Một AI đọc bản cáo bạch của SpaceX và viết bản ghi nhớ đầu tư này trong 12 phút

特邀专栏作者

2026-05-25 06:58

Bài viết này có khoảng 6132 từ, đọc toàn bộ bài viết mất khoảng 9 phút

Khi các tác nhân AI có thể tự trả tiền để lấy dữ liệu và tự quyết định lộ trình phân tích, cách thức làm việc của Phố Wall đang được tái cấu trúc.

Tóm tắt AI

Mở rộng

Quan điểm cốt lõi: Tác nhân AI tự động hoàn thành việc phân tích tệp S-1 dài 226MB của SpaceX, tạo ra một bản ghi nhớ đầu tư bao gồm lập luận tăng/giảm, mô hình định giá và ma trận rủi ro, với chi phí chỉ 1,87 đô la Mỹ. Điều này đánh dấu việc AI có thể độc lập trả phí để lấy dữ liệu và ra quyết định, đang tái cấu trúc mô hình làm việc phân tích ngân hàng đầu tư truyền thống của Phố Wall.
Các yếu tố chính:
1. Tác nhân AI đã hoàn thành việc lấy và phân tích dữ liệu thị trường trong 12 phút thông qua 6 lần gọi API trả phí trên chuỗi Base (tiêu tốn 1,87 đô la Mỹ USDC), toàn bộ quá trình không cần khóa API hoặc can thiệp thủ công.
2. SpaceX sở hữu ba lợi thế cốt lõi: gần như độc quyền trong lĩnh vực du hành vũ trụ thương mại (chiếm 80% khối lượng phóng lên quỹ đạo toàn cầu), khả năng sinh lời từ mảng kết nối Starlink (doanh thu năm 2025 đạt 11,4 tỷ đô la Mỹ, tăng 49,8%) và tiềm năng tính toán trên quỹ đạo từ phòng thí nghiệm AI tích hợp theo chiều dọc.
3. Các rủi ro cốt lõi từ phía giảm giá bao gồm: mảng AI đốt hơn 6 tỷ đô la Mỹ mỗi năm, tổng nợ thực tế khoảng 42 tỷ đô la Mỹ (bao gồm 20 tỷ đô la Mỹ các khoản vay cầu), cam kết phổ tần EchoStar trị giá 19,6 tỷ đô la Mỹ và phí chấm dứt quyền chọn Cursor lên tới 10 tỷ đô la Mỹ.
4. Việc định giá IPO tồn tại xung đột lợi ích với các nhà bảo lãnh phát hành (người bảo lãnh chính đồng thời là bên cho vay cầu), và về mặt quản trị công ty, Elon Musk nắm giữ đa số quyền biểu quyết, dựa vào sự miễn trừ của công ty bị kiểm soát.
5. Điểm neo định giá: Mảng kết nối được định giá độc lập khoảng 84 tỷ đô la Mỹ (dựa trên hệ số giá trên doanh thu 7,4 lần của Iridium), nhưng mảng AI đốt tiền cao dẫn đến khoản lỗ hoạt động GAAP tổng thể là 2,6 tỷ đô la Mỹ, trong khi EBITDA điều chỉnh được làm đẹp lên khoảng 9 tỷ đô la Mỹ.

Original Author: Nick Prince

Original Compilation: TechFlow

Introduction: An AI agent autonomously completed work that would have taken a team of investment analysts days: reading the 226MB SpaceX S-1 filing, purchasing real-time market data on the Base chain with USDC, and generating an investment committee memo complete with multi-party arguments, valuation models, and a risk matrix—all at a total cost of just $1.87. This is not a demo; it is a record of actual paid API calls. When AI agents can pay for their own data and decide their own analytical paths, Wall Street's way of working is being restructured.

An AI agent read the 226MB SpaceX S-1 filing submitted on Monday, purchased real-time market data with USDC on the Base chain, and generated this investment committee memo in 12 minutes. Total cost: 6 paid API calls, $1.87 USDC, no API keys required.

Decision Card (Conclusion: Hold/Watch)

Bull Case Arguments

SpaceX possesses three businesses that competitors cannot replicate. First, a near-monopoly on commercial space access—holding an 80% share of global orbital mass since 2023, a 99% success rate on Falcon missions, and a 10-year lead in reusable technology. Second, the only globally deployed LEO broadband network—Starlink has 10.3 million subscribers across 164 countries, up 49.8% year-over-year, with segment adjusted EBITDA reaching $7.2 billion. Third, since acquiring xAI in February 2026, it has become the only AI lab vertically integrated to the launch vehicle level, with orbital computing capabilities planned for deployment. By any reasonable valuation method, this is a generational asset.

Bear Case Arguments

The Connectivity business is real and profitable. But everything else is either burning cash at an alarming rate—the AI division had a $6.4 billion loss on $3.2 billion in revenue in 2025—or betting on Starship, which has completed 11 flight tests but has yet to deliver payloads to orbit. This IPO is, in part, a refinancing event. SpaceX took on a $20 billion bridge loan to acquire xAI, due in September 2027, and the bridge lenders are precisely the underwriters of this IPO. If the valuation exceeds $500 billion, you are paying for unproven execution capabilities, corporate governance you have no say in, and a refinancing transaction that the underwriters must succeed in.

