Robinhood Co-Founder Is Building Data Centers in Space. His Startup Just Hit a $2 Billion Valuation.

Aetherflux, the orbital data center startup founded by Robinhood co-founder Baiju Bhatt, closed a $120 million Series B led by Andreessen Horowitz in March 2026. The company is building computing infrastructure in low Earth orbit (400-600km altitude), powered by solar photovoltaic arrays and cooled by radiative heat rejection rather than atmospheric cooling. Bhatt presented the concept at NVIDIA GTC 2026, framing it as a solution to the two primary constraints on terrestrial AI data centers: energy cost and cooling capacity.

Why Data Centers in Space Are Not As Absurd As They Sound

The physics case for orbital computing rests on three facts. First, solar energy in orbit is approximately 5 to 10 times more efficient than terrestrial solar because there is no atmosphere to absorb photons, no weather to block panels, and no nighttime (a satellite in the right orbit receives near-continuous sunlight). Second, cooling is free in space: radiative cooling in vacuum is more efficient than any terrestrial cooling system. Data center cooling accounts for 30 to 40% of total energy consumption on Earth. In orbit, that cost drops to near zero. Third, orbital data centers face no land use restrictions, water consumption limits, or grid connection bottlenecks, all of which are becoming binding constraints on terrestrial data center construction in 2026.

The economics case is less clear. Launch costs (SpaceX Falcon 9: ~$2,700/kg to LEO, Starship target: ~$100/kg) determine whether orbital compute can compete with terrestrial pricing. At current launch costs, the capital expense of putting hardware in orbit exceeds terrestrial data center construction costs by 10x to 100x. At Starship’s target costs, the gap narrows significantly but does not close.

The Physics Case For and Against Orbital Compute

Real physics advantages: Radiative cooling to 3K cosmic background (vs. 15-35C ambient terrestrial). Solar irradiance ~1361 W/m² without atmospheric absorption. No land acquisition, zoning, or water use permits. Proximity to satellite communications infrastructure.

Hard unresolved problems: 230ms minimum round-trip latency from LEO (speed of light). 90-minute orbital period creates power intermittency. Hardware servicing requires launch ($2,000-5,000/kg to LEO). Radiation degrades semiconductors ~10x faster than terrestrial. The 230ms latency is not an engineering problem: it is a physics constraint. Any AI inference workload with real-time requirements cannot be served from LEO regardless of hardware quality.

The Three Questions the Pitch Deck Does Not Answer

1. Who is the customer? Training large models is latency-tolerant, but hyperscalers (Google, Microsoft, Meta) already have massive terrestrial training clusters and the capital to build more. The customer who cannot build terrestrial compute but can afford $5,000/kg launch costs does not obviously exist at scale.

2. How does hardware refreshing work? Terrestrial data centers refresh GPU hardware every 2-3 years. Orbital data centers require a launch for each hardware refresh. At current Starship pricing, a single rack refresh costs millions in launch fees alone.

3. What is the radiation hardening strategy? Standard NVIDIA H100s are not radiation-hardened. Rad-hard computing is 10-100x more expensive per FLOP than commercial silicon. Aetherflux has not disclosed their semiconductor strategy for radiation tolerance.

The Baiju Bhatt Pivot

Aetherflux originally focused on beaming solar power from orbit to terrestrial receivers via laser. The company pivoted to orbital computing in 2025 after concluding that the terrestrial power transmission economics were unfavorable. The pivot keeps the core capability (space solar power systems) while changing the customer: instead of selling power to terrestrial grids, sell compute powered by space solar to AI companies.

Bhatt’s credibility from co-founding Robinhood (which achieved a $32 billion valuation before its IPO) gives Aetherflux access to top-tier venture capital. The $2 billion valuation prices Aetherflux as a pre-revenue company with a potentially transformative technology, which is the same valuation framework that funded SpaceX before it had a single paying customer.

What Has to Go Right

For Aetherflux to succeed, several things must happen simultaneously. SpaceX’s Starship must achieve routine operation at prices near its $100/kg target. The satellite computing hardware must survive the radiation environment of low Earth orbit without unacceptable error rates. The latency from ground-to-orbit-to-ground round trips must be acceptable for the target workloads (batch training: yes; real-time inference: probably not). And the company must solve the data bandwidth problem: getting training data up to orbit and results back down requires high-throughput optical or radio links that do not yet exist at the necessary scale.

The competitors are literal: Lumen Orbit, founded in 2024, is pursuing a similar concept with a solar-powered orbital data center targeting 2027 deployment. Microsoft Azure Space and AWS Ground Station provide cloud-edge compute for satellite operators but do not offer orbital compute as a service. The market for orbital computing does not exist yet. Aetherflux and Lumen Orbit are both betting that terrestrial data center constraints (power, cooling, land, water) will create demand for orbital alternatives within 5 to 7 years.

The honest assessment: orbital data centers are a real technology with real physics advantages that face massive engineering and economic challenges. The $120M Series B funds a proof-of-concept deployment, not a commercial data center. The first data center satellite targeting 2027 will be a technology demonstrator, not a commercially competitive compute platform. If the demonstrator works, the path to commercial viability depends on launch cost reductions that are outside Aetherflux’s control. Bhatt knows this. The bet is that solving the technical challenges now positions Aetherflux to capture a market that will exist in 2030, even if it does not exist today.

Sources: Aetherflux Series B announcement; Bhatt GTC 2026 panel; Andreessen Horowitz portfolio blog; SpaceX Starship commercial pricing; NASA radiation effects documentation. Market context, not financial advice.