SpaceX After the Listing: Segment Economics, Capital Intensity, and the Limits of the Addressable-Market Argument

On June 12, 2026, Space Exploration Technologies Corp. (Nasdaq: SPCX) completed the largest initial public offering on record, pricing at $135 per share and raising $75 billion, or $86.2 billion including the exercise of the over-allotment option. Within four trading sessions the shares reached $225.64 and the company’s market capitalisation exceeded $2.7 trillion. As of July 10, the stock had declined to an all-time low of $145.07, below the offering price, reducing the market capitalisation to approximately $1.9 trillion, a drawdown of roughly 35% from the peak prior to the company’s first quarterly report as a public entity.

Our conclusion is as follows. SpaceX comprises two proven businesses of genuine quality and one unproven business of extraordinary claimed scale. The two proven segments, launch services and Starlink connectivity, can reasonably support a valuation measured in the hundreds of billions of dollars. The remainder of the current market capitalisation, in excess of one trillion dollars, is attributable to an artificial intelligence segment that generated $818 million of revenue in the first quarter of 2026 against an operating loss of $2.47 billion, and whose principal contracted revenues derive from leasing computing infrastructure to its two most capable competitors. In our assessment, the current share price embeds assumptions about market capture, unit economics, and capital efficiency that the company’s own disclosures do not support. We regard the equity as materially overvalued at present levels, while acknowledging the exceptional quality of the underlying operating franchises.

The S-1 registration statement discloses three reporting segments with fundamentally different economic characteristics, and the analysis benefits from treating them separately.

• Space (launch services). The segment generated revenue of $4.09 billion in 2025, approximately 22% of the consolidated total, and recorded an operating loss of $657 million after absorbing approximately $3.0 billion of research and development expenditure related to the Starship programme. The operational dominance of this franchise is not in dispute. Falcon vehicles flew 96 missions in 2023, 134 in 2024, and 165 in 2025, and the company placed more than 2,200 metric tons into orbit last year, representing over 80% of global mass to orbit. The introduction of reusable launch vehicles transformed the cost structure of the industry and established a durable competitive advantage that entrants have yet to replicate at scale.
• Connectivity (Starlink). This segment is the financial core of the enterprise. Revenue reached $11.39 billion in 2025, an increase of 49.8% year over year; operating income of $4.42 billion grew 120%, implying an operating margin of approximately 39%. The subscriber base expanded from 2.3 million at year-end 2023 to 8.9 million at year-end 2025, and stood at 10.3 million across 164 countries and territories as of March 31, 2026. The constellation comprises approximately 9,600 broadband and mobile satellites in low Earth orbit, roughly three-quarters of all active manoeuvrable satellites. By the first quarter of 2026, connectivity accounted for 69% of consolidated revenue.
• Artificial intelligence. The segment, consolidated through the February 2026 acquisition of xAI, generated revenue of $3.20 billion in 2025 against an operating loss of $6.36 billion. In the first quarter of 2026, revenue was $818 million and the operating loss was $2.47 billion. The segment therefore incurs approximately three dollars of operating loss for each dollar of revenue. It represents 17% of consolidated revenue and is the principal reason the company moved from net income of $791 million in 2024 to a net loss of $4.94 billion in 2025.

The constraint on the launch business is the size of the market rather than the company’s position within it. SpaceX itself quantifies the addressable market for space operations at $370 billion, the smallest of its three segments. Commanding market dominance within a structurally limited market is a source of strategic strength, but not, on its own, of trillion-dollar enterprise value.

The connectivity segment also exhibits the operating leverage characteristic of network infrastructure. Once the constellation is deployed, the marginal cost of serving an additional subscriber is minimal, which explains why operating income grew at more than twice the rate of revenue.

Two qualifications are warranted. First, average revenue per user declined from $99 per month in 2023 to $66 in the most recent quarter, indicating that subscriber growth is being achieved in part through geographic and price mix rather than through pricing power. Second, the segment’s expansion is concentrated in markets where terrestrial infrastructure is inadequate. The enterprise opportunity is substantive: United Airlines has equipped more than 400 aircraft and expects to reach approximately 1,000 by year-end 2026, with its complete widebody fleet targeted for mid-2027, and Lufthansa Group, IAG, and Emirates have announced fleet-wide commitments of their own. Aviation, maritime, and underserved rural markets constitute Starlink’s natural domain. In well-served urban markets, however, fibre and cable remain superior on both speed and cost, and substitution is unlikely. The company sizes the total connectivity opportunity at $1.6 trillion.

It is the artificial intelligence segment that accounts for substantially all of the valuation premium over the two established businesses.