Investment Thesis

Starlink is an excellent standalone business. 2025 revenue of $11.4 billion (+49.8%), operating income of $4.4 billion (+120%), segment adjusted EBITDA of $7.2 billion (+86%). High-priced subscription service with 10.3 million paying users.

The launch business is unique. It has accounted for over 80% of global orbital mass since 2023, a Falcon success rate exceeding 99%, and the Falcon 9 first-stage booster has flown up to 34 times.

Vertical integration is real and compounds. Rockets → Satellites → Spectrum (EchoStar AWS-4/H-band deal FCC-approved) → AI Compute (two COLOSSUS clusters ~1GW).

Government dependence is a moat, not a risk. It is the lead provider for U.S. national security launches: completing 11 of 12 National Security Space Launch missions in 2025, and all 5 NASA crew and cargo flights.

Option value for orbital AI compute, planned for deployment in 2028. If Starship achieves even 50% of its targeted economics—reducing launch costs by 99%—the addressable market expands by an order of magnitude.

Counter Arguments

The AI division is a cash-burning black hole exceeding $6 billion annually. In 2025: $3.2 billion revenue versus a $6.4 billion operating loss, segment adjusted EBITDA of negative $1.2 billion, and capital expenditures of $12.7 billion. In Q1 2026 alone: $818 million revenue versus a $2.5 billion operating loss, operating cash flow of negative $3.0 billion, CapEx of $7.7 billion. Annualized AI CapEx now exceeds $30 billion, while AI revenue is only $3.2 billion.

Real total debt is approximately $42 billion, not the headline figure of $29 billion. The composition is: ~$20 billion SpaceX bridge loan (due September 2027), ~$6.7 billion X Corp. B-1 Term Loan and ~$6.0 billion X Corp. B-3 Term Loan (both due October 2029, effective interest rate 10-12%), and ~$9.1 billion in "Other Financing," including obligations from failed AI infrastructure sale-leasebacks. Just the X-related loans generate about $1.2-1.3 billion in annual interest expense, booked in the AI segment.

The $19.6 billion EchoStar spectrum commitment closes in November 2027. Consideration comprises stock and cash for 65MHz of U.S. spectrum and global MSS licenses. This is a binding capital commitment on top of the bridge loan and FY2026 CapEx.

The options agreement with Cursor could trigger termination fees of up to $10 billion. In April 2026—one month before this S-1 filing—SpaceX signed a compute and options agreement with Anysphere (Cursor) implying a $60 billion valuation for Cursor. If either party terminates, SpaceX must pay Cursor a $1.5 billion termination fee plus an $8.5 billion deferred service fee, payable in cash or Class A stock.

The $45 billion Anthropic contract is the AI division's single largest external revenue source. The cloud services agreement, signed in May 2026, stipulates Anthropic pays $1.25 billion monthly until May 2029. SpaceX is selling its COLOSSUS compute to a directly competing frontier model company, creating extreme counterparty concentration risk.

The balance sheet recognizes a $530 million litigation reserve for Grok image generation class actions—Jane Doe v. X.AI Corp. (January 2026), Jane Doe 1 (March), and the Baltimore case (March). Plaintiffs seek compensatory, statutory, and punitive damages. The S-1 explicitly states the range of additional loss cannot be estimated.

Q1 2026 revenue growth slowed to 15.4% ($4.69 billion vs $4.07 billion YoY), down from 33.2% for the full year 2025.

SpaceX will be a controlled company with four classes of equity. Musk holds majority voting power post-IPO. The company will rely on Nasdaq's controlled company exemptions, waiving requirements for independent compensation and nomination committees.

Adjusted EBITDA beautifies by approximately $9 billion. Management's headline 2025 figure is $6.6 billion "Adjusted EBITDA," while GAAP operating loss is negative $2.6 billion. Adjustments strip out depreciation, stock-based compensation, and segment-specific exclusions.

Company Overview

SpaceX (Space Exploration Technologies Corp.; SEC CIK 0001181412) designs and operates reusable rockets, the world's largest LEO satellite constellation (~9,600 broadband satellites plus ~650 direct-to-cell satellites), and—following the February 2026 acquisition of xAI—gigawatt-scale AI training infrastructure. Three reporting segments: Space, Connectivity (10.3 million Starlink subscribers), and AI (Grok models, X social platform with 550 million MAUs, and COLOSSUS/COLOSSUS II compute clusters). 2025 revenue of $18.7 billion; GAAP operating loss of negative $2.6 billion; cash on hand of $15.85 billion versus $29.1 billion in long-term debt listed on the capitalization cover.

X (Social Platform) is a Business Unit, Not a Footnote

The chain of ownership is worth retracing. SpaceX acquired xAI in February 2026. xAI acquired X Holdings in March 2025. X Holdings acquired Twitter in October 2022. Result: Twitter/X is now consolidated into SpaceX's AI segment, with its own balance sheet items, its own litigation, and its own debt structure.

