SpaceX's $2 Trillion IPO: The AI Company That's Actually a Satellite Business

SpaceX filed its S-1 prospectus targeting a June 2026 listing at a $2 trillion valuation — a figure that would place it among the seven most valuable US IPOs in history if pricing holds. 

The filing runs nearly 300 pages, covers three core business segments, and discloses a $15 billion annual contract with Anthropic that nobody saw coming. It also reveals a company that lost roughly $5 billion last year while trading at over 100 times sales and 300 times trailing EBITDA.

The prospectus is, in important ways, two documents occupying the same filing. One describes an AI infrastructure company with a $28.5 trillion total addressable market, seven of twelve stated growth strategies tied to AI, and nearly half of all segment-specific language dedicated to AI-related terms. 

The other describes Starlink — a satellite internet business generating 61% of revenue, virtually all free cash flow, and a 63% EBITDA margin that would be exceptional in any industry.

Which version of SpaceX public investors are actually buying is the question the listing will answer. The math behind each is dramatically different.

The Anthropic Contract Changes the AI Narrative — But Not the Economics

The most consequential disclosure in the filing is one that almost wasn't in it. An earlier draft of the S-1 reviewed by PitchBook included cost data on SpaceX's first two Colossus II AI compute clusters, built at $2.7 million per megawatt — roughly a fourfold improvement on the industry benchmark for data center construction. 

That figure was deleted from the final version. What remained was the revenue side of the equation: Anthropic is paying SpaceX $1.25 billion per month — $15 billion annually — for access to compute infrastructure through May 2029.

Pair those two numbers and the implied economics are striking. At the disclosed construction cost, SpaceX recoups its AI infrastructure capital expenditure in under a month. Even if actual costs ran double the leaked figure, payback comes in roughly 2.2 months. For context, hyperscalers typically measure data center payback in years.

The commercial logic of the Anthropic deal cuts deeper than the financials alone. Anthropic — a direct competitor of Grok, the AI assistant built by Elon Musk's xAI — is renting the servers from SpaceX. If the compute infrastructure were primarily a moat for Grok, Anthropic would be the last company signing a multi-year contract to train on it.

The deal validates SpaceX's AI infrastructure as a standalone commercial asset capable of attracting third-party customers independent of any internal AI agenda. The $15 billion contract nearly matches the combined 2025 revenue of SpaceX's Space and Connectivity businesses — which makes it, structurally, a second company hiding inside the prospectus.

The gap between narrative and economics, however, remains wide. The AI segment generated just 6.7% of revenue in 2025 (excluding legacy X advertising) and posted a $14 billion free cash flow loss. The Anthropic contract represents future revenue, not historical performance. SpaceX is asking investors to value the AI business on its forward potential while Starlink funds the present.

Starlink Is the Business That Actually Makes Money

Strip the AI narrative out of the filing and what remains is one of the most commercially impressive satellite businesses ever built.

Starlink subscribers surpassed 10 million in Q1 2026, doubling year-over-year, on a 36% operating margin. The connectivity segment delivered 61% of total company revenue and virtually all free cash flow, operating at a 63% EBITDA margin. Revenue grew 32% year-over-year.

For reference, that growth rate and margin profile would make Starlink a highly valued standalone public company in its own right — the kind of business that attracts institutional capital at premium multiples without needing to position itself as anything other than what it is.

The commercial launch business remained profitable despite heavy Starship research and development spending — a balance that reflects the operational leverage SpaceX has built through reusable rocket technology.

The economics of reuse have been extensively analyzed, but the prospectus confirms what private market valuations had been implying: the launch business generates real margins even while absorbing the capital costs of developing humanity's largest rocket.

The consolidated entity's $5 billion annual loss and $40 billion in annual capital expenditure need to be read against this backdrop. With 76% of capex directed toward AI infrastructure rather than rockets, the loss is primarily a function of building data center capacity at extraordinary speed — capacity that the Anthropic contract suggests can be monetized faster than conventional infrastructure investments.

Whether that capital is being deployed wisely at scale is a question the Colossus II cost data — now deleted from the filing — would have helped investors answer directly.

The Valuation Math That Requires Careful Handling

$2 trillion is an abstraction until you anchor it to something. The S&P 500's current total market capitalization runs roughly $45–50 trillion, which means a full SpaceX float at the target valuation would represent just over 0.1% of the index — significant for index mechanics, not economy-altering. But the price-to-revenue multiple at $2 trillion is harder to make comfortable.

At over 100 times sales and 300 times trailing EBITDA, SpaceX is priced for a future that is much larger than its present. That's not unusual for transformative technology companies at the moment of public offering — Amazon traded at extraordinary multiples for years before its current earnings profile emerged. The question is always whether the business model scales the way the filing implies.

The $28.5 trillion total addressable market figure, with 80% attributed to enterprise AI workloads, deserves scrutiny. TAM figures in technology IPO prospectuses are routinely constructed to maximize the implied market opportunity, and a $22.8 trillion AI TAM claim is about as aggressive as these numbers get.