One accounting observation is appropriate here. The convention of expensing research and development as an operating cost systematically understates the current profitability of research-intensive enterprises, and any assessment of SpaceX should adjust for the approximately $3.0 billion of Starship development expenditure absorbed by the launch segment, without which that segment is profitable. The adjustment, however, does not rehabilitate the consolidated picture: the segment disclosures make clear that the losses originate not in the accounting treatment of research but in the operating economics of the AI business itself. What remains true, and is the more important point, is that the largest intangible asset on any balance sheet is future growth. Substantially the whole of SpaceX’s valuation derives from what the company is expected to do rather than from what it currently earns, and an asset of that character demands a correspondingly higher standard of evidence, not a lower one.

The prospectus states that the company has identified the largest actionable total addressable market in corporate history, quantified at $28.5 trillion. Of this figure, $26.5 trillion, or 93%, is attributed to artificial intelligence, within which $22.7 trillion is classified as enterprise applications. The estimate excludes China and Russia. For context, $28.5 trillion approximates the annual gross domestic product of the United States.

We do not consider it productive to dispute the magnitude of the estimate. Even a reduction of 80% would leave an addressable market exceeding $5 trillion, which remains extraordinary. The analytical deficiency lies elsewhere: the size of the market constitutes the entirety of the investment case as presented, and the size of a market is not, in itself, a determinant of value.

A total addressable market is the denominator of a question the company has not yet answered: what share will be captured, at what gross margin, and at what cost of invested capital. Growth creates shareholder value only where the return on incremental invested capital exceeds the cost of that capital. Where growth is accompanied by substandard margins and exceptional capital intensity, it may be neutral or negative to intrinsic value. Nothing in the current disclosures establishes that the AI segment satisfies this condition.

The capital intensity is documented in the filings. Consolidated capital expenditure rose from $4.4 billion in 2023 to $20.7 billion in 2025, of which $12.7 billion was attributable to the AI segment. In the first quarter of 2026 alone, AI capital expenditure reached $7.7 billion, an annualised rate above $30 billion, against quarterly segment revenue of $818 million. Evercore projects company-wide capital expenditure of approximately $360 billion in 2030; Goldman Sachs projects consolidated free cash flow of negative $105 billion in 2029. These are not the financial characteristics of a company growing into its valuation. They are the characteristics of a company whose valuation depends on surviving an extended period of negative free cash flow at unprecedented scale.

The most consequential disclosures in the filings concern the AI segment’s contracted revenues.

In May 2026, Anthropic, a direct competitor in frontier model development, agreed to lease the full capacity of the Colossus 1 facility in Memphis at approximately $1.25 billion per month through May 2029. In June 2026, Google agreed to pay approximately $920 million per month from October 2026 through June 2029 for access to roughly 110,000 Nvidia GPUs, citing demand for its Gemini Enterprise platform. Together, these agreements represent approximately $2.2 billion of monthly revenue, or roughly $26 billion annually, payable to SpaceX by the two firms it has identified as its principal competitors in the market it intends to win.

The strategic incoherence is difficult to overstate. A company cannot simultaneously assert that it will capture a dominant share of an industry and lease the majority of its productive capacity to that industry’s leading incumbents under multi-year agreements. These are two distinct strategies, and they are mutually exclusive in their implications for capital allocation.

The tension was underscored on July 9, when the chief executive, responding publicly to commentary regarding Anthropic’s dependence on SpaceX infrastructure, stated that he had been “clearly wrong about Anthropic” and acknowledged that firm as the current leader in artificial intelligence. The concession is notable: the executive of a company whose valuation rests principally on winning the AI market publicly acknowledged that a customer currently holds that position.

There exists a coherent resolution, which is the infrastructure model. SpaceX possesses a demonstrable and possibly durable advantage in constructing and powering computing capacity at scale, an advantage that would become structural should orbital data centres prove viable, given the increasing constraints of power availability, water usage, and local opposition affecting terrestrial siting. Supplying computational infrastructure to the AI industry is a legitimate and potentially attractive business. It is, however, a business whose revenues are measured in the tens of billions rather than the tens of trillions, and it does not support the current valuation. It should further be noted that both compute agreements are terminable on 90 days’ notice, meaning the segment’s most dependable revenues are also its most readily cancellable.

The $60 billion all-stock acquisition of Anysphere, developer of the Cursor coding platform and the largest acquisition of a venture-backed company on record, represents the company’s principal initiative in enterprise AI. The acquired business is substantive: approximately $4 billion in annualised revenue, of which roughly $2.6 billion derives from enterprise customers numbering approximately 50,000, achieved on the fastest revenue trajectory recorded in enterprise software.

