SpaceX IPO Filing Reveals Starlink Satellite Internet as Company's Core Business Bet

When SpaceX filed its S-1 registration with the SEC on May 19, 2026, the document answered a question that had lingered over the company for years: what is SpaceX, really? The answer, buried in 300-plus pages of financial disclosures, is unambiguous. SpaceX is a telecom company that happens to own rockets [1].

Starlink — the satellite internet service that began commercial operations in 2020 — generated $11.4 billion in revenue in 2025, accounting for 61% of SpaceX's $18.7 billion in total sales [2]. In Q1 2026, Starlink's share climbed to nearly two-thirds [1]. More critically, Starlink was the company's only profitable division, producing $4.42 billion in operating income while the launch business lost $657 million and the AI segment (absorbed from Musk's xAI acquisition) hemorrhaged $6.35 billion [2].

The S-1 marks the first time SpaceX has opened its books publicly. What those books show is a company whose future — and whose $750 billion-plus target valuation — rests almost entirely on the trajectory of a satellite internet service competing against fiber optics, 5G, and the laws of physics.

Source: SpaceX S-1 Filing / CNBC

Data as of May 21, 2026CSV

The Revenue Engine: Growth at the Cost of ARPU

Starlink's revenue growth has been steep. From $3.3 billion in 2022 to $3.9 billion in 2023, then $7.7 billion in 2024 and $11.4 billion in 2025, the business has roughly tripled in three years [3]. Subscriber counts tell a parallel story: 1 million in 2022, 2.3 million in 2023, 4.4 million in 2024, 8.9 million in 2025, and 10.3 million by March 2026 [2].

Source: SpaceX S-1 Filing

Data as of May 21, 2026CSV

But the growth has come at a price. Average revenue per user (ARPU) fell 18%, from $99 per month in 2023 to $81 per month in 2025 [4]. That decline reflects SpaceX's strategy of trading per-user economics for global volume, expanding into lower-income markets where pricing power is weaker. The company does not disclose subscriber churn rates, making it difficult to assess customer retention or compare directly to terrestrial broadband providers like Comcast, whose broadband ARPU runs roughly $60 per month but with significantly lower capital intensity [4].

EBITDA margins, however, have moved in the opposite direction — from 41% in 2023 to 50% in 2024 to 63% in 2025 — suggesting that Starlink's unit economics improve as the network scales [3].

Source: SpaceX S-1 Filing / Sacra

Data as of May 21, 2026CSV

Who's Paying: Segments and Government Contracts

The S-1 breaks Starlink's revenue into broad categories. For 2025, analysts estimate roughly $7.5 billion came from consumer services, $1.3 billion from hardware sales, and $3 billion from U.S. government contracts [5]. The government revenue figure is notable because it includes not just civilian broadband but military and intelligence applications routed through Starshield, SpaceX's classified satellite network.

The Pentagon holds a $537 million contract for Starlink services to Ukraine's military through 2027 [5]. The National Reconnaissance Office reportedly has a Starshield contract worth $1.8 billion [6]. Under the Space Force's Proliferated Low Earth Orbit (PLEO) program, SpaceX has captured 97% of awarded task orders [5].

Maritime connectivity, aviation partnerships, and enterprise sales represent a growing share of revenue, with longer contract terms and higher switching costs than residential subscribers. These segments generate substantially higher margins, though exact breakdowns remain unavailable in the S-1 [5].

The concentration of government revenue raises a question for potential investors: how much of Starlink's profitability depends on defense contracts that could shift with political winds, particularly given Musk's proximity to the current administration?

The Satellite Treadmill: Physics and Costs

SpaceX operates approximately 10,200 Starlink satellites — roughly 65% of all active satellites orbiting Earth [7]. The FCC has licensed SpaceX for significantly more, and in January 2026 approved an additional 7,500 satellites [8]. But the constellation carries a structural maintenance burden that the S-1 doesn't fully quantify.

Starlink satellites orbit at about 550 kilometers altitude. At that height, atmospheric drag limits operational lifespan to roughly five to seven years, with empirical data showing many satellites retiring before the five-year mark [9]. SpaceX shut down nearly 500 satellites in the first half of 2025 alone [9].

