Thesis Summary

SpaceX is the only company simultaneously operating a dominant orbital launch franchise, a fast-growing global broadband business, and a frontier AI compute platform — each one enabling the next. The current financial picture is deliberately distorted by an investment cycle of historic magnitude: AI segment capex of $12.7B in FY2025 and $7.7B in Q1 2026 alone are obscuring a Connectivity segment that generated $7.2B in Segment Adjusted EBITDA in FY2025, growing 86% year-over-year. The near-term catalyst is Starship: its commercial payload delivery, targeted for the second half of 2026, unlocks V3 satellite deployment that could drive a step-change in Starlink network capacity and cost per subscriber, while beginning the path to orbital AI compute.

Key thesis points:

• Starship commercialization is the critical near-term event. Payload delivery in H2 2026 will trigger a cascade: V3 Starlink satellites (1 Tbps per satellite, 20x the downlink capacity per launch vs. Falcon 9), V2 Mobile satellites for Starlink Mobile expansion, and the beginning of orbital AI compute deployment.
• Starlink's cash generation engine is growing faster than the headline numbers suggest. Connectivity Segment Adjusted EBITDA grew 86% to $7.2B in FY2025 despite ARPU declining 11%, because subscriber volume growth of 100% dominated. That dynamic has continued into Q1 2026 with 105% YoY subscriber growth to 10.3M.
• The AI compute build is the largest bet in corporate history. SpaceX is spending over $30B annualized on AI capex to become the lowest-cost provider of compute tokens, with a unique long-term pathway to orbital compute powered by unlimited solar energy. The $1.25B/month Anthropic agreement validates the market for its compute infrastructure.
• The flywheel is real and reinforcing. Launch economics fund Starlink deployment. Starlink cash funds AI compute. AI compute infrastructure funds Grok model quality. Grok quality drives subscriptions and enterprise adoption. Each layer compounds the next.

Starship Commercialization: The Near-Term Catalyst

Starship has completed 12 flight tests as of May 2026, with the 12th debuting the next-generation vehicle and booster. SpaceX expects payload delivery to orbit in H2 2026.

The financial implications of Starship commercialization are non-linear:

• V3 Starlink satellites require Starship — Falcon 9 cannot deploy them. V3 satellites offer 1 Tbps downlink each, vs. approximately 50 Gbps for V2 Mini, and Starship can deploy up to 60 per launch vs. approximately 22 V2 Mini per Falcon 9 launch. A single Starship launch represents a 20x increase in downlink capacity deployed per flight.
• Cost per subscriber served will inflect. SpaceX has already reduced satellite manufacturing cost per Gbps of downlink by approximately 3x from V1 to V2 Mini. V3 targets a total 9x reduction vs. V1. Lower cost per bit is the mechanism to sustain profitability as ARPU continues declining internationally.
• Starlink Mobile Gen2 depends on V2 Mobile satellites, also requiring Starship. SpaceX has 30 MNO partnerships covering approximately 1.9 billion people; Starlink Mobile had approximately 7.4M monthly unique devices as of Q1 2026. V2 Mobile satellites are expected to begin deployment in 2027, enabling broadband data and 5G connectivity to unmodified phones.
• Orbital AI compute — targeted for 2028 initial deployment — is only economically viable with full Starship reusability. SpaceX argues this eventually solves the terrestrial power grid bottleneck constraining AI scaling.

Starship is absorbing $3.0B in R&D annually (FY2025), holding the Space Segment Adjusted EBITDA to $653M vs. $1.2B in FY2024. Once Starship is commercialized, Starship costs shift from R&D expense to capitalized cost of revenue, and the Space segment's reported profitability inflects upward. This is a straightforward but timing-uncertain earnings recovery.

Starlink: The Cash Engine Is Stronger Than It Looks

The headline ARPU trend obscures the operating leverage of the Connectivity business:

• FY2025 Connectivity revenue grew 49.8% to $11.4B while Connectivity income from operations more than doubled to $4.4B (120% growth).
• Segment Adjusted EBITDA grew 86.2% to $7.2B in FY2025, and $2.1B in Q1 2026 alone.
• Subscribers grew 105% YoY to 10.3M as of March 31, 2026; monthly ARPU was $66 in Q1 2026.

The ARPU decline is a deliberate strategy, not a competitive or pricing failure. SpaceX is expanding into lower-income international markets where it faces limited competition from terrestrial broadband, trading per-user revenue for volume. The bet is that Starlink's satellite manufacturing and launch cost structure continues to improve as constellation scale increases — a bet that V3 satellites should validate.

