Morgan Stanley initiates coverage of SpaceX: Orbital AI compute is the key incremental growth driver; rated Overweight with a $300 price target

Overview

This is a Morgan Stanley company research and initiation-of-coverage report on SpaceX, focused on the engineering feasibility, capital expenditure, cost curve, and long-term valuation implications of orbital AI compute. The report extends the views expressed in the July 7 report "SpaceX: AI’s Final Frontier; Initiate at Overweight, PT $300" and primarily addresses investor questions about thermal management for space-based AI infrastructure. The core conclusion is that while convection cooling using air or water is not possible in a vacuum, satellites and spacecraft can manage heat through radiative cooling, fluid loops, and large radiators; therefore, thermal dissipation is not a fundamental physical impediment to orbital compute.

Core views

The report's core views include: First, orbital compute can leverage the abundant space, solar power supply, and parallel deployment capabilities of low Earth orbit, providing a differentiated scaling path as AI infrastructure is constrained by terrestrial power grids, land, water resources, and permitting. Second, SpaceX could achieve gigawatt-scale orbital compute around 2030, with orbital compute becoming the majority of deployed compute after 2032. Third, orbital compute has higher upfront capex than terrestrial compute, but because it lacks significant power, cooling, land, maintenance, and site operating costs, its all-in costs could become superior to current industry terrestrial costs beginning in 2031. Fourth, if SpaceX can use Starship to reduce launch costs, achieve hardware cost reductions through scaled satellite manufacturing, and improve efficiency through specialized chips and solar power systems, long-term AI EBIT margins could reach more than 50%.

Analysis framework

The report uses a bottom-up engineering and financial modeling approach. It breaks orbital compute capex into launch, satellite hardware, and compute, and further estimates radiator area, solar-array area, compute payload, satellite mass, cost per watt, and depreciation. The engineering assumptions cover three generations of orbital AI satellites: AI1 is based on AI satellite designs recently disclosed by Elon Musk and SpaceX, while AI2 and AI3 are assumed to be larger, have higher power density, and achieve lower unit mass costs. On the financial side, the report compares capex, depreciation, cash costs, revenue per watt, and EBIT margins for orbital and terrestrial compute.

Methodology notes

• equity_research_modelMorgan Stanley ModelWare

  Morgan Stanley valuation and financial forecasting framework

  The report states that, unless otherwise noted, all metrics are based on the Morgan Stanley ModelWare framework, incorporating Morgan Stanley Research estimates and Refinitiv Estimates consensus data.

• engineering_physicsStefan-Boltzmann Law

  Radiator-area estimation

  The report uses the Stefan-Boltzmann law to estimate the surface area required for space radiators at a given temperature, emissivity, and effective background temperature, emphasizing that radiative capacity increases with the fourth power of temperature.

• bottom_up_capexOrbital Compute Capex Build

  Orbital compute capital expenditure breakdown

  The report breaks orbital compute capex into launch costs, satellite hardware costs, and compute costs, modeling Starship payload capacity, satellite counts, GPU cost premiums, a five-year useful life, and straight-line depreciation item by item.

• cost_comparisonAll-in Cost per Watt

  Cost comparison between orbital and terrestrial compute

  The report argues that looking only at capex understates the advantages of orbital compute, so it compares capex depreciation with differences in power, cooling, land, personnel, maintenance, and site operating opex for terrestrial compute.

Asset mapping & comparison

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

• SpaceX (SPCX.O)

  Core covered company and investment asset

  Strengths

  Possesses potential integrated advantages spanning Starship, Starlink satellite manufacturing, a path to lower launch costs, space-based solar power, and optical communications, enabling orbital compute to progress from concept to scaled deployment.

  Weaknesses

  Orbital AI compute remains at an early stage; technical designs may undergo multiple iterations, upfront capex and GPU cost premiums are high, and in-orbit maintenance capabilities are limited.

  Comparison

  Compared with conventional terrestrial data centers, orbital compute has higher upfront capex but could have long-term advantages in power, cooling, land, maintenance, and deployment speed.

  Risks

  Slower-than-expected launch cadence, insufficient satellite reliability, radiation damage, orbital debris, constrained chip supply, regulatory or security issues, and customer demand or pricing falling short of expectations.

• Orbital Compute / Starmind

  Long-term AI infrastructure growth engine for SpaceX

  Strengths

  Can leverage low Earth orbit space, near-continuous sunlight, repeatable infrastructure deployment, and lower long-term opex, making it suitable for large-scale AI inference workloads.

  Weaknesses

  Requires solving systems-engineering issues involving radiator mass, solar-array area, radiation-hardened compute hardware, remote redundant maintenance, and in-orbit network connectivity.

  Comparison

  The report believes orbital compute will not immediately be cheaper than terrestrial compute on capex, but will gradually demonstrate advantages in all-in cost and scaling speed.

  Risks

  Overly optimistic engineering assumptions, slower-than-expected cost declines, useful life below five years, and compute yield or degradation rates exceeding model assumptions.

• Terrestrial AI Data Centers

  Primary comparison benchmark for orbital compute

  Strengths

  Mature technology, convenient maintenance, and well-established supply chains and operating models; suitable for model training and workloads requiring low-latency terrestrial access.

  Weaknesses

  Constrained by grid access, land, water resources, cooling, personnel, maintenance, and permitting timelines, which may limit expansion speed.

  Comparison

  The report expects terrestrial compute to be used more for model training over the long term, while most inference workloads could migrate to orbit.

  Risks

  If terrestrial data-center costs decline faster or power constraints ease, the relative advantage of orbital compute could weaken.

