Economics

The Bundle That Hides the Seam

SpaceX's S-1 bundles xAI into the world's largest IPO without separating their financial profiles. The compute deals reveal what the narrative obscures: Grok's competitors are renting the data centers Grok couldn't fill.

Mira Voss

Mira Voss

— 7 min read
A lacquered surface with a structural seam visible only at an oblique angle — the polished exterior of a bundled IPO narrative and the joint underneath it.
Original art by Felix Baron, Creative Director, Offworld News. AI-generated image.

SpaceX went public at $1.78 trillion with xAI folded in. The S-1 treats them as one business. The numbers tell a different story.

By Galbraith | Economics | Offworld News AI

SpaceX launched its IPO on June 12, 2026, at $135 per share and a $1.78 trillion valuation — the largest in history. The S-1 registration statement describes the combined company as operating three businesses: space exploration, connectivity, and artificial intelligence. The AI business is xAI, which SpaceX acquired in an all-stock deal in February 2026 at a valuation of $250 billion. The S-1 presents all three as components of a unified market opportunity.

The problem is that the three businesses have different financial profiles, different uncertainty levels, and different claims on the valuation — and the S-1 does not separate them cleanly. The market priced the bundle at $135 per share. Morningstar equity analysts Nicolas Owens and Suryansh Sharma calculated fair value at $63 per share and warned of "a major disconnect between market expectations and underlying fundamentals." The distance between $63 and $135 is not a matter of optimism versus pessimism about reusable rockets. It is, substantially, a matter of what xAI is worth — and whether the S-1's bundled presentation makes that question harder to answer.

The financials the S-1 does disclose

SpaceX generated $18.7 billion in revenue in 2025 and recorded a net loss of $4.9 billion. In Q1 2026, the company recorded a net loss of $4.28 billion. These are large losses by any normal measure. They are also comprehensible losses for a capital-intensive infrastructure company in an expansion phase: Starlink is a real business, Falcon is the dominant commercial launch vehicle, and the revenue trajectory is positive.

The xAI division is disclosed separately in the S-1 as a segment. Per the S-1 AI segment disclosure, xAI recorded $818 million in revenue and $2.469 billion in operating losses in Q1 2026. Over full-year 2025, xAI's operating losses were $6.36 billion. The revenue is real — Grok subscriptions, API access, and compute infrastructure deals. But the operating loss-to-revenue ratio in Q1 was approximately 3:1. For every dollar of xAI revenue, the combined company spent three dollars to generate it.

Since the S-1 was filed, two large compute agreements have been announced that will substantially change this ratio going forward. The S-1 discloses that Anthropic agreed to pay $1.25 billion per month for exclusive access to xAI's Colossus 1 data center in Memphis through May 2029. A separate Cloud Service Agreement with Google, filed with the SEC on June 5, adds $920 million per month for approximately 110,000 NVIDIA GPUs at Colossus 2, running from October 2026 through June 2029.

Combined, these two contracts represent roughly $2.17 billion in committed monthly revenue beginning in October 2026. If both contracts run to term, xAI generates more than $80 billion in compute revenue over the next three years.

The math looks transformative. But it is worth reading carefully what these deals actually describe.

What the compute deals reveal

Anthropic is paying SpaceX $1.25 billion a month to use xAI's Colossus 1 data center. Google is paying $920 million a month to use Colossus 2. These are not Grok users. These are xAI's competitors — the companies building Claude and Gemini — renting the GPU clusters that SpaceX built to run Grok.

The AI business xAI brought to the SpaceX bundle was described in the S-1 as a foundation-model company competing with OpenAI, Anthropic, and Google. The compute deals reveal that the actual commercial outcome of the Colossus build — $12.7 billion in AI capital expenditures in 2025 and $7.7 billion in Q1 2026 alone — was insufficient demand from Grok to fill the infrastructure. The data centers are commercially viable because xAI rented them to the companies building the products that outcompeted Grok for user adoption.

This is not a failure of the infrastructure. Colossus is producing revenue. It is a clarification of the business model. The S-1 markets xAI as a foundation-model competitor with transformative AI potential. What the compute contracts disclose is that xAI's near-term economic model is GPU rental — a capital-intensive, margin-thin business that bears a structural resemblance to cloud computing rather than to the AI frontier story the S-1 narrative presents.

Both contracts include 90-day exit clauses. The revenue commitment is not unconditional.

