “like an auto manufacturing line,” with chassis receiving components as they move.
• Molly puts total capital raised at $450 million. Johnston says part is reserving constrained launch capacity, particularly from 2028–2029 onward; the remainder supports factory fit-out, hiring and materials. He says the company is making about a million dollars a month in interest. The cash will increasingly be spent as the build-out progresses.
• Staffing could reach 50 by year-end and maybe 100 by the following year. Engineers alternate between CAD work upstairs and hands-on wiring and testing below. A technician group—four, or perhaps five, Johnston says—handles fabrication, including the aluminum compute-module enclosure. The production model keeps design and physical assembly closely connected.

9. Johnston’s SpaceX prediction depends on recursive manufacturing

• Johnston thinks SpaceX “might be worth more than $100 trillion in 10 years.” Molly explicitly conditions that prospect on plans working and no world war. His mechanism is Optimus building Optimus factories: once “Optimus is producing Optimuses,” manufacturing growth accelerates. He compares the process with recursive self-improvement transferred into the physical world.
• His Tesla prediction is categorical: “This isn’t a hot take. This is just a fact. SpaceX is going to buy Tesla.” Johnston ties timing to stock unlocks and concerns that participants would dispute the price. He predicts few Optimus announcements for the next year, with more following a merger. The conversation provides no transaction evidence beyond his assertions.
• On the Roadster announcement expected that evening, Johnston says he has heard about cold-gas thrusters from the spacecraft and a compressor compressing air to fire out the back, enabling extraordinary acceleration and cornering. He also thinks lift is possible. Molly suggests it might be an eVTOL. He disagrees: “I would be extremely surprised if it’s an eVTOL,” arguing that such a development would already be known.

10. Fermi Explorer extends the cheap-spacecraft thesis to another star

• Johnston describes a separate nonprofit effort, Fermi Explorer, to send the first spacecraft toward Alpha Centauri, about four light-years away. Starcloud 1’s cost experience encouraged the idea. An MIT physical AI lab analyzed a trajectory he says could put a spacecraft on that journey for about $10 million–$15 million.
• The mission would use existing technology: solar-powered Hall-effect ion thrusters, five years spiraling toward the Sun, then another five years firing around perihelion into expanding ellipses before departure. It targets where the moving star system will intersect the spacecraft’s path in about 80,000 years, rather than its present location.
• Johnston hopes it becomes both “the first to leave” and “the last to arrive,” overtaken by later propulsion technologies. Even propulsion only 20% faster, he imagines, might allow a colony to have existed for 15,000 years before arrival. Its inhabitants could applaud the spacecraft across the finish line and put it in a museum.
• The underlying motivation is the Fermi paradox. Johnston estimates even today’s technology could settle the galaxy in roughly five to ten million years, making the lack of visible evidence troubling. He sees two possibilities: humanity is exceptionally rare, which he welcomes, or “intelligent life doesn’t last very long.” The mission is intended to make people consider that choice.