Anthropic Is Paying SpaceX $1.25B/Month — and 14 Other Things Hidden in the S-1

TL;DR — What you need to know

• SpaceX is no longer just rockets. The filing consolidates three segments: Space (launch and vehicles), Connectivity (Starlink) and — after the February 2026 xAI acquisition — a massive AI division.
• Starlink is the cash machine: $11,387 M of FY 2025 revenue (+49.8% YoY) and $4,423 M of income from operations (+120.4% YoY), even as monthly ARPU is declining and the filing flags more declines ahead.
• The headline surprise is the Anthropic deal: $1.25 billion per month through May 2029 (~$45 B over three years) in exchange for compute capacity on Colossus and Colossus II. That's more than the current annual Starlink revenue.
• Cursor is an option, not yet an acquisition: the filing discloses an option agreement signed in April 2026 with an implied valuation of $60.0 billion, payable in Class A stock.
• Terafab is a massive bet: joint venture with Tesla (March 2026) and Intel (April 2026) to produce one terawatt of compute hardware per year in-house.
• Macrohard is also new in this filing: an agentic AI platform under development with Tesla, designed to "fully emulate digital workflows" and augment human work.
• Extreme voting concentration: a dual-class (actually tri-class) structure with Mr. Musk in control. Post-IPO SpaceX will be a Nasdaq "controlled company".
• The IPO has no price yet: this is a "no-price" filing — the $2T valuation and $50 B raise floated in the press are market expectations, not SEC disclosures.

Key 2025 / Q1 2026 numbers

The consolidated story told in the S-1 plays at three speeds: Starlink scaling aggressively, Starship absorbing capex and R&D, and the AI segment (in the perimeter for only about two months) burning cash to chase capacity.

Source: pages 17–19 ("Summary Consolidated Financial Data") and pages 87–110 (MD&A) of the S-1.

Segment breakdown (official numbers)

The S-1 formalizes three operating segments. The stories they tell are very different.

Segment Adjusted EBITDA — historical view

Starlink: the cash machine with declining ARPU

Starlink is today the piece that funds the rest of the group. User growth is explosive and the 38.8% operating margin would be the envy of any terrestrial ISP — but the filing is crystal clear on one point: ARPU will keep going down.

Starlink official numbers

• Subscribers (Mar 31, 2026): ~10.3 million, +105% YoY vs. ~5.0 M on March 31, 2025.
• Constellation: ~9,600 broadband and mobile LEO satellites (≈9,000 broadband-only), serving 164 countries and markets.
• Median residential peak-hour download speed: 225 Mbps.
• Share of all active LEO satellites: ~75%.
• Starlink launches in 2025: roughly 3,100 (5× the entire second-largest LEO constellation in the world).
• Starlink Mobile active in ~30 countries, deployed from January 2024.

ARPU: the one curve heading down

Source: page 89 of the S-1 ("Starlink Subscriber ARPU").

The Starship → Starlink V3 storyline

The S-1 explicitly ties Starlink's fate to the success of Starship. The V3 satellites, deploying from H2 2026, are designed to deliver 1 Tbps of downlink per satellite. A single Starship will be able to put up up to 60 V3s in orbit, versus a handful with Falcon 9: the filing talks about a potential twenty-fold increase in Starlink capacity per launch. If Starship slips, the whole constellation roadmap slips with it.

Starship: a 99% cost-per-kg reduction target

The Space segment closes 2025 with $4,086 M of revenue and an operating loss of just $(657) M — strikingly contained given the Starship program. The explanation: Falcon 9 is a wildly mature cash cow that is bankrolling the development of its successor.

Falcon numbers (as of March 31, 2026)

• Cumulative orbital launches: ~650, of which ~620 with Falcon 9 and 11 with Falcon Heavy.
• Mission success rate: above 99%.
• Re-flight record for a single Falcon 9 first stage: 34 flights.
• Cumulative mass to orbit: ~7,400 metric tons.
• Cost-per-kg reduction: from $18,500/kg (historical industry average) to $2,700/kg with Falcon 9 (NASA, 2010): -85%.

Annual mass to orbit: the exponential curve

Starship: what's at stake

• Test flights to date (March 31, 2026): 11.
• V3 fully reusable payload: 100 t to orbit; future generations designed to double to 200 t.
• First commercial payload to orbit expected: second half of 2026.
• FY 2025 R&D dedicated to Starship: $3,004 M (of which $930 M in Q1 2026 alone).
• Steady-state cost target: a 99%+ reduction versus the historical industry average.
• Long-term vision in the filing: "thousands of launches per year and the transport of approximately one million metric tons to orbit annually".

The real scoop: the $1.25B/month Anthropic deal

The piece of the filing that made analysts fall off their chairs is a single clause inside a Cloud Services Agreement signed in May 2026. It's worth quoting in full, verbatim from the document:

"In May 2026, we entered into Cloud Services Agreements with Anthropic PBC […], an AI research and development public benefit corporation, with respect to access to compute capacity across COLOSSUS and COLOSSUS II. Pursuant to these agreements, the customer has agreed to pay us $1.25 billion per month through May 2029, with capacity ramping in May and June 2026 at a reduced fee. The agreements may be terminated by either party upon 90 days' notice. The customer will retain ownership and intellectual property rights in its content, AI models, and related data."

