Related on Kindle
In July 2026 the Institute published a practitioner reading of the SpaceX S-1 against a $1.75 trillion ask, and put the base case at $1.55T. SpaceX priced at $135.00 on June 12 — about $1.78T — ran to $225.64 intraday on the third session, and closed at $134.00 on August 21, marginally below its own offer price. Most publishers quietly update the number. This page grades the call instead: what the framework got right, what the base case got wrong, and what the first Form 10-Q shows that no pre-IPO reading could have.
The case study is 62 pages — too big to scroll in a browser. Download a PDF to read on your own time.
The three S-1-era PDFs have not been revised. The Full Case Study, Library Crosswalk and One-Page Summary were written and posted 13 July 2026, before SpaceX priced, and none has been updated since. They are kept as the timestamped record of the call, not as a current reading. The Post-IPO Scorecard is the current document.
Everything below the next panel is the original reading, published 13 July 2026 against the Form S-1 and left standing. It has not been edited to look better in hindsight. What follows here is the grade.
The short version. The valuation framework held and the point estimate was low. The Institute published five price bands describing what a buyer would be paying for at each level, and said the $1.7–2.2T zone meant paying a premium for optionality while anything above $2.2T required the bull case to be true. SpaceX priced at roughly $1.78T, at the bottom of the optionality band. By the third session it traded at $225.64 — near $2.97T — well inside the zone the framework said required the bull case. It then gave the entire move back, trading as low as $104.83 — roughly $1.38T, two bands below where it priced — before closing at $134.00 on 21 August, marginally below its own $135.00 offer price. In ten weeks the market walked four of the five bands. The bands were the right instrument. The $1.55T base case, sitting one band lower in fair value, was about 13% below where the stock priced and 12% below where it has settled.
| Event | Detail | Implied equity value |
|---|---|---|
| Institute base case, 13 Jul 2026 | SOTP across three segments plus optionality. Bull $2.6T, bear $750B. | $1.55T |
| IPO priced, 12 Jun 2026 | 638.9M Class A shares at $135.00 incl. full over-allotment; gross $86,250M, net $85,675M | ~$1.78T |
| First close, 12 Jun 2026 | $160.95, up 19% on the offer | ~$2.12T |
| Intraday high, 16 Jun 2026 | $225.64 — the 52-week high | ~$2.97T |
| Low since listing | $104.83 — the 52-week low, set between the June peak and August | ~$1.38T |
| Close, 21 Aug 2026 | $134.00, below the $135.00 offer price | ~$1.77T |
Offering terms Filed — final prospectus under Rule 424(b)(4), 12 June 2026, and Form 10-Q for the quarter ended 30 June 2026. Prices are market data. Implied equity values are Institute arithmetic on 13,176 million shares outstanding at 30 June 2026 (Class A 7,607M, Class B 5,569M) and are stated to three figures only; they exclude the 391,041,680 Class A shares issued in the Cursor merger on 14 August 2026, which would lift each figure by roughly 3%.
| What the Institute published | What the filings and the tape show | Grade |
|---|---|---|
| The price-band framework. $1.7–2.2T means paying a premium for optionality; above $2.2T the bull case must be true. | Priced into the optionality band, spiked deep into the bull-case band by the third session, then fell through fair value into the undervaluation band at $104.83 before recovering to the optionality floor — four of the five bands in ten weeks. | Held |
| The Anthropic compute agreement is the single largest swing factor in the valuation. | AI solutions and infrastructure revenue went from $311M to $2,194M in a single year. The AI segment is now 32.8% of consolidated revenue against 18.1% a year earlier. | Held |
| Customer concentration flag. Two customers anchor most of AI revenue, and the Anthropic compute agreement carries a mutual 90-day termination right. | The 10-Q discloses two customers above 10% and names neither. Customer B — below the disclosure threshold in every prior period shown — is now 19.5% of quarterly revenue, entirely within the AI segment. The two together are 37.8%. The Anthropic agreement runs at $1.25 billion per month through May 2029 and is terminable by either party on ninety days’ notice after an initial three-month period [Filed — 424(b)(4)]. The 31 December 2026 exit right cited in the original reading below belongs to the Google compute agreement discussed in that panel, not to Anthropic. The panel attributes those Google terms to the amended S-1; no filed document read for this scorecard names Google as a compute customer, and no dollar figure attached to that agreement has been filed. The Institute carries the Google agreement as reported, not filed. | Held |
| The optionality bucket: Cursor, Terafab and orbital compute, credited only above $1.7T. | One of the three now has a price. The Cursor merger closed 14 August 2026 at an implied equity value of $60.0 billion, paid in 391 million Class A shares. That is roughly 3.4% of the offer-price market capitalisation for one of three named options. | Held |
