SpaceX Orbital AI Computing: Thermal Dissipation Is Not the Core Bottleneck; Long-Term Focus Is on Scalability and the Cost Curve

Overview

This report provides Morgan Stanley's in-depth analysis of the SpaceX orbital AI computing theme and follows its initial coverage report dated 2026-07-07. The core view is that although space is close to 2.7K, the effective radiative background temperature in low Earth orbit is higher due to Earth's infrared radiation and albedo. Even so, radiative cooling based on the Stefan-Boltzmann Law can manage the thermal load of AI satellites within engineering design limits. Therefore, the central debate over orbital data centers should move beyond whether they can dissipate heat and instead focus on cost, reliability, launch cadence, chip supply, orbital safety, and the ability to deploy at scale.

Core views

The report is bullish on SpaceX's long-term orbital AI infrastructure potential. Morgan Stanley believes SpaceX could use Starship, its Starlink satellite manufacturing experience, space-based solar power, and optical networks to develop orbital computing into one of the most scalable sources of AI computing in the 2030s. The report expects the first orbital computing deployment in 2028, GW-scale capacity by 2030, orbital computing to represent the majority of the company's deployed computing capacity by 2032, and its share to rise to 96% by 2040. In terms of cost, the report believes orbital computing will have higher initial capex than terrestrial computing, but because it avoids substantial power, cooling, land, maintenance, and site operating costs, it could gain a cost advantage over current industry terrestrial computing on an all-in basis by 2031 and support long-term EBIT margins above 50%.

Analysis framework

The report uses a bottom-up engineering and financial modeling framework: it first estimates radiator area, solar arrays, satellite mass, launch mass, chip costs, and depreciation based on assumptions for three generations of orbital computing satellites, AI1, AI2, and AI3. It then aggregates launch, satellite hardware, and compute capex into cost per watt, and finally compares these figures with the capex and opex of terrestrial AI infrastructure. The thermal analysis uses the Stefan-Boltzmann Law, the launch analysis uses SpaceX's internal launch cost-per-kilogram curve, and the financial analysis uses the Morgan Stanley ModelWare framework.

Methodology notes

• valuation_modelMorgan Stanley ModelWare

  Sell-side modeling framework

  The report states that, unless otherwise specified, all metrics are based on the Morgan Stanley ModelWare framework.

• physics_engineeringStefan-Boltzmann Law

  Radiative cooling area estimation

  The report uses the Stefan-Boltzmann Law to estimate the radiator area required for a given thermal load, surface temperature, emissivity, and effective background temperature.

• bottom_up_modelOrbital Compute Capex Build

  Bottom-up breakdown of orbital computing capex

  The report divides orbital computing capex into Launch, Satellite Hardware, and Compute, and estimates each component using satellite-generation specifications, launch mass, GPU costs, and five-year depreciation.

• cost_comparisonAll-in cost parity analysis

  All-in cost comparison with terrestrial AI computing

  The report emphasizes that focusing only on capex understates the advantages of orbital computing, because its ongoing costs are primarily depreciation, whereas terrestrial computing also incurs power, cooling, land, maintenance, and operating costs.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• SpaceX (SPCX.O)

  Core covered company and investment target

  Strengths

  Starship launch capability, satellite manufacturing experience, accumulated Starlink technology, space-based solar power, and optical communications networks could provide advantages for an orbital AI computing platform.

  Weaknesses

  Orbital AI computing remains at an early design and deployment-assumption stage, with high initial capex and dedicated GPU costs and substantial engineering iteration uncertainty.

  Comparison

  The report believes orbital computing will have higher initial capex than terrestrial computing, but on an all-in basis it could outperform current industry terrestrial computing costs by 2031 because it avoids power, cooling, land, and maintenance opex.

  Risks

  Insufficient launch cadence, constrained chip supply, radiation damage, orbital debris, difficulty of remote maintenance, asset security, and weaker-than-expected commercial demand.

• Terrestrial AI data centers

  Primary comparison group

  Strengths

  Mature technology pathway, convenient maintenance, and clearer supply-chain and customer-use models.