The competitive position, however, was deteriorating at the time of purchase. Corporate expenditure data from Ramp indicate that Cursor’s share of the AI coding category declined from approximately 41% in June 2025 to approximately 26% in May 2026, over a period in which Anthropic captured roughly half the category. The transaction was consummated at approximately fifteen times revenue, paid in equity issued days after the offering, and it converts a deliberately model-agnostic product into a captive component of a competing model developer’s ecosystem, thereby compromising the neutrality that constituted a material element of its enterprise appeal.

We do not regard the transaction as irrational. It secures distribution, proprietary usage data, and engineering personnel at a juncture when all eleven co-founders of xAI had departed the firm. It is, however, an acquisition executed with appreciated equity at a moment of competitive weakness, and characterising it as validation of the AI strategy inverts the causality.

Ten days after completing the equity offering, SpaceX issued $25 billion of senior unsecured investment-grade notes in five tranches, with coupons ranging from 5.35% to 6.65% and maturities extending from 2031 to 2056. Proceeds were applied to the repayment of a $20 billion bridge facility incurred in connection with the debt of X and xAI assumed in the February merger. Order books reportedly approached $89 billion, and Oppenheimer, in initiating coverage, projected net debt in excess of $400 billion by 2031.

The sequence merits emphasis. The largest equity raise in history proved insufficient for the company’s funding requirements; within two weeks, the company was issuing thirty-year obligations to finance a segment currently generating three dollars of operating loss per dollar of revenue.

This bears directly on a systemic consideration that distinguishes the present cycle from the technology boom of 1999 to 2000. The dot-com expansion was financed almost entirely with equity, and when valuations corrected, the losses, however severe, were confined to shareholders. The current AI expansion constitutes the largest infrastructure capital cycle in a century, and a growing proportion of it is financed with debt across public bonds, private credit, vendor financing, and special-purpose vehicles. Equity absorbs disappointment; debt does not. A correction in AI-related revenues would therefore transmit through credit channels, with the potential for distress and default extending well beyond the equity holders who knowingly assumed the risk.

The credit markets have begun to price this consideration. SpaceX’s 2036 tranche was issued at a spread of approximately 140 basis points over Treasuries, roughly 40 basis points wider than comparably rated BBB issuance, indicating that fixed-income investors required compensation for execution risk that the equity market, at the time of issuance, had not yet incorporated.

SpaceX is an enterprise of exceptional operational quality. The launch franchise represents one of the significant engineering and commercial achievements of the era, and Starlink is an asset of the first rank: rapidly growing, genuinely profitable, and possessed of widening competitive advantages in aviation, maritime, and underserved terrestrial markets.

The valuation, however, must be assessed on a sum-of-the-parts basis. Starlink, valued generously as a high-growth infrastructure franchise with a 39% operating margin, plausibly supports several hundred billion dollars of enterprise value. The launch segment, dominant within a $370 billion addressable market but currently loss-making owing to Starship development, supports a further sum in the low hundreds of billions at most. The residual, comfortably in excess of one trillion dollars of the current market capitalisation, rests entirely upon an AI segment that generates $818 million of quarterly revenue, incurs $2.47 billion of quarterly operating losses, consumes $7.7 billion of quarterly capital expenditure, has been publicly ranked behind Anthropic by its own chief executive, and derives its most dependable revenues from supplying infrastructure to the competitors it proposes to displace.

The constructive case, properly stated, is that the AI market is of sufficient scale that even a modest share, ultimately earned at moderate margins, would justify the present price. That case may eventually prove correct, but it must be substantiated with evidence concerning relative model quality, the cost of inference, enterprise willingness to pay, and the return profile of the capital programme. What the market purchased in June was not that substantiation. It was the largest addressable-market figure ever presented in an offering document, together with a founder whose record of confounding expert scepticism is genuine but is not a valuation input.

Narratives that terminate at the addressable market are, in our experience, not investment analyses but rationalisations: the decision to own the asset precedes the argument for owning it, and the argument is then assembled in support of the decision. This tendency is universal. Quantitative investors are as susceptible to it as narrative investors; the numbers are simply recruited later in the process. The function of disciplined valuation is to slow that sequence down, to give the analytical faculty an opportunity to interrogate the instinct before capital is committed.

Our recommendation to investors is accordingly one of disciplined patience. The obligation of the analyst is to require that the investment case be completed: that the addressable market be translated into projected revenue, margin, and return on capital before a price is paid for it. Until that translation is available, the appropriate posture toward the equity is abstention, and the price consistent with the evidence currently disclosed is materially below the prevailing quotation.