Manufacturing costs have dropped to an estimated $500,000 per satellite, down from roughly $1 million in earlier generations [9]. Each Falcon 9 launch costs approximately $67 million and can carry several dozen satellites [9]. One widely cited analysis estimates that maintaining a 12,000-satellite constellation at full strength requires 2,400 replacement satellites per year across 104 launches, implying annual maintenance costs of roughly $8.2 billion [10]. That figure approaches Starlink's current EBITDA of $7.2 billion, which would leave little margin for network expansion or shareholder returns.

SpaceX disputes these estimates implicitly — the company's vertical integration (building both the satellites and the rockets) means internal costs are lower than external price sheets suggest. But the S-1 acknowledges that "any failure or delay in the development of Starship at scale or in achieving the required launch cadence" would cascade through the entire Starlink business [11].

The Starship Dependency

This creates what may be the most underappreciated risk in the S-1. Starlink's next-generation satellites — larger, more capable, and essential for the direct-to-cell service SpaceX calls Starlink Mobile — require Starship to launch. Falcon 9 cannot carry them. SpaceX invested $930 million in Starship R&D in Q1 2026 alone, on top of $3 billion in 2025 [11].

Starship remains in the testing phase. The S-1 warns of potential "rapid unscheduled disassembly" — SpaceX's euphemism for explosions [12]. If Starship's development stalls, SpaceX cannot deploy next-generation Starlink satellites, cannot expand Starlink Mobile, and cannot build the orbital data centers the company has teased for 2028 [1].

For potential IPO investors, there's a structural wrinkle. The S-1 covers only the publicly traded entity; Musk will retain 85% voting control [12]. The launch business — the very infrastructure Starlink depends on — operates as a cost center within the same company. Outside investors would own a minority stake in a company whose most critical asset (launch capability) generates losses and whose capital allocation is controlled by a founder with stated ambitions to colonize Mars.

Jay Ritter, a finance professor at the University of Florida, put the concern bluntly: "Even if Starlink generates tens of billions of dollars per year in profits, the money may be squandered on sending people to Mars rather than sending the money to shareholders" [12].

The Physics Ceiling: Can Satellites Beat Fiber?

Starlink's bull case rests on serving the roughly 3 billion people worldwide who lack reliable broadband. Its bear case is equally straightforward: satellite physics impose hard limits that make Starlink structurally inferior to fiber and 5G in the dense, wealthy markets where broadband revenue is concentrated.

Fiber latency runs 5–10 milliseconds. Starlink's signal must travel 550 km up, then 550 km back down — a minimum 1,100 km round-trip that produces latency of 30–40 ms [13]. The gap is irrelevant for web browsing but material for gaming, video conferencing, financial trading, and other latency-sensitive applications.

More critically, fiber provides dedicated bandwidth per connection via individual glass strands. Starlink shares bandwidth across all users in a geographic cell via radio beams from orbit [13]. A 1 Gbps fiber plan typically delivers symmetrical speeds — 1 Gbps up and down. Starlink upload speeds cap at 20–40 Mbps even on its MAX plan [13].

As population density increases, the bandwidth-per-beam constraint becomes more binding. Each satellite can serve only so many users within its coverage footprint before speeds degrade. This means Starlink's strongest performance comes in precisely the markets with the fewest potential subscribers — rural and remote areas — while its weakest performance comes in cities where broadband spending is highest.

The implication for investors: Starlink is a strong product for a limited addressable market, and a limited product for the large addressable market.

Regulatory Headwinds Across Continents

The S-1 devotes 47 pages to risk factors, with regulatory and spectrum risks featuring prominently [12]. SpaceX states that its expansion "is dependent upon regulatory and licensing approvals" and that "there can be no assurance that our applications to renew, modify, or expand our authorizations will be granted on a timely basis, or at all" [14].

India has demanded that Starlink store user data locally, grant intelligence agencies lawful access, allow government audits, and comply with Indian encryption standards. SpaceX accepted these conditions, but India's telecom incumbents — Jio and Airtel — have fiercely contested spectrum allocation for satellite providers, arguing LEO operators should bid for spectrum at auction rather than receiving it administratively [15].

