What if the next big leap in computing power did not happen on the ground at all? Imagine a rocket that refuses to treat its upper stage as disposable hardware. Instead of letting that expensive piece of engineering burn up or drift away, the whole thing stays in orbit and turns into a working data center. That is the core idea behind a project that just stepped into the public eye, and it has me thinking about how quickly the boundaries between launch vehicles and satellites are starting to blur.
The Big Iron Concept Takes Shape
Cowboy Space Corporation, the company formerly known as Aetherflux and founded by Robinhood co-founder Baiju Bhatt, recently pulled the curtain back on its rocket called Big Iron. The vehicle stands roughly 86 meters tall. Its stainless-steel first stage relies on nine liquid oxygen and methane engines named Clydesdale. Each of those engines is rated at about 300,000 pounds of thrust, giving the stack a total of 2.7 million pounds. According to the company, that power should allow more than 15,000 kilograms to reach a sun-synchronous orbit.
Once the first stage does its job, the upper stage does not separate and float off as a conventional payload. It becomes the satellite itself. In its final configuration the structure measures about 132 by 73 meters and can host five racks of Nvidia Vera Rubin GPUs. The design tries to squeeze multiple functions out of the same hardware. The rocket skin doubles as a radiator. Ammonia propellant also serves as coolant. Nitrogen used for pressurization helps unfold the large solar wings. In my view that kind of multi-use thinking is exactly what space hardware needs if costs are ever going to come down in a meaningful way.
Our design puts us on a path to cost parity with terrestrial data centers.
– Baiju Bhatt
That statement is ambitious. Cost parity with ground-based facilities has been the holy grail for anyone talking about orbital compute for years. Whether the claim holds up remains to be proven, of course. Nothing from this program has flown yet. The company aims to finish the first rocket by the end of 2027 and attempt a first flight in 2028. Those timelines feel aggressive for a brand-new vehicle, yet the team has already shown it can move quickly on smaller pieces of hardware.
Engine Progress and Early Testing
Part of a Clydesdale engine was tested not long ago at NASA’s Stennis Space Center. A small team of about thirty people built and tested the combustion chamber in roughly ninety days. That pace is impressive by traditional aerospace standards. Still, a combustion chamber is only one piece of a full engine, and a full engine is only one piece of a complete vehicle. Plenty of hard work sits between a successful chamber test and a reliable orbital flight.
Government customers, according to company executives, want to see a completed rocket before they commit serious business. That preference makes sense. Many alternative launch vehicles have announced bold plans only to struggle when the calendar reaches the first flight date. Big Iron sits in that same category for now. The difference, if there is one, may lie in how tightly the launch system and the data-center payload are integrated from day one.
From Launch Vehicle to Orbiting Compute Platform
The upper stage is meant to operate in a dawn-dusk orbit. That choice keeps the solar arrays in near-continuous sunlight while the structure itself stays relatively cool on the shaded side. Five racks of high-end GPUs represent a meaningful amount of compute for an early orbital system. The company is not claiming it will replace terrestrial hyperscale facilities overnight. The pitch is more measured: start with a useful amount of capacity, prove the thermal and power architecture, then scale.
Power and cooling remain the two biggest headaches for anyone trying to put serious compute in space. Big Iron’s approach tries to solve both by reusing the rocket’s own systems. The skin radiates heat. The propellant absorbs heat. The pressurant gas helps deploy the solar arrays. It is an elegant set of dual-use decisions on paper. Reality will test every one of those assumptions once the vehicle reaches orbit and the GPUs start drawing power.
A Larger Constellation Ambition
Big Iron is only the launch half of a broader plan. Earlier this year the company filed with regulators for a constellation of roughly 20,000 satellites under the name Stampede. Approval has not yet been granted. The filing asks for waivers on the usual deployment schedule and milestone requirements. The company states that commercial service could begin with as few as one satellite in orbit. That is a low bar, but it also reflects a practical approach: get something useful flying, then grow from there.
Funding has been secured. A Series B round brought in 275 million dollars this past spring. The first demonstration satellite, called Reason-1, launched on October 1 aboard a rideshare mission. Its job is to test laser power beaming from orbit down to Earth. A second demonstration vehicle, Reason-2, is planned for 2027 and is intended to run GPUs in orbit. Those two missions should give the team real data on both power delivery and orbital computing before the full Big Iron vehicle is ready.
I find the power-beaming experiment particularly interesting. If you can move energy wirelessly from orbit to the ground with acceptable efficiency, a whole set of new business models opens up. The same technology might eventually support orbital data centers that need extra power or that want to sell energy as a side product. It is still early days, but the direction of travel is clear.
Selling Excess Capacity and Courting Defense Customers
While the company builds its own constellation, it is also offering launch services to others. The Big Iron Express program invites outside customers who need up to a megawatt of power in orbit to ride along. That is a smart way to fill capacity and generate revenue while the internal data-center business matures. At the same time, executives have begun conversations about competing for national security launch work. Existing providers dominate that market today, yet the door is never completely closed to new entrants that can demonstrate reliability and unique capabilities.
One competitive pressure sits in plain sight. Another major space company has already filed paperwork describing plans for a million data-center satellites that would deliver on the order of 100 gigawatts of compute. That is an enormous number. Whether those satellites ever reach orbit in the quantities described is another question. What matters for Cowboy Space is that the market for orbital compute is no longer theoretical. Multiple teams are now racing to prove the concept works at useful scale.