  Weaknesses

  Subject to power, cooling, land, water, grid-connection, and permitting constraints, with relatively high ongoing opex.

  Comparison

  The report believes terrestrial computing is better suited to model training, while the majority of inference workloads could migrate to orbit over the long term.

  Risks

  If terrestrial infrastructure costs decline faster or grid-connection bottlenecks ease, the cost and time-to-deployment advantages of orbital computing could narrow.

Key data

• RatingOverweightMorgan Stanley's stock rating for SpaceX.
• Target Price$300.00The price target listed in the report.
• Closing Price$119.85Closing price on 2026-07-20.
• First-Year Orbital Computing Deployment160 MW in 2028The report assumes orbital computing deployment will begin in 2028.
• Orbital Computing Capacity in 2030Approximately 2.7 GWThe report expects orbital computing to reach GW scale in 2030.
• Orbital Computing Share in 203221.2 GW, 58% of total computing capacityThe report expects orbital computing to become the majority of deployed computing capacity by 2032.
• Orbital Computing Capacity in 2040364 GW, 96% of total computing capacityThe report's long-term orbital computing capacity assumption.
• Incremental Annualized Cost in 2031Approximately $6.5/W/yearThe report estimates this by dividing $32.4/W of capex by a five-year useful life and compares it with approximately $6.8/W/year for industry Blackwell terrestrial computing.
• Long-Term Cash Cost per WattBelow $9/W in 2035, approximately $3.7/W in 2040The report estimates that orbital computing will drive a continued decline in cash cost per watt.
• Long-Term EBIT MarginApproximately 50%+ in the 2030sThe report believes declining costs will support long-term AI business EBIT margins above 50%.
• AI1 Satellite Assumption150 kW peak power, 120 kW compute, approximately 2.1 tonsThe report bases its initial estimate on the AI1 satellite design disclosed by Elon Musk/SpaceX.
• AI1 Radiator Estimate110 m² per side, 767 kg total massBased on assumptions of approximately 90°C, 0.91 emissivity, a 250K LEO background temperature, and two-sided radiation.

Impact & implications

If the report's assumptions hold, SpaceX's investment narrative would expand from traditional launch services and satellite internet to an AI infrastructure platform. Orbital computing could not only alleviate power, land, water, permitting, and grid-connection constraints affecting terrestrial data centers, but could also command premium pricing during periods of scarce computing capacity. For investors, the key question is whether SpaceX can convert low launch costs, mass-produced satellites, space-based solar power, and optical networks into commercializable, maintainable, and scalable AI inference infrastructure.

Risks

• Orbital computing satellites face radiation damage and higher fault-tolerance requirements, which could increase chip and system costs.
• Orbital assets lack on-site maintenance capabilities and must rely on remote redundancy and replacement mechanisms.
• Starship's initial launch cadence may provide only single-digit MW-scale increments, affecting the speed of scaling.
• Orbital debris and congestion in low Earth orbit could increase operating risks.
• Space thermal management feasibility does not imply commercial feasibility; cost, reliability, demand, and time to power remain key constraints.
• Tightness in the AI chip supply chain could limit orbital computing deployment.
• If SpaceX passes more of the cost savings on to customers, declining revenue per watt could limit further margin expansion.

What to watch

• Further technical details on SpaceX's AI1 satellite design, particularly power, radiators, solar arrays, and mass metrics.
• Effective payload to orbit, launch frequency, and launch cost-per-kilogram curves for Starship V3/V4/V5.
• Thermal tolerance, radiation resistance, fault-tolerance design, and cost-reduction pace of orbital GPUs.
• Pricing, customer demand, and delivery progress for SpaceX neocloud transactions.
• Policy changes related to LEO orbital debris, regulation, and space asset security.
• Actual deployment capacity and availability of orbital computing in 2028-2030.
• Whether power-grid interconnection, cooling, land, and equipment supply bottlenecks at terrestrial AI data centers persist.