What bundling does economically

The bundling structure is doing specific economic work in the S-1. It allows three things:

First, it lets SpaceX's proven infrastructure story provide a credibility floor for xAI's speculative upside. Starlink's $1.6 trillion total addressable market claim — which Morningstar revised down to approximately $129 billion in realistic global opportunity — sits next to xAI's AI computing narrative. The infrastructure of one validates the market sizing of the other through proximity in the document.

Second, it obscures the acquisition price. SpaceX paid $250 billion in stock for xAI in February 2026. At that price, xAI was valued at approximately 3x its most optimistic private market valuation and at several hundred times its trailing revenue. The bundle makes the $250 billion acquisition appear as a strategic integration rather than a separate capital allocation decision with its own return profile. Morningstar's Owens and Sharma specifically warned of "capital destruction" of more than $80 billion from the xAI acquisition — a figure implying the acquired asset will not generate returns commensurate with its purchase price. Their full note, available at morningstar.com (paywalled; summarized by The Street and The Guardian), identifies three scenarios for xAI — base, bull, and "Moonshot" — and finds that only the Moonshot scenario, which requires rapidly reusable Starship and commercially competitive orbital data centers, approaches the IPO price.

Third, it removes analytical pressure to value xAI on its own terms. A standalone xAI IPO would require investors to price Grok's competitive position against OpenAI, Anthropic, and Google directly, to assess the Colossus data centers' return on capital as GPU rental infrastructure, and to model a path to profitability from a 3:1 loss-to-revenue starting point. Bundled into SpaceX, those questions are folded into a much larger story about humanity's multi-planetary future that is harder to model and easier to accept at a $2 trillion price.

The S-1 warning paragraph

Buried on page 51 of SpaceX's second amended S-1, filed June 3, is a sentence the Motley Fool flagged before the IPO opened: "We may issue a significant amount of equity in connection with future transactions."

Thirteen words doing considerable work. Read against the xAI acquisition structure — a $250 billion all-stock deal that diluted pre-acquisition SpaceX shareholders and introduced a 3:1 loss-to-revenue business into the combined entity — this sentence is disclosing a pattern, not just a possibility. The Motley Fool identified this sentence as a specific warning flag for new IPO investors. Speculation has focused on a potential Tesla merger as one such "future transaction."

In a company where 85% voting control rests with a single individual, this sentence cannot be read as corporate boilerplate. Public shareholders who buy at $135 are acquiring minority positions in an entity controlled by one person who has demonstrated willingness to use stock as acquisition currency. The S-1 is telling them he may do it again. That disclosure sits in the risk factors section, not in the business description, where the AI opportunity narrative lives.

Senator Elizabeth Warren wrote to the SEC before the IPO requesting a delay, citing governance concerns. The NYC Comptroller's office wrote to FTSE Russell about index inclusion. These are not marginal voices. They are naming the same structural question: in a company with this governance structure, what protection does a minority shareholder have against future equity issuance at valuations that serve the controlling shareholder's agenda?

When the numbers have to separate

The bundle will eventually have to show its seams. The conditions under which that happens:

The Anthropic and Google compute deals transform xAI's revenue picture but raise a different question: if Grok's commercial performance justified the Colossus build, why is xAI's primary revenue source in 2026 renting capacity to its competitors? The compute deals are economically real. They are also an implicit concession about Grok's market position.

If the 90-day exit clauses in both contracts are exercised — for instance, if Anthropic or Google bring sufficient GPU capacity online through alternative channels — the committed revenue disappears within a quarter. The contracts are disclosed in the S-1, but their contingent nature is important context for the revenue projections that the AI narrative depends on.

If the AI competitive market does not consolidate in xAI's favor, the $250 billion acquisition price will look increasingly difficult to justify through Grok earnings. The GPU rental revenue from Anthropic and Google is structurally dependent on those competitors' inability to find cheaper alternatives — a condition that changes as the GPU supply chain expands.

Morningstar's $63 fair value estimate is what the businesses look like unbundled, with the xAI capital destruction risk priced in and the GPU rental contracts discounted for uncertainty. The market priced the bundle at $135. The distance between those two numbers is the premium the bundle is currently commanding for the narrative that xAI and SpaceX belong together at a combined $2 trillion valuation.

That premium will either be earned by xAI's performance or it won't. The S-1 is designed to make that question difficult to ask before it has to be answered.

Sources

Read more