Technical Choices That Stand Out
The decision to use stainless steel for the first stage echoes a design philosophy that has gained popularity in recent years. Steel is heavier than advanced composites or aluminum-lithium alloys, yet it is cheaper, more forgiving during manufacturing, and better at handling high temperatures. For a vehicle that aims to fly often and eventually support a large constellation, those practical advantages can outweigh the mass penalty.
Methane as a fuel also fits a modern trend. It is denser than hydrogen, easier to handle than kerosene in some respects, and can be produced from relatively simple chemical processes. The combination of steel and methane has already been proven on other vehicles, so the technical risk on the propulsion side feels manageable compared with the risk of making a data center work reliably in orbit.
- Nine Clydesdale engines providing 2.7 million pounds of total thrust
- Upper stage converting into a 132-by-73-meter orbital structure
- Capacity for five racks of high-performance GPUs
- Dual-use systems for radiation, cooling, and solar array deployment
- Target of more than 15 metric tons to sun-synchronous orbit
Those numbers look solid on a presentation slide. The real test will come when the first vehicle stacks on the pad and the countdown begins. Until then, every claim remains aspirational.
Challenges That Still Need Answers
Thermal management in orbit is unforgiving. Heat rejection depends almost entirely on radiation. The idea of using the rocket skin itself as a large radiator is clever, but the surface area, surface properties, and internal heat-transfer paths all have to work together under real orbital conditions. Dust, atomic oxygen, and thermal cycling will age those surfaces over time. The company will need robust testing data before customers trust multi-year missions.
Radiation hardening of the compute hardware is another open item. Commercial GPUs are not designed for the space environment. Shielding adds mass. Error-correction software helps, yet it also reduces effective performance. Finding the right balance between protection, performance, and cost will be critical if the system is ever to approach terrestrial economics.
Then there is the simple question of launch cadence. Building one rocket is hard. Building many rockets on a predictable schedule is much harder. The constellation plans imply a need for frequent flights. That requirement places pressure on manufacturing, supply chain, and range operations all at once. History shows that most new launch companies underestimate how long it takes to reach high flight rates.
Why the Idea Still Feels Compelling
Despite the challenges, the basic logic is hard to dismiss. Energy in space is abundant once you have solar arrays. Cooling becomes a design problem rather than a water-and-electricity problem. Latency to certain users can be lower than routing traffic through distant terrestrial facilities. And the ability to place compute close to Earth-observation sensors or to other space assets opens applications that ground data centers simply cannot serve as well.
I have followed several orbital-compute concepts over the past few years. Most of them treat the launch vehicle as a pure transportation service and the data center as a separate satellite. Cowboy Space is trying something different by making the upper stage the data center. That integration reduces interfaces, eliminates a separate deployment step, and lets the same structural mass serve multiple purposes. Whether the savings prove large enough to reach cost parity remains an open question, but the approach is more elegant than many competing ideas I have seen.
The personal angle is hard to ignore. Baiju Bhatt left a highly successful fintech company and chose to pour energy into space-based power and computing. That transition from consumer finance to heavy aerospace is unusual. It suggests a long-term bet on infrastructure that sits above the atmosphere rather than inside data-center buildings. Investors who wrote the recent large check appear to share at least some of that conviction.
Looking Ahead to the First Flights
The next couple of years will be decisive. Reason-2 is scheduled to fly in 2027 and run actual GPUs. That mission should answer basic questions about thermal performance, radiation effects, and software reliability. If those results look promising, the pressure to complete Big Iron on schedule will only increase. A successful first flight in 2028 would put the company in rare company among new launch providers.
Even a partial success would move the conversation forward. The industry needs more data points on what it actually costs to keep high-performance computers running in orbit for months or years. Every new vehicle that reaches space and publishes honest performance numbers helps the entire field. In that sense, the arrival of Big Iron is welcome regardless of whether every ambitious claim ultimately proves correct.
Perhaps the most interesting aspect is how quickly the discussion has shifted. A few years ago orbital data centers were mostly the stuff of concept papers and conference talks. Today multiple teams are building hardware, filing for spectrum and orbital slots, and raising serious capital. The technology is still immature, the economics are still unproven, and the regulatory path is still incomplete. Yet the direction of travel feels unmistakable.
Big Iron represents one concrete attempt to turn that direction into flying hardware. The rocket is tall, the engines are powerful, and the upper stage is designed to stay useful long after the first stage has returned to Earth. Whether the finished product lives up to the early renderings will depend on execution over the next several years. For now the concept itself is enough to keep the conversation lively and the industry watching closely.
The path from a successful chamber test to a reliable megawatt-class orbital data center is long. Many technical and operational hurdles remain. Still, the decision to treat the upper stage as permanent infrastructure rather than disposable hardware feels like a genuine step forward in thinking. If the team can deliver on the thermal, power, and reliability challenges, the idea of cost parity with terrestrial facilities may move from marketing language into engineering reality. That possibility alone makes the project worth following as the hardware moves from design screens toward the launch pad.
In the end, space has always rewarded those willing to rethink assumptions that everyone else treated as fixed. Discarding upper stages after a single use is one of those assumptions. Big Iron questions it directly. The coming years will show whether that question leads to a practical new class of orbital infrastructure or simply another ambitious design that never quite reaches the flight rate needed for commercial success. Either way, the conversation about where computing power belongs has become more interesting.
