Morgan Stanley has initiated coverage of SpaceX with a $300 price target, arguing that the company's vertically integrated AI infrastructure business, combining terrestrial and orbital compute capacity, could make it one of the most powerful infrastructure platforms of the artificial intelligence era. The brokerage's analysis reveals a dramatic shift in how Wall Street is valuing Elon Musk's space company: less as a rocket manufacturer and more as a hybrid AI infrastructure provider that spans from Earth to orbit.

The investment thesis rests on a counterintuitive premise. While SpaceX's Starship rocket and Starlink satellite network grab headlines, Morgan Stanley argues the real value driver is SpaceXAI's compute infrastructure, particularly its COLOSSUS and COLOSSUS II clusters. These facilities collectively provide around 1.0 gigawatt (GW) of computing power, with COLOSSUS brought online in just 122 days and COLOSSUS II in 91 days. The speed matters because in the AI infrastructure race, time-to-power is as critical as cost-per-watt.

What Makes Orbital Compute SpaceX's "Secret Weapon"?

The most audacious part of Morgan Stanley's model involves orbital compute, a concept that sounds like science fiction but is grounded in real engineering. The bank estimates that SpaceX will deploy AI satellites beginning in 2028, scaling to 2.7 gigawatts by 2030, 21 gigawatts by 2032, and eventually 364 gigawatts by 2040. The critical milestone: orbital compute reaching cost parity with current terrestrial data centers by 2031.

To understand why this matters, consider the economics. Today, building a terrestrial data center costs roughly $6.8 per watt per year for high-end GPU clusters. Morgan Stanley's model shows orbital compute capex falling from $166 per watt in 2028 to just $9 per watt by 2040, translating to approximately $6.5 per watt per year on an annualized basis. That's cheaper than Earth-based alternatives, without the land constraints or cooling challenges that plague traditional data centers.

But cost parity alone may not be the primary advantage. Morgan Stanley emphasizes that "scalability and time-to-power is the bigger constraint versus pure cost." Recent deals in the "neocloud" market, where companies rent access to large GPU clusters, have commanded premium pricing precisely because high-end compute capacity is scarce and difficult to access quickly. If SpaceX becomes the most scalable AI infrastructure player available, it could charge a material premium even if orbital compute isn't the absolute cheapest option.

Morgan Stanley

How Does Starship Enable the Entire AI Infrastructure Vision?

Starship remains the foundational technology enabling SpaceX's long-term economics. Morgan Stanley assumes Starship becomes operational in the fourth quarter of 2026 and models launch costs per kilogram falling to roughly $500 by 2030, under $200 per kilogram by 2035, and below $150 per kilogram by 2040. For context, historical averages before SpaceX's Falcon 9 rocket were about $18,500 per kilogram.

The reusability profile is central to achieving those cost curves. SpaceX has already caught the Super Heavy booster three times and reused it twice. The next critical milestone is physical recovery of the Starship upper stage, which has already performed controlled ocean landings. Morgan Stanley expects Ship recovery "sometime before year-end, potentially as soon as Flight 14," but stresses that investors should focus less on the landing itself and more on the turnaround time required for reuse, as launch cadence will be the key indicator of cost progress.

By 2040, Morgan Stanley models roughly 6,000 Starship launches per year, implying about 16.5 launches per day across five pad complexes and 10 towers. Each Ship would be reused around 40 times and each booster about 130 times, supported by a fleet of roughly 170 Ships and 56 boosters. Critically, 75 percent to more than 90 percent of Starship launches are likely to be used internally between 2027 and 2040, meaning launch economics flow directly into Starlink V3 deployment, orbital compute, and other infrastructure programs.

Steps to Understanding SpaceX's Four Key Performance Metrics

• Revenue per Watt: Measures how much revenue SpaceX generates from each unit of computing power deployed, directly tied to pricing power in the AI infrastructure market and whether the company can command premium rates for orbital compute.
• Cost per Watt: Tracks the capital and operational expenses required to deploy compute capacity, including manufacturing, launch, and orbital deployment costs, with the goal of reaching cost parity with terrestrial data centers by 2031.
• Cost per Kilogram to Orbit: Reflects Starship's reusability and launch efficiency, with Morgan Stanley modeling costs falling from current levels to under $150 per kilogram by 2040, enabling affordable deployment of AI satellites and infrastructure.
• Starlink Subscribers and Connected Nodes: Measures adoption of Starlink broadband and the proliferation of Starlink-connected devices, including autonomous vehicles, drones, and robots, which will generate connectivity revenue and create demand for compute services.

How Does Starlink Fit Into the AI Infrastructure Story?

Starlink is positioned as the "connectivity layer for every data-transmitting device that needs reliable coverage beyond terrestrial networks." With Starship-enabled V3 broadband and Mobile Gen 2 satellites, Morgan Stanley estimates saleable capacity rising approximately 2.5 times in 2027 on just 35 Starship launches, nearly 19 times by 2030, and roughly 600 times by 2040. The V3 satellites offer more than 10 times the downlink capacity per satellite and over 20 times the capacity per launch versus current V2 Mini satellites on Falcon 9.

Connectivity revenue is modeled to climb from $11.4 billion in 2025 to $120.6 billion in 2030 and $687.7 billion in 2040. A significant portion of this growth comes from Starlink-connected robots and embodied AI devices. Morgan Stanley estimates a global robot base of about 2.2 billion by 2040, with Starlink penetration rates of 40 percent for autonomous vehicles, 33 percent for drones and eVTOLs, and 20 percent for other robots, contributing roughly 34 percent of 2040 connectivity revenue.

To reach broad adoption, the bank's model assumes global household broadband penetration peaking near 11.5 percent in 2040 and mobile penetration at around 6 percent. Consumer and enterprise pricing is expected to decline from roughly $92 per month in 2025 to $47 in 2030 and $36 in 2040 as international markets grow in the mix.

What Is the Enterprise AI Opportunity?

Morgan Stanley calls Enterprise AI "the largest opportunity" in its SpaceX model, breaking it into three layers: renting compute through neocloud agreements, selling managed AI infrastructure and tools, and delivering end-to-end enterprise applications. The bank forecasts AI revenue rising from about $22 billion in 2026 to $190 billion in 2030 and $2.6 trillion in 2040, with enterprise use cases accounting for the majority of the segment over time.

Near term, neocloud deals dominate the revenue picture. SpaceX's regulatory filing disclosed a major partnership with Anthropic, an AI safety company, and the company has signed additional agreements with other enterprises seeking access to large-scale compute capacity. These deals command premium pricing because high-end GPU clusters are scarce and difficult to scale quickly, giving SpaceX a competitive advantage if it can deliver reliable, rapidly deployable infrastructure.

The broader implication is that SpaceX's valuation increasingly depends on execution across multiple fronts: Starship's reusability, Starlink's adoption, and SpaceXAI's ability to deliver compute capacity faster and cheaper than competitors. Morgan Stanley's $300 price target assumes all three pillars succeed, with orbital compute becoming a meaningful driver of lower cost per watt by the early 2030s. If the company delivers on even a portion of this vision, the investment thesis suggests SpaceX could become one of the most valuable infrastructure platforms in the AI era.