• Starship

  Launch and cost foundation for scaling orbital compute

  Strengths

  Greater effective payload to orbit and lower launch cost per kilogram can reduce orbital compute capex and support larger AI satellite iterations.

  Weaknesses

  The model depends on Starship V3/V4/V5 continuing to improve payload capacity and launch economics.

  Comparison

  The report assumes Starship V3/V4/V5 achieve effective payloads to orbit of 80 tonnes, 120 tonnes, and 160 tonnes, respectively, corresponding to the deployment of 37, 30, and 26 AI satellites of different generations per launch.

  Risks

  Starship technology, regulatory approval, launch frequency, or reusability costs falling short of expectations would directly affect the orbital compute deployment curve.

Key data

• Stock ratingOverweightMorgan Stanley assigns SpaceX an Overweight rating.
• Price target$300.00The report lists a price target of $300.00.
• Closing price$119.85Closing price on July 20, 2026.
• Orbital compute deployment start160MW in 2028The model assumes SpaceX begins deploying orbital compute in 2028.
• Orbital compute in 2030Approximately 2.7GWThe report expects gigawatt-scale orbital compute by 2030.
• Orbital compute share in 203221.2GW, 58% of total computeThe report expects orbital compute to become the majority of deployed compute in 2032.
• Orbital compute in 2040364GW, 96% of total computeThe long-term projection assumes that most AI inference compute migrates to orbit.
• Timing of orbital compute cost advantage2031The report estimates that the all-in operating cost of newly added orbital compute will fall below current industry terrestrial compute costs in 2031.
• Annualized orbital compute cost in 2031Approximately $6.5/W/yearDerived from $32.4/W capex amortized over a five-year life.
• Industry terrestrial Blackwell cost referenceApproximately $6.8/W/yearBased on a recent estimate from Morgan Stanley's internet team.
• Cash cost in 2035Below $9/WThe report expects total cash costs to decline steadily from approximately $35-40/W currently.
• Cash cost in 2040$3.7/WLong-term cost reductions support AI EBIT margins.
• Long-term AI EBIT marginMore than 50%The report expects AI business EBIT margins to reach approximately 50% or higher in the 2030s.
• AI1 satellite peak power150kWAssumption for the first-generation orbital AI satellite, with compute power of approximately 120kW.
• AI1 satellite massApproximately 2.1 tonnesCorresponds to the initial Starship deployment assumptions.
• AI1 radiators110 square meters per side, 767kg total massAssumes two-sided radiation, approximately 0.91 emissivity, and an operating temperature of approximately 90°C.
• AI1 solar array605 square meters, 1,028kg total massThe report assumes 25% cell efficiency, 95% sunlight exposure, 95% packaging ratio, and 95% power-distribution efficiency.
• Orbital GPU cost premiumMore than 5x terrestrial costs in 2028, approximately 2.3x in 2030, and 5-10% higher from 2032 onwardReflects early requirements for high-temperature tolerance, fault tolerance, and radiation resistance, as well as small-batch production costs.

Impact & implications

For investment implications, the report expands SpaceX's value narrative from launch and connectivity businesses to AI infrastructure. If orbital compute develops as modeled, SpaceX could become one of the most scalable AI compute suppliers of the 2030s and receive a valuation premium through lower long-term cash costs and faster deployment. The report also cautions that investors should not treat the thermal-management debate as the sole key variable; the factors requiring continued validation are Starship launch cadence, satellite manufacturing scale, chip supply, orbital reliability, radiation damage, debris risk, and customers' willingness to pay a premium for scarce compute.

Risks

• Upside risk: Starship launch costs decline faster than expected, satellite hardware and GPU costs improve faster than modeled, and orbital compute customers are willing to pay a higher premium for scarce compute.
• Upside risk: SpaceX achieves high-power-density AI satellites faster, while radiator and solar-array mass are lower than assumed, allowing orbital compute deployment to reach gigawatt scale earlier.
• Downside risk: Although thermal management is solvable, radiator size, mass, coolant, long-term reliability, or temperature control fail to meet the requirements of high-density AI computing.
• Downside risk: Radiation damage, orbital debris, lack of human maintenance, physical security, and remote redundancy requirements result in useful life or availability below the five-year assumption.
• Downside risk: Starship launch cadence, payload capacity, unit launch cost reductions, or satellite-manufacturing scale fall short of expectations.
• Downside risk: Orbital GPUs require greater high-temperature tolerance, fault tolerance, and radiation resistance, causing the cost premium to persist longer than modeled.
• Downside risk: Terrestrial AI data-center costs decline or grid bottlenecks ease, narrowing the relative advantage of orbital compute.
• Downside risk: Regulation, spectrum, orbital resources, international security, and space-debris governance may constrain deployment scale.

What to watch

• Parameters of the AI1, AI2, and AI3 satellites disclosed by SpaceX, including compute power, satellite mass, radiator area, and solar-array area.
• Effective payload to orbit, launch frequency, reusability performance, and internal cost per kilogram for Starship V3/V4/V5.
• Actual cost, radiation resistance, high-temperature design, and supply-chain expansion progress for orbital AI chips.
• Whether the first orbital compute deployment can begin in 2028 and whether it approaches the 2.7GW model assumption by 2030.
• Contract pricing, scale, delivery timing, and service-availability terms between SpaceX and neocloud or other AI customers.
• Measured performance of radiator materials, coolants, emissivity, effective LEO background temperature, and two-sided radiation designs.
• Low Earth orbit debris, collision avoidance, regulatory approvals, and physical-security incidents.
• Changes in terrestrial AI data-center power access, cooling, land, equipment supply, and industry-average Blackwell costs.