Where Will Orbital Computing Be by 2035? Cowboy Space's 'Big Iron' and the 20,000-Data-Center Ambition as a Test of Space-Based AI Infrastructure

The orbital compute thesis is, at root, a launch-cost and mass-per-watt thesis. Everything else follows from those two numbers.

Industry Implications: A Decade Scenario

2026-2028: Vertical integration and demonstration

Expect a period of announcements outpacing hardware. Big Iron's late 2028 debut sits at the optimistic end of new-vehicle history, and the sector has seen slips, with a medium-lift vehicle moving from a 2024 target to 2027. A reasonable planning assumption is that first launches slip by 6 to 18 months, which would put Big Iron's first flight in 2029. Early orbital compute is likely to be small demonstrators, such as AI-run payloads and edge processing aboard existing satellites. Examples already in the news include autonomous maneuvering software tested on NASA's Starling satellites and an AI-run plant growth chamber headed to the ISS. These are not data centers, but they show the software maturing.

2029-2031: Hundreds, not thousands

If Big Iron flies by 2029-2030, a realistic build-out is dozens to low hundreds of operational compute satellites by 2031, constrained by launch cadence and production lines. Growth to thousands requires reusable launch, high flight rates and satellite manufacturing at a pace seen so far only in broadband constellations. Falcon 9 shows the cadence benchmark: it launched 27 Starlink satellites on a booster flying for the 26th time on Oct. 27. A new entrant must approach that kind of reuse to hit its numbers.

2032-2035: Niche or infrastructure

There are two plausible end states by 2035.

1. Infrastructure case: Orbital compute serves latency-tolerant AI training and inference workloads, with power and launch costs low enough to compete with terrestrial sites in power-constrained regions. Constellation size in the thousands. Revenue is concentrated among a few vertically integrated operators.
2. Niche case: Orbital compute serves space-native workloads (processing Earth observation data before downlink, secure government processing, in-orbit autonomy) while terrestrial data centers absorb most AI demand through new power generation. Constellations number in the hundreds.

The second case is the more conservative baseline. The first requires a convergence of launch and hardware economics that has not yet been demonstrated.

Effects on adjacent sectors

• Launch: Demand for high-cadence, heavy-lift reusable vehicles rises, favoring operators with proven reuse. Government support for launch infrastructure, such as the loan guarantee for a second Florida pad, indicates policy tailwinds for capacity.
• Earth observation: On-orbit processing reduces downlink needs. Companies such as EarthDaily, which just won a $3 million NASA task order to test compatibility with Landsat, could pair data with orbital analytics.
• Communications: Orbital compute needs inter-satellite and ground links, increasing demand for optical terminals and interoperable networks.
• Servicing and debris: Large constellations raise the need for autonomous rendezvous and removal, linking to work on failure-tolerant servicing and to the UK's active debris removal mission.
• Defense: Governments may buy secure orbital processing, but the DIA's warning about counterspace threats shows that such assets would be targets.

Key Stakeholders

• Cowboy Space and Baiju Bhatt: vertically integrated aspirant.
• SpaceX: incumbent launch leader with a growing AI business.
• Stoke Space, Rocket Lab and other launch developers: potential suppliers or competitors in heavy and medium-lift capacity.
• FCC and ITU: regulators for spectrum and orbital filings at 20,000-satellite scale.
• Space Force and national security buyers: potential customers and threat assessors.
• Insurers and debris-removal firms: bearers of collision and failure risk.

Potential Disruptions and Wildcards

• Launch cost curve: If reusable heavy-lift reaches steady low cost per kilogram, the scenario tilts toward infrastructure. If vehicles keep slipping, it tilts toward niche.
• Radiation and reliability: Commercial AI accelerators are not designed for space. Failure rates could force costly shielding or frequent replacement.
• Thermal limits: Radiator mass scales with power, and large compute loads need large radiators. This is a physical constraint, not a business one.
• Regulation: Orbital filings at this scale could face scrutiny on debris, astronomy interference and spectrum. The FCC's ongoing licensing reform could speed or complicate approvals.
• Terrestrial alternatives: Cheaper power generation or nuclear build-outs could erode the case for space.
• Conflict and counterspace: Directed energy or kinetic threats raise insurance and hardening costs.
• Capital markets: Funding of a rocket plus 20,000 satellites is enormous. A capital drought would delay milestones.

Future Outlook: Milestones to Watch

The most useful leading indicators are not announcements but flight rates, satellite production rates and the first disclosed prices for orbital compute. If those appear and are credible by the early 2030s, the infrastructure case gains weight. If not, the sector's center of gravity will remain terrestrial, with orbital processing serving specialized needs.

Key Takeaways

• Cowboy Space unveiled Big Iron, targeting a late 2028 debut to launch a 20,000-satellite AI data center constellation.
• Orbital compute depends on launch cost and mass per watt, and new vehicle schedules often slip, so a 2029-2030 first flight is a sensible planning assumption.
• By 2035, a niche outcome (hundreds of satellites serving space-native and secure workloads) is the conservative baseline, and an infrastructure outcome (thousands) requires proven reusable heavy-lift.
• Radiation, thermal limits, regulation and orbital congestion (about 19,440 satellites today) are the main wildcards.
• Track flight rates, production rates and disclosed pricing rather than announcements.