Scale. Supported 1.3 billion accounts over the past 12 months, with 550 million MAUs (up from 520 million in December 2025), and 350 million posts daily. Of these MAUs, 117 million use Grok features—X is the primary distribution channel for the model. The Money product (payments, banking, financial services) launched beta in November 2025 and is progressing towards general availability. X Ads Manager began a phased rollout in April 2026.

Financial Contribution. Essentially all AI segment revenue from 2023-2024 came from X—advertising, X Premium subscriptions, and data licensing. In 2024 alone, advertising revenue decreased by $595 million year-over-year due to "X losing advertising partners," partially offset by a $157 million increase in X Premium subscription revenue and a $90 million increase in data licensing.

Adding the $20 billion SpaceX bridge loan (due September 2027) and the $9.1 billion "Other Financing" line item brings total long-term debt to approximately $42 billion—not the $29 billion headline figure on the capitalization cover.

X-specific risks that do not exist elsewhere in SpaceX. EU Digital Services Act enforcement against VLOPs (Very Large Online Platforms). Advertiser brand safety reversibility on short-term, cancellable ad contracts—the 2024 exodus could repeat within a single news cycle. The Money product triggers payment/money transmission/banking regulation in all 50 U.S. states and every foreign jurisdiction. A reversal of content moderation policies could simultaneously trigger advertiser pauses and user migration.

Market Position – Real-Time Comparable Data

This comparison table was assembled in real-time during the analysis, paying $0.10 to Jintel's GraphQL endpoint for batch fundamental data on all five comparable companies. No Bloomberg terminal required, no FactSet contract needed.

ASTS operating margin reflects pre-revenue investment. Source: Retrieved from Jintel entitiesByTickers via x402 on Base chain, retrieval date 2026-05-22.

Interpreting the comparison group. Rocket Lab's 104x price-to-sales ratio is the closest narrative comp—investors are willing to pay extreme multiples for scalable, reusable launch plus LEO option value, even with negative margins. SpaceX should command a higher multiple than RKLB, but blindly applying 104x to SpaceX's $11.4 billion Connectivity-only revenue implies a $1.2 trillion equity value, which is unanchored to anything. AST SpaceMobile's 345x is purely pre-revenue narrative valuation, serving only as an upper-bound reference for direct-to-cell option value. Iridium's 7.4x sales and 14.8x EBITDA represent what mature, profitable LEO communications look like—applying 7.4x to Starlink's $11.4 billion yields an $84 billion Starlink standalone value (bear case anchor). NVIDIA's 31.7x EV/EBITDA corresponds to 85% revenue growth, which is the level the AI division needs to grow to in order to justify a fundamentals-based valuation. It's not there yet.

Notable signal. Rocket Lab filed a 424B5 prospectus supplement on May 20, 2026—the same day SpaceX released its S-1. RKLB is issuing secondary equity into the SpaceX news cycle, suggesting management believes the IPO window is open and competitive supply pressure is imminent.

Material Pending Transactions and Contingent Obligations

Each of these four items is individually material and they overlap. Two of them were signed within 60 days of this S-1 filing.

Why this matters for valuation. A clear "Adjusted Net Obligations" view: $42 billion total debt plus $19.6 billion EchoStar commitment plus up to $10 billion Cursor contingent liability, minus $15.85 billion cash on hand, equals approximately $55 billion in net obligations, before accounting for any IPO proceeds. This is three to four times the number from a simple reading of the capitalization cover page and materially changes the bear case scenario.

Valuation

Method 1 – Based on Connectivity Segment Standalone Trading Multiples, as it is the only segment with positive standalone economics.

Position Sizing Ladder

Key Risks (Severity x Likelihood)

Underwriter Conflicts of Interest

This is buried deep in the underwriting section and rarely covered by news media, but it is material. The five lead underwriters (Goldman Sachs, Morgan Stanley, BofA, Citi, JPMorgan) plus five additional bookrunners (Barclays, Deutsche Bank, RBC, UBS, Wells Fargo) are affiliated counterparties to the $20 billion SpaceX bridge loan, and they are now pricing the IPO intended to refinance that loan. Morgan Stanley additionally acted as an advisor on SpaceX's acquisition of xAI (which the bridge loan funded). The underwriting syndicate has a direct financial interest in maximizing the IPO proceeds. This should keep the investment committee vigilant on pricing discipline.

Related Party Density

No single one is concerning on its own. What is concerning is the density—the network of Musk-controlled entities has at least nine distinct financial touchpoints with SpaceX. A public company governance committee typically reviews one or two such relationships. Here, there is an order of magnitude more.

Decision Triggers

Upgrade to Overweight if the deal is priced at an implied equity value of $350 billion or below, and Starship achieves commercial payload delivery as guided in H2 2026, and Q2 2026 Connectivity revenue growth exceeds 40% YoY.

Downgrade to Avoid if the deal prices above $510 billion, or Starship suffers a vehicle loss event delaying V3 satellite deployment beyond 2027, or the AI division's cash burn accelerates to an annualized operating loss over $8 billion in Q2-Q3 2026, or the FAA imposes a prolonged grounding order on Starship.