SpaceX is engineered as a three-engine flywheel where Connectivity throws off cash, Space builds the next-generation capability, and AI creates the demand pull. Every revenue stream, every deal, every capital decision serves the same compounding loop. Read at this resolution, the business model is one of the most carefully designed in modern industry.
What Business Is SpaceX In?
For most of the company’s history the answer was “space access.” For the last five years it has been “satellite internet with a rocket business attached.” Neither frame survives contact with the current architecture.
SpaceX runs three reporting segments — Space, Connectivity, AI — but those labels misread the structure. They are not three businesses sharing a balance sheet. They are three engines in a single machine, each performing a distinct economic function, each gating the next.
Connectivity is the cash spine. It throws off the operating income that funds everything else. Recurring revenue, low marginal cost per subscriber, expanding geographic coverage.
Space is the capability development platform. It runs near breakeven by design. Its purpose is to build the next-generation deployment vehicle — Starship — that the other engines need.
AI is the demand pull. It is the heaviest consumer of capital. Its value to the company is not its current revenue line but its existence as an internal customer with effectively unlimited demand for orbital compute.
This is the operating reality. Read every revenue stream, every cost line, and every deal through this lens. None of them stand alone. All of them serve the loop.
The Revenue Stack, Engine by Engine
Each engine carries multiple revenue streams. The composition tells you where the business model is going.
Connectivity — about $11 billion last year, growing 50%
Five distinct revenue lines sit underneath the Starlink brand, each with different unit economics:
Consumer Broadband. Around ten million subscribers across 164 countries. Monthly subscription plus hardware (Starlink Kit) revenue. The bulk of the segment today. Median 225 Mbps peak-hour speeds, declining unit cost per subscriber as the constellation matures.
Enterprise Solutions. Aviation, maritime, agriculture, energy, hospitality, retail, telecom backhaul. Multi-site contracts with significantly higher ARPU than consumer and longer commitment periods. The fastest-growing line and the one with the strongest pricing power.
Government Solutions. Public services, humanitarian deployments, disaster response. Lower margin than enterprise but operationally important — proves the platform in adverse conditions.
Starlink Mobile. About 7.4 million monthly unique devices across 30 countries, served via partnerships with roughly 30 mobile network operators. SMS, voice over-the-top, light data today; broadband and IoT after V2 Mobile satellites deploy on Starship in 2027. Revenue currently flows largely through MNO partners, which the EchoStar spectrum deal is designed to change.
Starshield. A dedicated secure satellite network for US government customers and national security applications. Separately engineered from commercial Starlink, sold on cost-plus contract structures, very high gross margin and low elasticity. Materially smaller than commercial today but strategically significant.
What’s notable about this stack is how it climbs. Consumer Broadband is the volume layer. Enterprise is the margin layer. Government and Starshield are the regulatory and strategic-relationship layers. Each tier monetises the same underlying constellation at different price points and margin structures.
Space — about $4 billion last year, but the wrong number to focus on
Three revenue lines sit underneath Space, and all three are subordinated to the R&D investment running through the segment:
Third-party launch services. Commercial satellite operators, international space agencies, US government NSSL missions. Last year SpaceX flew 11 of 12 NSSL medium and heavy lift missions and all five US crew and cargo missions to the ISS. Roughly 650 cumulative orbital launches with >99% mission success.
Dragon crew and cargo missions. Sole US capability to fly humans to and from the International Space Station. 78 crewmembers flown across 20 countries since 2020. Long-term NASA contracts. Strategically irreplaceable, commercially modest.
Internal Starlink launches. Not reported separately — eliminated in consolidation — but they represent the majority of actual launch cadence. This is where Space’s real economic value lives.
Roughly $3 billion of the segment’s revenue was reinvested into Starship R&D last year. This engine is not run for cash; it is run as a capability buildout funded by the cash spine. The unit economics of Falcon 9 are excellent at this point — booster reflight up to 34 times, mature operations — but the segment-level losses reflect a deliberate choice to convert every dollar of operating margin into the next-generation vehicle.
AI — about $3 billion last year, but spending close to $13 billion of capex
This is the newest engine, the most expensive engine, and the one with the most fragmented revenue stack:
Consumer subscriptions. SuperGrok Lite, SuperGrok, SuperGrok Heavy — three tiers of model access with escalating usage limits. X Premium+ as a bundled premium subscription. Direct-to-consumer monthly revenue.
Enterprise subscriptions. Grok Business (organisational AI deployment), xAI Gov (government applications). Higher ARPU, contractual commitments, integration revenue.
Advertising revenue from X. The legacy revenue model from the X platform — display advertising, promoted posts, premium publisher arrangements. Materially smaller than at acquisition due to advertiser exodus, but the platform’s role is increasingly as distribution and data substrate for Grok rather than as an ad business in its own right.
Compute services revenue. This is the newest and most interesting line. SpaceX is selling raw compute capacity on COLOSSUS and COLOSSUS II to third-party customers — the Anthropic deal being the public example, but the filing explicitly says more such contracts are expected.
The composition matters. The first three lines are application-layer revenue subject to model competition. The fourth — compute services — is infrastructure — as explored in the economics of AI compute infrastructure — -layer revenue subject to capacity supply. The economic gravity of the AI segment is shifting from application revenue toward infrastructure revenue as the compute buildout scales. This is the same direction Amazon’s AI business is moving, and for the same reason: infrastructure margin compounds while application margin is fought over.
The Cost Structure and Capital Allocation
Total capital expenditure across the company ran about $24.7 billion last year, allocated as follows:
AI segment: $12.7 billion (51%). Data centre construction, GPU procurement, power infrastructure, satellite-compute R&D, model training compute.