Translation: Anthropic will pay SpaceX roughly $15 billion per year to rent compute capacity on Colossus and Colossus II for the next three years. That's more than the entire FY 2025 Starlink revenue.

What it means for SpaceX

• The AI segment goes from cash-burner to (potential) third revenue pillar.
• The numbers will land in the AI segment: the $3,201 M of FY 2025 revenue is pre-Anthropic; once fully ramped, the run-rate alone adds ~$15 B/yr from Anthropic.
• Anthropic "will retain ownership" of models and data: SpaceX is a pure compute provider, not a co-owner of IP.

What it means for Anthropic

• Confirms the "compute crunch" Dario Amodei has been talking about publicly: demand has exploded beyond expectations.
• There's an exit hatch: either side can terminate with 90 days' notice. An optionality the market tends to underweight when penciling in the run-rate.

Two token factories: Colossus and Colossus II

The filing devotes dozens of pages to AI infrastructure. The exact numbers, as reported by the official document:

The S-1 introduces a new metric: Nameplate Compute Draw, defined as "the number of GPUs installed at period end multiplied by their respective all-in power draw". As of March 31, 2026, this metric reaches 1.0 gigawatt. For industrial context, the filing notes that an industry benchmark for bringing a greenfield 100 MW data center online is roughly two years: SpaceX brought the first Colossus II cluster online in 91 days, the second in 64.

Cursor: an option, not an acquisition

The most viral take on social — "SpaceX bought Cursor" — is inaccurate. The filing talks about a compute and option agreement signed in April 2026 with Anysphere, Inc. (the company behind Cursor).

• Compute agreement: SpaceX supplies GPU capacity to Cursor and collaborates on improving existing models (including Grok), potentially co-developing future models.
• Option agreement: SpaceX has the right, not the obligation, to acquire Cursor at a pre-determined price, or to pay a termination fee.
• Implied valuation of Cursor (if the option is exercised): $60.0 billion.
• Payment form: SpaceX Class A stock, based on the 7-trading-day VWAP prior to closing.
• Termination fees if SpaceX walks away or materially breaches:
  • $1.5 B termination fee
  • $8.5 B deferred services fee
  Payable in cash or, if the IPO hasn't closed yet, in Class A stock.

The strategic rationale, as spelled out by the filing, is to acquire "context-rich, verifiable data" generated by Cursor's coding workflows, to feed the training of Grok and SpaceX's own models. In exchange, Cursor gets access to GPU infrastructure at scale. An integration between infrastructure, models and applications — the strategic leitmotiv that recurs throughout the prospectus.

Terafab and Macrohard: the long-term bets

Terafab — the one-terawatt-a-year fab

Among the most ambitious initiatives announced in the filing is Terafab, defined as "a chip manufacturing initiative with a long-term goal of producing one terawatt of compute hardware each year".

• Framework agreement with Tesla in March 2026.
• Intel joins in April 2026, contributing expertise in design, fabrication and packaging of ultra-high-performance chips.
• No specific project defined yet: the document is clear that every step will require "separate negotiations and agreements".
• Neither Tesla nor Intel is required to remain part of the project.

Macrohard — the agentic AI platform

The filing also mentions Macrohard: "an agentic AI platform designed to be capable of fully emulating digital workflows and augmenting human work", developed in collaboration with Tesla. This is the first time the name appears in an official document; the prospectus discusses it in the context of the convergence between models, compute and consumer/enterprise applications.

Capital structure and debt (official numbers, p. 70)

The balance sheet at March 31, 2026 paints a company well capitalized at the top of the curve and meaningfully indebted underneath.

Condensed balance sheet (March 31, 2026)

Capitalization table — long-term debt

Source: capitalization table on pages 69–70 of the S-1.

SpaceX Bridge Loan — the details

• Signed: March 2, 2026, with a syndicate of banks; Goldman Sachs Bank USA as administrative agent.
• Aggregate principal: $20,000 M unsecured.
• Rate: Term SOFR + 0.75%–1.75% margin (rating-dependent).
• Maturity: September 2, 2027, with two 3-month extension options (0.25% fee each) → maximum maturity March 2, 2028.
• Use of proceeds: full repayment of legacy xAI/X loans (X B-1 Term Loan, X B-3 Term Loan, xAI Fixed/Floating Rate Loans, xAI 12.5% Senior Secured Notes).

SpaceX Credit Facility (revolving)

• Originally signed February 2025, $1,500 M capacity.
• Amended and waived in March 2026 to accommodate the Bridge Loan.
• Further amended in May 2026: capacity raised to $5,000 M, maturity May 19, 2031.
• As of March 31, 2026: no drawn amount.

