Full Analysis of SpaceX IPO Documents: When Satellite Cash Flows Feed the AI Black Hole

Author: Ada, Deep Tide TechFlow

The public market is about to face an unprecedented package of assets: a cash-cow-level satellite internet business, a monopolistic rocket-launch operation, and an AI lab that burns cash equivalent to 4 times its total annual revenue—everything crammed into a single income statement.

According to disclosures in the prospectus, SpaceX’s consolidated revenue in 2025 was $18.67 billion, but it posted a net loss of $4.94 billion; in Q1 2026, revenue was $4.69 billion with a net loss of $4.28 billion. Compared with the $791 million net profit on $1.4 billion revenue in 2024, the trajectory of this curve is clear—one destination: the all-stock merger of xAI completed in February 2026.

It is this transaction that redefines SpaceX—from a “profitable space company” into a “cash-consuming AI infrastructure company.”

Starlink booked $3.26 billion in a single quarter, propping up the group’s cash flow

The data disclosed in the prospectus for the first time lays bare Starlink’s profitability. In 2025, the business recorded $11.4 billion in revenue, up about 50% year over year; operating profit was $4.42 billion; the adjusted EBITDA profit margin reached 63%; and it generated approximately $7.17 billion in operating cash flow over the year.

Entering 2026, growth accelerates even further. In Q1, Starlink revenue reached $3.26 billion and operating profit was $1.19 billion. Subscription users surpassed 10.3 million, distributed across 164 countries and regions, with about 9,600 satellites in orbit. Payload, an independent analytics firm, predicts Starlink’s full-year 2026 revenue will grow by about 80% to $18.7 billion, by which time it is expected to account for 79% of SpaceX’s total revenue.

But the risk lies in the continued decline of per-user value. According to prospectus data cited by BigGo Finance, Starlink’s average monthly ARPU for individual subscribers has fallen from $99 in 2023 to $81 in 2025—an 18% fade over two years. SpaceX has chosen an expansion path of “trading price for volume.” The U.S. lowest-tier plan was cut from $120/month to $50/month, and in some regions terminal equipment is even given away for free. This strategy works in capturing market share, but it drags down unit economics.

The rocket launch business, by contrast, is relatively peripheral. In 2025, it contributed $4.1 billion in revenue—the lowest among the three major segments—but with NASA’s human landing system contracts and 170 Falcon 9 launch missions throughout 2025, this business provides a “can’t-be-replaced” strategic moat rather than cash ammunition.

xAI burns $7.7 billion in a single quarter

If Starlink is a money-printing machine, xAI is the black hole behind it, draining the power grid.

The prospectus shows that xAI’s 2025 revenue was only $3.2 billion, with an operating loss of $6.35 billion, and capital expenditures as high as $12.73 billion. The CapEx of this single division exceeds the combined capital expenditures of SpaceX’s core aerospace business ($3.83 billion) and the Starlink division ($4.18 billion). In Q1 2026, xAI revenue was $818 million and operating loss was $2.47 billion. Capital expenditures for a single quarter surged to $7.72 billion—far faster than last year’s pace.

In 2025, xAI consumed a total of about $14 billion in cash—nearly matching the cash scale generated by all of SpaceX’s other divisions combined. Asset-liability data cited by SpaceWar indicates that the group currently has $23.385 billion in servers and network equipment, $2.97 billion in data center infrastructure, and $14.05 billion in construction in progress—most of which are stacked up for xAI.

xAI is also reshaping the debt structure. According to PitchBook disclosures, in 2025 xAI added $16 billion of new debt for GPU procurement. In March 2026, SpaceX immediately provided a $20 billion bridge loan to refinance xAI’s debt onto the parent company’s balance sheet at a lower cost. In essence, this move uses credit supported by the cash flows from Starlink and the launch business to underwrite AI compute capacity expansion.

Anthropic pays $1.25 billion per month, turning a rival’s training compute cluster into its own customer base

The most dramatic disclosure in the prospectus is the compute contract between SpaceX’s xAI and Anthropic. The former is a direct competitor of the latter in frontier models—and also its largest single paying customer.

Per the S-1 filing, Anthropic agreed to pay xAI $1.25 billion per month to buy 300 MW of compute at the Colossus 1 data center in Memphis, Tennessee, and the contract runs through May 2029. The data center is built specifically for xAI infrastructure, and will open to Anthropic about 220,000 GPU usage rights. Either party may terminate the contract with 90 days’ notice.