Space: $3.8 billion (15%). Starship development and infrastructure, launch facilities, Raptor engines, Starbase buildout.
Remaining 17% distributed across shared infrastructure, integration costs, working capital.
Two observations stand out.
The first is that AI capex now exceeds Connectivity and Space capex combined. This is not the cost structure of a launch company with side projects. It is the cost structure of a compute infrastructure company with launch and satellite operations attached.
The second is that R&D is structurally embedded inside the segment cost lines rather than reported separately at the corporate level. Roughly $3 billion of Space segment cost was Starship R&D last year — not capex, but operating expense pushed through the segment. This is a deliberate accounting choice that depresses Space segment profitability while keeping the company’s overall R&D investment composable inside the engine that incurs it. The cost structure is engineered to make the segments tell the strategic story.
The Vertical Integration Map
The defining structural feature of SpaceX is the depth of its vertical integration. The list of things SpaceX makes itself is unusually long:
AI training infrastructure (COLOSSUS, COLOSSUS II)
Power infrastructure adjacent to data centres
Frontier AI models (Grok)
Distribution platform (X)
Consumer applications (Grok app, SuperGrok tiers)
Sourced externally:
GPU chips (NVIDIA, with custom configurations)
Foundry services for custom silicon (TSMC primarily)
Some specialised components (radiation-hardened parts, certain sensors)
Cloud services for non-critical workloads
Legal, tax, audit, regulatory advisory
The pattern is consistent — every supplier dependency that constrains capability or extracts margin is targeted for internalisation as soon as scale permits.
The Customer Stack
Four distinct customer types pay SpaceX directly:
Consumers. Starlink residential subscribers, SuperGrok subscribers, X Premium+ subscribers, Grok app users. High volume, low ARPU, monthly subscriptionmodel — as explored in the shift from SaaS to agentic service models — . Approximately 10 million Starlink subscribers plus 117 million MAUs using Grok features.
Governments. Starshield for US national security, Starlink Government for public services and disaster response, NASA crew/cargo contracts, NSSL launch contracts, xAI Gov for government AI applications.
Internal customers. Starlink as a launch customer for SpaceX itself, the AI segment as a compute deployment customer for Space, the X platform as a distribution customer for Grok. The largest “customer” by transaction volume and the engine that justifies the full capacity buildout.
The Five Deals as Business Model Instruments
The xAI Merger — the founding move. Stock-for-stock merger, completed February 2026. Integrates AI development, distribution platform, and substrate provider into one corporate entity.
The Anthropic Compute Services Agreement — dual monetisation. Monthly compute services contract worth approximately $1.25 billion through May 2029.
The Cursor Option — acquisition disguised as partnership. Compute services agreement plus unilateral option to acquire Cursor at $60 billion implied equity value.
The EchoStar Spectrum Transaction — asset swap diplomacy. Acquisition of AWS-3, AWS-4, and H-Block spectrum licenses.
The Terafab Initiative — internalising silicon. Joint venture framework with Tesla and Intel. Target: 1 terawatt of compute hardware production annually.
The pattern across all five is the same: option-based vertical integration paid for with future equity value rather than current cash.
The Flywheel and Its Feedback Loops
Five loops running simultaneously:
The cost-down loop. Each Starlink subscriber added reduces unit cost per subscriber.
The launch cadence loop. Each additional internal payload increases the launch cadence, amortising fixed cost over more flights.
The compute monetisation loop. Each gigawatt of compute built for Grok creates surplus capacity for sale to third parties.
The data feedback loop. Each Grok interaction generates training data. X provides 350 million daily posts.
The substrate compounding loop. Each successful Starship flight reduces the cost of orbital deployment.
The five loops feed each other. The cost-down loop pays for the launch cadence loop. The launch cadence loop enables the substrate compounding loop. Cut any single loop and the others slow; sustain all five and the system compounds.
The Moats
Five distinct moat types:
Sequential capability moat. 24 years to build the stack. Capital alone cannot compress that timeline.
Vertical integration moat. Each owned layer captures margin competitors pay to suppliers.
Network effect moat in Starlink. The constellation is now large enough that no plausible competitor can match coverage.
Regulatory and spectrum moat. Spectrum licenses, FCC authorisations, international landing rights across 164 countries.
Founder concentration moat. Dual-class structure with founder control allows decisions quarterly markets would not tolerate.
The fragility is concentrated in one specific place: Starship as the gating dependency.
Key Takeaways and Mental Models
The Three-Engine Flywheel. Connectivity is cash spine, Space is capability, AI is demand pull.
Engineered Cost Allocation. Heavy on R&D and capex, light on SG&A and marketing.
Internal Customer Demand. SpaceX is its own largest customer across all three segments.
Option-Based Vertical Integration. Use structured options and deferred consideration to pre-commit to every layer.
Dual Monetisation of Substrate. Sell to competitors what you also use.
The Five Compounding Loops. Most companies have one flywheel; SpaceX runs five concurrently.
The Single Point of Failure. Starship gates everything. The same architecture that makes the upside compound makes the downside concentrate.
The honest summary: SpaceX is one of the most carefully engineered business models in modern industry. The corporate architecture has been designed with the same first-principles discipline as the rockets. Either the system runs as designed and produces a generational compounder, or one of the gating dependencies slips and the loop breaks.
Go Deeper with Free Tools
Explore where SpaceX and its competitors sit on the full landscape: Map of AI →
Run a free Defensibility Audit on any company: Try it →
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