The more grounded reference point is the $15 billion Anthropic contract: that's real money, contracted for a real term, from a real counterparty. It's the empirical foundation under what is otherwise a very large narrative.

Governance risk adds a structural complexity. The filing discloses superclass voting shares that give Elon Musk ten votes per share — an arrangement that preserves founder control regardless of public ownership percentage.

Institutional investors with ESG mandates or governance screens will face a genuine decision about whether Starlink's economics and the AI infrastructure opportunity justify accepting a governance structure that offers limited accountability mechanisms. For passive index vehicles forced to hold the stock on inclusion, the choice doesn't exist. For active managers, it does.

How Fund Managers Are Positioning Ahead of the Listing

Goldman Sachs analyst John Flood documented a consistent pattern: cash balances at large US mutual funds rose measurably before each of the four largest historical US IPOs. The same phenomenon appears to be playing out ahead of SpaceX's June listing, with both Nasdaq 100 and S&P 500 adopting accelerated rules for adding newly listed megacaps to reduce the index inclusion disruption.

The rotation pressure that concerns existing large-cap holders is real. Fund managers who need to build SpaceX positions on inclusion have to source that capital from somewhere — and the most likely source is trimming existing large-cap technology positions.

Alphabet (GOOGL) has been specifically mentioned by analysts as a candidate for rotation out, given its scale, liquidity, and some thematic overlap with AI infrastructure investment themes. The net effect on Alphabet's stock would depend on the magnitude of the rotation and whether organic demand for GOOGL absorbs the selling.

Deutsche Bank analysts pointed to elevated household cash balances persisting from the pandemic period as a potential source of retail demand for the offering — a dynamic that could support pricing even at historically stretched multiples if individual investors treat SpaceX as a generational technology investment in the same category as early Amazon or Google.

Lockup expiration risk is the post-listing variable that tends to be underweighted in IPO analysis. With typical 90 to 180-day lockup periods, insider selling pressure arrives roughly three to six months after the June listing — October through December 2026.

The magnitude of that pressure depends on how many pre-IPO employees and early investors choose to liquidate at what will presumably be a significant premium to their cost basis. For a company where hundreds or thousands of employees hold substantial equity, coordinated or sequential selling at lockup expiration can weigh on secondary market prices even when underlying business performance remains strong.

For investors seeking exposure before the June listing itself, analysts have pointed to Voyager Technologies (VOYG) as a publicly traded proxy — a pure-play contractor with exposure to orbital launch and defense themes that overlap with SpaceX's core businesses.

What the Deleted Disclosure Actually Reveals

The decision to remove the $2.7 million per megawatt construction cost figure from the final S-1 is worth sitting with. Companies delete information from prospectus drafts for various reasons — legal exposure, competitive sensitivity, or simply the preference not to provide benchmarks that could complicate future negotiations. But the Colossus II cost data was arguably the single most compelling evidence that SpaceX's AI infrastructure business has genuinely differentiated economics.

A fourfold improvement on the industry benchmark construction cost means SpaceX can build compute capacity at roughly one-quarter the capital intensity of conventional hyperscale data centers.

Combined with the Anthropic contract's $1.25 billion monthly revenue, the implication is a capital-efficient, fast-payback AI infrastructure business that doesn't exist at this cost structure anywhere else. Removing that number from the final filing obscures the strongest purely financial argument for the AI valuation premium.

What remains is the qualitative case: Anthropic signed a four-year, $15 billion contract despite being a competitor to xAI's Grok, which signals something about the perceived quality and pricing of the infrastructure. That inference supports the AI thesis even without the deleted construction cost data.

But investors pricing SpaceX at $2 trillion have to make that inference themselves rather than working from disclosed numbers — which is a different kind of underwriting decision.

The Question the June Listing Will Actually Answer

SpaceX at $2 trillion is asking the market to accept one of two propositions. The first: Starlink is a highly profitable satellite internet business with growing AI infrastructure attached, and the combined entity deserves a premium multiple because both segments are growing rapidly and the AI infrastructure economics are genuinely superior.

The second: SpaceX is primarily an AI infrastructure company that happens to own Starlink, and the $28.5 trillion TAM justifies pricing that would be indefensible on current revenues alone.

The filing is written almost entirely to support the second proposition — 47% of segment language devoted to AI, seven of twelve growth strategies tied to AI, a total addressable market where 93% is attributed to AI workloads. The financial statements support the first proposition — Starlink at 61% of revenue and 63% EBITDA margins, with AI at 6.7% of revenue and a $14 billion free cash flow deficit.

Neither proposition is obviously wrong. The Anthropic contract suggests the AI infrastructure business is commercially viable at a scale that wasn't visible from the outside. Starlink's subscriber growth and margin profile suggest the connectivity business has more room to run than its current contribution implies.

The question is which one is doing the work at $2 trillion — and whether the multiple attached to the AI narrative is supported by the economics that survive the deletion of the most favorable cost disclosure in the draft.

June will produce an answer. It won't necessarily be the final one.