Measured on a monthly basis, the annualized value of this contract is $15 billion, and the full-cycle value is expected to exceed $40 billion. SpaceWar’s analysis directly provides a quantified comparison: “$15 billion in annual revenue exceeds Starlink’s total revenue for all of 2024.” In other words, a compute contract from an external AI customer, by volume alone, can already rival the biggest, most profitable standalone segment of SpaceX’s original business.

This arrangement reveals the core of the “vertically integrated AI infrastructure” business model: xAI builds clusters, trains Grok, and sells idle compute to all buyers—including competitors. SpaceX uses Starlink profits to fund construction costs. Anthropic gains a stable supply of compute and avoids lock-in with mega cloud providers such as Microsoft and Amazon.

It is also worth noting the asymmetry in the 90-day termination clause. For a $15 billion annualized contract, such a short-term exit right makes Anthropic’s commitment look more like an “option on compute” than a long-term lock-in. Investors will have to evaluate whether this contract is the start of xAI’s commercialization of compute, or a transitional plan before Anthropic completes its own data center.

18,712 bitcoins sitting on the books—no additional purchases since 2024

Another unexpected disclosure in the S-1 is that, as of March 31, 2026, SpaceX holds 18,712 bitcoins on its balance sheet, with a fair value of $1.29 billion, which is roughly worth $1.45 billion at current prices.

The total cost basis is about $661 million, corresponding to an average purchase price of $35,324 per coin. CoinDesk data shows this holding has not changed since the end of 2024. SpaceX first added Bitcoin to its balance sheet in 2021, and at its peak it held 25,724 BTC; the current scale is smaller than the peak. By comparison, Tesla held 11,509 BTC over the same period—about 60% of SpaceX’s holdings.

This disclosure places SpaceX among the world’s top 7 to top 11 corporate Bitcoin holders (rankings vary slightly depending on methodology). In public remarks made in 2024, Musk defined Bitcoin as a “fundamental currency based on energy,” consistent with his ongoing “energy infrastructure” narrative—solar power, Starship launches, and orbital data centers in particular.

But the detail that it has not added to its Bitcoin holdings for two years is a point worth underlining. In the cycle in which Bitcoin’s price rose from about $35,000 to around $77,000—an increase of more than 120%—SpaceX chose not to add or reduce, treating this $1.45 billion asset as a locked-in strategic reserve rather than a liquid position. Considering the company’s net loss of $4.94 billion in 2025 and xAI burning billions of dollars in cash each quarter, this “inaction” itself is a stance: these bitcoins are not meant to fill the AI funding hole; they are viewed as hard-asset hedges in an uncertain monetary environment.

A closed loop of satellite–compute–AI: can it be priced in the public market?

Putting the four pieces together, what Musk delivers to public market investors is an unprecedented list of assets.

Starlink keeps printing money with a 63% EBITDA margin, and expected 2026 revenue is $18.7 billion. The rocket launch business provides a strategic foothold at a national-security level. xAI burns $14 billion annually for entry into the AI race, and it has already signed a $15 billion/year compute order with Anthropic. On top of that, it also holds $1.45 billion in Bitcoin on the books as non-dollar exposure. SpaceX’s 2025 valuation in private markets jumped from $350 billion to about $800 billion. After the xAI merger, the overall valuation is set at around $1.25 trillion, and the IPO target price points straight to $1.75 trillion.

The logic of this business loop is that the cash flow from the satellite internet feeds the self-built AI compute infrastructure. Part of the compute is used internally (training Grok), and part is sold externally (to paid customers such as Anthropic). From 2028 onward, data centers will be moved to space again, using Starship’s transport capacity and solar power to bypass ground power bottlenecks. Each link cycles internally, minimizing dependence on external suppliers and the capital markets.

But risks also exist within the closed loop itself. xAI’s burn rate far exceeds its current revenue. The 90-day termination clause in the Anthropic contract leaves room for rapid cash drainage. Starlink’s per-user value continues to decline. In Q1 2026, the group’s consolidated net loss hit $4.28 billion—more than 86% of the full-year 2025 loss. For underwriters including Morgan Stanley, Goldman Sachs, Bank of America, Citi, and J.P. Morgan, the question is what investors are actually paying for: an already proven satellite internet business, or an AI compute bet that is still in the burn phase?

Musk currently owns 12.3% of Class A shares and 93.6% of Class B shares, for a total of 85.1% of voting power. This means that no matter how the public market prices this “vertically integrated AI infrastructure” experiment, decision-making power remains entirely in his hands.