| Base case $1.55T against a $1.75T ask — a $200B gap. | It priced at $1.78T and has settled near $1.77T. The base case was roughly 13% low and sat one band below where the market cleared, in fair value rather than premium for optionality. | Low |
| Space as one of three growing segments in the sum-of-the-parts, at $4.1B of FY2025 revenue. | Space revenue declined: $1,581M in the first half of 2026 against $1,611M a year earlier, while the company as a whole grew 54%. It is 12.3% of revenue and loses money at the operating line — $(542)M in the quarter, on $1,174M of capital expenditure (both Furnished; the revenue lines are Filed). | Missed |
The capital expenditure. Nothing in the original case anticipated the scale of the build. SpaceX spent $28,476 million of capital in six months. That consolidated figure is Filed — it is the purchases of property, plant and equipment line in the statement of cash flows. The split beneath it is Furnished: $23,551 million went into the AI segment against $3,316 million a year earlier, and that breakdown exists only in the earnings release. Gross property, plant and equipment now stands at $83,071 million, and the composition is the tell: $34,771 million of servers and networking equipment against $13,788 million of satellites. Two and a half times as much computing hardware as spacecraft, on the books of a rocket company. A sum-of-the-parts built on FY2025 revenue could not see this, and it changes the shape of the free cash flow the base case was discounting.
The tax shield now rests on a footnote no post-IPO filing has confirmed, and the throttle we flagged is a live question. The original reading put the federal NOL shield at $4–5B of present value, walked from Note 15 of the S-1, and warned that a §382 ownership-change limitation would throttle the xAI portion. The IPO plus the automatic conversion of all preferred stock is precisely the kind of event §382 is written to catch, so the modeling adjustment we flagged is now a live question rather than a hypothetical. No filed document says a limitation has been triggered. But the Q2 2026 Form 10-Q, read end to end, contains zero occurrences of “net operating loss,” “carryforward” or “Section 382.” The figure is not in the document. Deferred tax assets rose to $354 million at 30 June from $141 million at 31 December, and the interim report states no effective tax rate. The honest position is that the $4–5B estimate stands on the S-1 footnote, has not been updated by any filed figure since, and should be treated as of that date rather than as current. Two SpaceX NOL figures circulating in secondary coverage are not traceable to a filing and are not used here.
The straightforward move, when a published valuation is overtaken by a market print, is to edit the number and say nothing. The Institute publishes a framework rather than a rating, and a framework that is never scored is a marketing document. The original reading below stands as it went out on 13 July 2026, marked only where the market has since walked the map, and the S-1-era PDFs in the download strip are the unrevised timestamped record of what was said before the market said anything. They have not been revised.
Not investment advice. The Baratelli Institute is a publisher operating under the Lowe v. SEC publisher exception and does not provide personalised investment advice or issue buy, sell or hold ratings. Nothing on this page is a recommendation to transact in SPCX or any other security.
The single most useful table this company produces is the segment build. Revenue by segment is in the Form 10-Q; the full cost build beneath it — cost of revenue, research and development, selling, general and administrative, down to income from operations — is in the earnings release furnished as Exhibit 99.1. It is unusually generous disclosure, and it shows three businesses in three completely different financial states.
| $ millions, quarter ended 30 Jun | Space | Connectivity | AI | Consolidated |
|---|---|---|---|---|
| Revenue, Q2 2026 | 962 | 4,291 | 2,561 | 7,814 |
| Revenue, Q2 2025 | 746 | 2,588 | 737 | 4,071 |
| Share of consolidated revenue | 12.3% | 54.9% | 32.8% | 100% |
| Income / (loss) from operations, Q2 2026 | (542) | 1,656 | (1,257) | (143) |
| Capital expenditure, first half 2026 | 2,226 | 2,699 | 23,551 | 28,476 |
| Revenue, first half 2026 | 1,581 | 7,548 | 3,379 | 12,508 |
| Revenue, first half 2025 | 1,611 | 5,062 | 1,465 | 8,138 |
Revenue lines Filed — Form 10-Q for the quarterly period ended 30 June 2026. Every operating-income line, segment and consolidated, and every capital-expenditure line are Furnished — Q2 2026 earnings release, Exhibit 99.1 to Form 8-K. The filed income statement carries no loss-from-operations line. Three checks were run: the three segment revenue lines sum to consolidated revenue; the three segment operating results sum to the consolidated result; and consolidated capital expenditure of $28,476M for the half ties exactly to purchases of property, plant and equipment in the Filed statement of cash flows.