Enterprise and government revenue is also a growing and under-appreciated contributor. Enterprise and government Connectivity revenue grew $1.4B in FY2025, driven by aviation, maritime, land mobility, and Starlink Mobile. The aviation market alone (approximately 23,900 commercial aircraft and 24,500 private aircraft globally) is early in adoption; enterprise customers have shown near-zero voluntary churn since 2023.

Starshield (the national security segment) remains opaque externally but represents a high-value, long-duration revenue stream from the U.S. government. The $19.6B EchoStar spectrum acquisition, expected to close November 2027, sets up a step-change in Starlink Mobile's addressable spectrum and 5G capability.

The AI Compute Build: A Call Option on the Largest Market in History

The AI segment is deeply loss-making: $(6.4B) operating loss in FY2025 and $(2.5B) in Q1 2026. But the strategic logic deserves separate evaluation from the current P&L.

SpaceX's argument is that cost per AI token will be the primary competitive differentiator in AI, and that cost per token is a function of compute hardware cost, data center construction cost, and energy cost. SpaceX argues it has structural advantages in all three:

• Data center construction speed and cost. SpaceX brought the first COLOSSUS cluster online in 122 days, and the first COLOSSUS II cluster in 91 days, at construction costs considerably below industry benchmarks. The industry benchmark for a 100 MW greenfield data center is approximately two years. COLOSSUS and COLOSSUS II collectively provide approximately 1.0 GW of nameplate compute draw as of Q1 2026.
• Chip access and vertical integration. Terafab (a collaboration with Tesla and Intel) targets producing 1 terawatt of compute hardware annually, reducing dependence on NVIDIA and optimizing chips for the space environment.
• Orbital compute and effectively zero marginal energy cost. Space-based solar arrays can generate more than 5x the energy per unit area of terrestrial solar due to continuous illumination. If orbital AI compute is achieved, SpaceX argues the marginal energy cost approaches zero — a structural cost advantage vs. any terrestrial competitor facing grid interconnect constraints.

The Anthropic agreement ($1.25B/month through May 2029, terminable after 90 days' notice) provides immediate monetization of stranded compute capacity and serves as a proof-of-concept that the market will pay for large-scale GPU clusters at premium rates.

Grok had 117M MAUs using AI features and 6.3M paid subscribers as of Q1 2026. The monetization base is thin relative to the capital deployed, but the X platform's 550M MAUs provide distribution at scale that most AI startups cannot access.

The Structural Flywheel: Why This Converges

SpaceX's capital allocation history is instructive: the Space segment became sustainably Segment Adjusted EBITDA positive in 2018, the Connectivity segment in 2023. Both segments followed an investment-then-harvest pattern. The AI segment is in the investment phase today.

The flywheel logic:

1. Falcon 9 launch economics (165 launches in FY2025, 80%+ of global mass to orbit) fund internal Starlink satellite deployment at zero external launch cost.
2. Starlink's $7.2B Segment Adjusted EBITDA funds AI compute build-out and Starship R&D simultaneously.
3. Grok/AI compute infrastructure, once monetized, funds the next layer — orbital compute, Terafab, and eventually lunar manufacturing.

SpaceX has executed this pattern twice. The question for AI is whether the time horizon to positive AI Segment Adjusted EBITDA is acceptable — management has indicated it expects a multi-year investment horizon. But unlike a startup, SpaceX has $7B+ of annual operating cash flow from its existing businesses to fund the wait.

Near-Term Earnings Dynamics to Watch

Several near-term developments could materially move segment-level results:

• Starship payload delivery in H2 2026 shifts Starship costs from R&D expense to capitalized satellite costs, improving Space Segment Adjusted EBITDA and beginning V3 constellation deployment.
• Anthropic compute revenue ramp: $1.25B/month in committed revenue from Anthropic begins in Q2 2026 (at a ramp rate), flowing into AI segment revenue. At full run rate, this represents approximately $15B in annual revenue from a single customer. Even with high infrastructure costs, this materially changes the AI segment's revenue-to-investment ratio.
• X advertising platform overhaul: Q1 2026 advertising revenue was $343M, down 23% YoY, partly due to a platform rebuild. Management attributed some of the decline to a temporary disruption during the rebuild. If the new X Ads Manager (rolled out in April 2026) recovers advertiser spend, it could be a revenue tailwind in H2 2026.
• EchoStar spectrum closing (November 2027): Adds 65 MHz of U.S. spectrum and global Mobile Satellite Service licenses, enabling 5G-like direct connectivity to unmodified phones and a generational upgrade to Starlink Mobile.