Baiju Bhatt walked on stage in Palo Alto on 26 October and described a rocket that refuses to die. Its first stage lands. Its second stage never comes home. It becomes the payload.
That is Big Iron, unveiled at an event called Big Sky Rising. The company began in 2024 as Aetherflux, a space-solar venture, and rebranded this spring. Bhatt co-founded Robinhood and left in March 2024 to build power infrastructure in orbit. The pitch has narrowed since. Beaming sunlight to Earth was the warm-up. Running GPUs above the weather is the business.
A second stage that clocks in
The vehicle is 86 metres of stainless steel, sized to the width of a road so it can be built in one place and shipped to another. Nine methalox Clydesdale engines power the first stage, each rated at 300,000 pounds of thrust. Cowboy says it will lift more than 15,000 kilograms to sun-synchronous orbit.
The upper stage is the product. After separation it becomes a satellite 132 metres long and 73 metres wide, carrying five racks of Nvidia Vera Rubin GPUs. Cowboy puts the payload at about one megawatt. SpaceX's planned Starmind satellites are quoted at roughly 250 kilowatts each.
If you build a small satellite with GPUs on it, you're going to have to put radiators on it. And so you're paying for that same metal twice to lift it into space.— Joseph Yaffe, chief operating officer, Cowboy Space
Every kilogram does two jobs. The rocket's skin becomes the data centre's radiator. The ammonia propellant becomes coolant. The nitrogen that pressurises the stage also unfolds the solar wings. Mass is the whole game in orbit. Cowboy's bet is that a vehicle designed around that constraint beats a general-purpose rocket hauling a separate satellite.
The idea is not new in principle. It is new in packaging. As we wrote in October, space-based solar just signed its first paying customer, a 20-year PPA for orbital baseload. That deal sells energy down to Earth. Cowboy keeps the energy up there and sells the compute instead.
The bill comes due in launches
One megawatt per launch sets the arithmetic. Twenty thousand satellites means twenty thousand launches, plus replacements. Cowboy wants to fly multiple times a day from sites it calls Trailheads, each with about four pads.
It also sells rides. Big Iron Express offers outside customers up to a megawatt of orbital power for compute or anything else. That revenue matters while the company scales its own constellation.
The capital base is thin against the ambition. Cowboy raised a $275 million Series B in spring 2026, led by Index Ventures, after $80 million earlier. Backers include Breakthrough Energy Ventures, Construct Capital and IVP. SpaceX already flies the rockets and has filed for a far larger compute constellation.
Bhatt says the design puts Cowboy on a path to cost parity with terrestrial data centres, with orbital compute cheaper than AI compute on the ground. The company has pointed to parity around 2031. Those are targets, not results.
Permission is the missing part
Regulators have not signed off. Cowboy filed with the FCC in May for Stampede, a constellation of up to 20,000 satellites. The Space Bureau sent a deficiency letter on 28 August asking for details including satellite dimensions. The company is also seeking a waiver of deployment milestones. Commercial service, it argues, could start with one satellite in orbit.
Hardware is early. Cowboy tested part of a Clydesdale combustion chamber at NASA's Stennis Space Center in October. A team of about 30 built and tested it in 90 days. The company claims it is almost a year ahead of schedule. First flight is targeted for December 2028. Bhatt's own caveat: a lot has to go right.
Reason-1, the first satellite, launched on 1 October aboard a SpaceX Falcon 9 rideshare. Its job is laser power beaming from orbit to a ground receiver. Cowboy says the first attempt is weeks away and testing has shown zero red flags. Reason-2, due in 2027, is meant to run GPUs in orbit, including weather data for NOAA.
The scarcity Cowboy is selling is not chips. It is power. Grid connection queues for large US data centres run five to seven years in some markets, according to the company. A solar array in dawn-dusk orbit sees the sun almost continuously. Whether that beats a terrestrial megawatt depends on launch cost, thermal management and a regulator who has not yet said yes.
The rocket is the easy part to describe. The hard part is flying it 20,000 times.