SemiAnalysis’ Space Datacenter TCO Model

The SemiAnalysis AI Space Datacenter TCO Model is used to evaluate the economics, engineering constraints, and supply-demand dynamics of orbital compute, and to determine if and when deploying AI datacenters in space becomes cost-competitive with terrestrial datacenters. It can be used to build up the full cost stack of a space datacenter from first principles – from launch vehicle physics and thermal rejection limits to GPU-level cost of ownership – and to stress-test the terrestrial power, datacenter, and silicon constraints that would push compute off Earth in the first place.

Our model spans 2026 to 2050 on an annual basis, with dynamic scenario modeling driven by user-controlled assumptions.

Two headline scenarios are provided, both fully adjustable:

• Base Case – bullish chip and Earth datacenter capacity ramp
• Elon Musk Case – constrained terrestrial supply and aggressive launch/Terafab assumptions

We compute the Total Cost of Ownership for both space and terrestrial deployments and output on a $/hr/GPU, $/PFLOP-hr, and $/billion-token basis. Cost buckets covered:

• IT Capital Cost of Ownership (servers, networking, storage, software, burn-in)
• Datacenter Capital Cost of Ownership (including launch)
• Operating Cost of Ownership
• Bridged from TCO to Levelized Cost of Compute (LCOC) via radiation availability and GPU redundancy / SLA provisioning
• Adjustable WACC, useful-life, utilization, and PUE assumptions

Space system capital costs are built line-by-line from first principles, with every item adjustable for cost and technology scaling. Covered subsystems include:

• Solar array (cell technology, efficiency, areal density, specific power, BOL/EOL sizing)
• Radiator hardware (Stefan-Boltzmann heat rejection, operating temperature, droplet-radiator roadmap, specific power)
• Cold plates, pumps, heat exchanger and transport loop
• Battery and power management & distribution (PMAD)
• Bus hardware (ADCS, onboard computer, star trackers, reaction wheels, magnetorquers)
• Propulsion hardware, feed system, propellant and tank
• Shielding (radiation and MMOD)
• Communications hardware and ground segment
• Structure & integration, and Assembly, Integration & Test (AI&T)

Launch economics for Falcon 9 and Starship, including:

• Cost per launch, payload to LEO, and cost per kg
• Reuse maturity scenarios (early / maturing / mature)
• Launch-cost trajectories (slow, base, Elon) and their sensitivity to LCOC

We model the terrestrial supply and the silicon constraint as five layers of incremental power supply. This includes user-adjustable foundry and HBM wafer capacity additions and Terafab scenarios.

Demand, supply, and installed base, including:

• Unconstrained vs. silicon-constrained AI critical IT power demand
• Terrestrial datacenter global critical IT power capacity
• Implied space datacenter demand and cumulative install base, by scenario
• Cost crossover and parity-ratio outputs, plus full launch-cost / terrestrial-cost sensitivity tables

The model also includes orbital mechanics (LEO, SSO, dawn-dusk, eclipse and solar irradiance) and 3-D interactive CAD models of major space datacenter components.

These models are not included with the annual membership to the SemiAnalysis Newsletter. All models are separate institutional offerings.