Read the loss carefully. The six-month net loss of $4,817 million includes a $1,545 million loss on debt extinguishment — a refinancing charge incurred once, in connection with restructuring the capital stack around the offering. Roughly a third of the half-year loss is that charge. The second-quarter loss on its own is $541 million against $1,008 million a year earlier. The operating trend and the headline half-year number point in opposite directions and only one of them is about the business.
And one line nobody discusses. Inside an AI segment that grew 131% in the half and 247% in the quarter, X's advertising revenue is still falling — $367 million against $426 million in the quarter, $710 million against $870 million for the half, down 18%. Consolidation does not fix a business. It hides one.
Backlog stood at $47,461 million at 30 June, of which $14,286 million sits in deferred revenue, with roughly 56% expected within one year. Starlink reported 12.0 million subscribers at an ARPU of $66 per month — a figure that appears in the earnings release rather than the 10-Q, and is tagged Furnished for that reason.
It was written against the Form S-1 and its amendments, before pricing. It has not been edited to look better in hindsight. The figures, the base case, the segment framing and the price bands all stand as they were published; the only additions are the hindsight note added 21 August 2026 and the ◀ markers on the price-band table, each labelled as a marking so it cannot be mistaken for original text. The scorecard above grades it against what the Form 10-Q for the quarter ended 30 June 2026 and the market actually did. Where the two disagree, the scorecard is current and this section is the record.
The full memo tags every figure to its S-1 page. Disclosed numbers are anchored to filing text; practitioner estimates are tagged with the reasoning. The model’s Sources tab has 40+ tagged assumptions.
SpaceX’s amended S-1 disclosed a compute-lease agreement with Google of $920 million per month — roughly 110,000 Nvidia GPUs (plus CPUs, memory, and related capacity) at SpaceX / xAI data centers, running October 2026 through June 2029 (~$32 billion total), with a reduced-fee ramp June–September 2026. This is in addition to the Anthropic cloud-services agreement (~$1.25 billion / month) already in the base case above. Together the two contracts represent ~$2.17 billion per month — about $26 billion of annualized contracted AI-infrastructure revenue.
Practitioner read: this materially enlarges the AI segment and shifts the mix toward contracted, hyperscaler-grade compute. It is also a concentration and durability flag — two customers now anchor most of AI revenue, and either party can exit after December 31, 2026 on 90 days’ notice (Google frames it as a short-term bridge for Gemini Enterprise demand). Re-run the SOTP with the AI segment re-based for both contracts, then stress it for customer-concentration and early-termination risk. Figures per SpaceX’s amended S-1 (June 2026); verify the exact terms against the filing on SEC EDGAR before relying on them.
The Institute does not publish Buy / Sell / Hold ratings. The practitioner question is at what price does the math support allocation? The reader walks his own portfolio against the five-band map.
Marked in hindsight: the stock priced at $135.00 on 12 June 2026 — about $1.78 trillion on 13,176 million shares — landing at the floor of the optionality band. It ran to $225.64 intraday on 16 June — the third session — at roughly $2.97 trillion, deep inside the bull-case band. It then fell to a low of $104.83, roughly $1.38 trillion, inside the undervaluation band. At $134.00 on 21 August 2026 it sits at about $1.77 trillion, back at the optionality floor and marginally below the offer price.
| Price band | What the math implies | What the reader is paying for |
|---|---|---|
| < $1.0T | Math implies clear undervaluation | Practically nothing — the Anthropic deal alone covers it |
| $1.0–1.4T | Math implies undervaluation ◀ Touched at the $104.83 low | Connectivity ramp + Space + base AI; no optionality credit |
| $1.4–1.7T | Math implies fair value ◀ Traversed on the way down, June–August | Three segments fully credited; optionality bucket modest |
| $1.7–2.2T | Math implies a premium for optionality ($1.75T target sits here) ◀ Priced here, and trades here today | Cursor + Terafab + orbital + Starship monetization |
| > $2.2T | Math implies bull case is required ◀ Reached by the third session; round-tripped in ten weeks | Mars-by-2030 + orbital AI compute + bull-case multiples on all three |
No rating. No price target. The math is on paper; the decision is the reader’s.