Other Financings — the $9.1B "failed" sale-leaseback

Note (3) of the capitalization table states that the $9,105 M of Other Financings "include obligations related to certain AI infrastructure assets recorded as failed sale-leaseback transactions". Translation: SpaceX tried to structure the sale of AI assets with a leaseback, but the terms did not allow derecognition of the underlying assets. In accounting terms, these remain liabilities in every respect.

Share structure: three classes and a "controlled company"

The prospectus introduces a three-class issued share structure (Class A, B, C), plus a Class D authorized but not yet issued. The point is simple: Elon Musk controls a majority of the voting power.

Source: capitalization table on page 69 of the S-1. Mr. Musk's final voting-power percentages and the float percentage are shown as placeholders (___%) pending pricing.

Stated TAM: $28.5 trillion (with an asterisk)

The filing quantifies its addressable market as follows:

Source: page 18 ff. of the S-1. The global residential ARPU assumption is $31/month ($43 high-income, $16 upper-middle, $9 lower-middle/low-income), weighted across 1.8 billion households.

S-1 TAM numbers should be taken with a grain of salt: they're theoretical opportunity, not guidance.

Risk Factors: the 10 most material (summarized)

The filing devotes more than 60 pages to Risk Factors. Here are the 10 we think matter most, paraphrased and translated from the official document (pp. 16–17):

• 1. Starship delays. "Any failure or delay in the development of Starship at scale or in achieving the required launch cadence, reusability and capabilities thereafter would delay or limit our ability to execute our growth strategy, including the deployment of next-generation satellites, global satellite-to-mobile connectivity, and orbital AI compute."
• 2. Control concentrated in Mr. Musk. After the IPO, Mr. Musk controls a majority of voting power. "Dual class structure concentrates voting control with Mr. Musk […]. This will limit or preclude your ability to influence corporate matters."
• 3. Substantial indebtedness. "Our substantial level of indebtedness could materially adversely affect our financial condition." The $20 B Bridge Loan maturing in 2027 (max 2028) is the first stress test.
• 4. Dependence on power and AI processors. "Our ability to scale our AI products relies on our terrestrial and orbital AI compute infrastructure, which depends on the availability of power, AI processors, and other critical components."
• 5. AI segment is newly formed and integration-heavy. "The Company's AI segment is recently formed, still being integrated, operates in a rapidly evolving industry and is subject to integration, execution, competitive and operational risks."
• 6. Technologies not yet commercially viable. "Many of our initiatives, including those to develop orbital AI compute at scale, manufacture AI chips at scale, establish a lunar economy, develop human augmentation systems, and transport humans and cargo to the Moon and Mars, involve significant technical complexity, unproven technologies, or technologies that do not exist."
• 7. Regulatory licenses (FAA, FCC, spectrum). Delays or difficulties on renewals (e.g., FAA launch/re-entry licenses, FCC and international spectrum) could "materially delay or disrupt our operations".
• 8. AI and X regulation. "Our AI products and X platform are subject to complex and evolving U.S. and foreign laws and regulations" (EU AI Act, California Frontier Artificial Intelligence Act, NY Responsible AI Safety and Education Act, ePrivacy/GDPR). The FTC has already opened an inquiry into chatbots.
• 9. "Brazil-style" jurisdictional risk. The filing explicitly cites the Brazil Asset Seizure: Brazil's Supreme Court had moved against X (before SpaceX ownership) and seized assets. The risk of spillover across entities related to Mr. Musk is formalized as a risk factor.
• 10. Bylaw restrictions on forum and arbitration. The bylaws include forum-selection clauses and mandatory arbitration for certain shareholder actions, limiting class-action possibilities.

What is this IPO actually worth?

This is a "no-price" filing: it does not contain the price per share, the total offering amount, or the valuation. All the numbers ($2T target valuation, ~$50 B raise, 5% float) circulating in the press and on Hacker News are market expectations, not SEC disclosures.

The multiples the market is pricing

• FY 2025 revenue: $18,674 M
• Speculated valuation: ~$2,000 B → P/S multiple ~107x
• FY 2025 Adjusted EBITDA: $6,584 M → EV/EBITDA multiple ~300x+
• Net loss: present, and getting worse

What the bulls lean on

• Adjusted EBITDA growing in real terms ($6,584 M in 2025, $1,127 M in Q1 2026 alone, annualized ~$4.5 B off a higher base).
• Connectivity growing +49.8% YoY with expanding margins.
• An Anthropic run-rate of ~$15 B/yr just kicking in (not yet reflected in any historical period).
• Starship → Starlink V3 pipeline with potential for a 20× increase in capacity per launch.

What the bears worry about

• Starlink ARPU in structural decline, acknowledged by the company itself.
• Dependence on a single customer (Anthropic) in the AI segment, with a 90-day exit clause.
• Voting concentration: buyers are pricing Musk risk, not SpaceX risk.
• Very low initial float and forced index inclusion: expect turbulence at first lock-up expiry.
• "Speculative" initiatives (orbital AI compute, lunar economy, Terafab) flagged by the filing itself as "unproven or do not exist".