Brazil demonstrated enforcement willingness in 2024 when regulatory authorities temporarily blocked Starlink assets over a dispute involving Musk's X platform and Brazilian court orders [15].

The EU is building IRIS², a competing sovereign satellite constellation backed by SES, Eutelsat, Airbus, and Deutsche Telekom. A direct spectrum acquisition by Starlink in European markets faces political headwinds as member states prioritize European operators [15].

At the ITU level, Iran and Russia successfully forced the inclusion of "unauthorized satellite operations" as an agenda item at the 2023 World Radiocommunication Conference, setting a precedent for future governance constraints on LEO operators like Starlink [16].

SpaceX's acquisition of EchoStar's AWS-4 and H-block spectrum licenses — paid for with 7.96 million SpaceX shares worth roughly $11 billion at the S-1 reference price — illustrates both the strategic importance and the cost of securing spectrum [14].

The FCC Subsidy Question

Starlink's relationship with U.S. broadband subsidies has been contentious. In December 2020, SpaceX won $885.5 million in bids through the FCC's Rural Digital Opportunity Fund (RDOF) auction, committing to provide 100/20 Mbps service to 642,925 locations across 35 states [17].

In August 2022, the FCC denied Starlink's RDOF funding, concluding that the company could not demonstrate it would meet the required speed thresholds. At the time, Starlink's median U.S. download speed was 64.54 Mbps — well below the 100 Mbps commitment — with median upload speeds of 9.72 Mbps against a 20 Mbps requirement [17]. The FCC upheld its denial on appeal in December 2023 [17].

The denial is a double-edged fact for IPO investors. On one hand, Starlink grew to 10.3 million subscribers and $11.4 billion in revenue without the subsidy — proving commercial viability independent of government handouts. On the other hand, it confirmed that Starlink's speeds in 2022–2023 fell short of basic broadband benchmarks, and it eliminated a potential $885 million revenue stream tied to serving the hardest-to-reach rural Americans.

The S-1 does not disclose how many subscribers in FCC-designated underserved areas currently use Starlink, nor does it detail any current subsidy obligations beyond the government contracts mentioned above.

The Competition: Amazon Leo and OneWeb

Starlink's first-mover advantage is substantial but no longer unchallenged. Amazon rebranded Project Kuiper as Amazon Leo in November 2025 and has launched 212 production satellites, with commercial service expected by mid-2026 [18]. Amazon plans a 3,236-satellite constellation and, critically, can bundle satellite broadband with AWS cloud services, Prime memberships, and its logistics network — a cross-subsidy playbook Starlink cannot match.

OneWeb, now owned by Eutelsat, operates a 618-satellite constellation focused on B2B and government contracts rather than consumer service [18]. Its technical capacity already trails Kuiper, which analysts say has surpassed OneWeb's 3.6 terabits per second of throughput despite having fewer satellites in orbit [18].

The LEO satellite internet market is projected to exceed $30 billion by 2030 [18]. Whether that market supports one dominant player or fragments among several will determine whether Starlink's current margins are sustainable.

What the S-1 Really Says

SpaceX has burned more than $37 billion since inception [11]. It lost $4.9 billion in 2025 on $18.7 billion in revenue [12]. Its only profitable division is Starlink, and Starlink's continued growth depends on Starship — a rocket that is still exploding occasionally during tests.

The S-1 is, at bottom, a bet that satellite internet can scale faster than its physical and regulatory constraints can bind it. The financial disclosures show a business with genuine momentum: subscribers doubling annually, margins expanding, government contracts growing. They also show a business whose founder will control 85% of the votes, whose most critical infrastructure operates at a loss, and whose competitive moat faces a well-funded challenge from Amazon.

For public market investors evaluating what could become one of the largest IPOs in history, the question is not whether Starlink works — it plainly does, for millions of subscribers in dozens of countries. The question is whether a satellite network with a five-year replacement cycle, physics-imposed bandwidth limits, and sovereign regulatory exposure in 160 markets can justify a valuation north of $750 billion. The S-1 provides the data. The market will provide the answer.