SpaceX’s S-1 Note 15 confirms a full federal valuation allowance against US deferred tax assets and a Q1 2026 effective rate of (0.1)%. The accumulated deficit on the balance sheet is approximately $41 billion.
The flat-statutory-rate read says “$41B × 21% = $8.7B of federal tax shield.” The practitioner read says: the shield is finite (depletes within the explicit forecast window), throttled (TCJA 80% cap + §382 ownership-change limitation on the xAI NOL), conditioned (COBE + potential NUBIG on the xAI side), and amplified (§174 R&D capitalization since 2022 makes the tax NOL larger than the GAAP loss base).
Run the mechanics inside the explicit forecast window, present-value at the 11% WACC: $4–5 billion in PV cash tax savings. Small in the context of a $1.75 trillion ask — roughly 25-30 basis points of equity value — but real, sourced, and undisclosed by the prospectus. Add it to the DCF as a discrete line. Do not apply it to the SOTP comparison metrics (those are pre-tax multiples; the after-tax shield is a double-count).
The full mechanics — including the practitioner flag for the post-IPO valuation-allowance release in the first three quarters — are in the 2-page NOL Addendum that ships with the case study.
Each chip below names a move from the SpaceX work and the Institute chapter that taught it. The full Library Crosswalk — with the application to the SpaceX package walked move by move — is the free download at the top of this page.
The full case study (62 pages) is a single PDF you can read offline, share with a client, or hand to a board. It stitches together six artifacts:
Every figure tagged to S-1 page. Three scenarios. Reader-side price-band map. The reading framed for the family-office principal.
Download memoThe federal tax-shield mechanics. TCJA 80% cap, §382 throttle, COBE conditions, §174 R&D capitalization wedge. PV cash tax savings $4–5B.
Download addendumBoard / IC-meeting form. Cover, S-1 in 8 numbers, three-segment business, segment details, Anthropic deal, revenue arch, FCF rollup, SOTP, scenarios, comps, risks, price-band map, library crosswalk.
Download deckNOL-adjusted SOTP + DCF + Comps + NOL Schedule + Drivers + Sources. 40+ tagged assumptions.
Download model PDFThe analytical techniques used in the case, named and mapped to the Institute chapter and page that teaches each one. 33 techniques, 8 guides cited.
Download crosswalkShare-ready summary for LinkedIn / journalist outreach / partner forwards. The whole case in one page.
Download one-pager“In 2006 I met one of the founding partners of what’s now one of the most respected boutique investment banks in the country — when the firm was just a few people in a room. This is the practitioner-shelf version of the bet they were making.”
If the work is compelling and you want the library on your own shelf, these are the five flagship guides the SpaceX educational case study was built from. Each is priced for a single working practitioner; the library is built to compose.
“If the package fits the portfolio, the reader buys. If it does not, the reader passes. Either is honorable. The dishonor is in deciding without doing the work.”
Financial Modeling & M&A · Pick the level that fits
Same practitioner voice at every tier — from the free acquisition records to the 800-page CFO reference.
Related free tools & libraries
Every one of these is a working calculator or reference — no signup, no gate.
Interactive weighted-average cost-of-capital calc with the Baratelli methodology.
73 company references plus the full methodology. Refreshed quarterly.
YTM, YTC, current yield — the fixed-income back-solve.
Macaulay, modified, and effective duration plus convexity adjustment.
Rolling maturity structure with cash-flow, yield, and duration outputs.
Tax-equivalent yield and after-tax comparison for muni holdings.
Sponsor-return math on a leverage-funded distribution.
Stock-based comp valuation via the closed-form Black-Scholes.
Discrete-time lattice for exotic vesting and early-exercise features.
Simulation-based value for TSR-linked performance awards.
Multi-period reinvestment and yield-on-cost across scenarios.
Full library — Apple to Nestle to Reliance. Snapshot inputs sourced to filings.
Berkshire, Danaher, Constellation, Blackstone, Apollo, LVMH, and 37 more.
Board-grade practitioner memos — LVMH, SpaceX, Dell, Disney, Belron.
One-page cheat sheet — formulas, conventions, sensitivity rules.
