Will the SpaceX IPO Beat the Market?

• Aug 24 · The supply model was watching the wrong variable. The rebound gave most of itself back — $136.97 on August 21, and the second unlock (319M shares, August 20) knocked the stock 6% intraday and back under the offer for a session. The bigger finding is that on August 14 SpaceX issued about 391M new Class A shares to close its $60B all-stock purchase of Cursor's parent — more stock than that unlock, six days ahead of it, and nowhere in Step 4's schedule. Also corrected: revision 1's funding gap was measured against the IPO raise alone, ignoring a $25B bond sold on June 23. Details in the newest update.
• Aug 13 · The lockup came and went — the revision was the mistake. 911.5M insider shares came free on August 6 and SPCX rose 6.1% that day, 15.8% the next, and 9.7% on August 12 after Musk told staff AI revenue would pass everything else by September. From the August 5 low close it is +35% in five sessions, back above the offer at $146.15 and now slightly ahead of the S&P from the offer. Revision 1 was published the morning after the low close. Scoring it, plus a correction to the August 6 text, is in the August 13 update.
• Aug 6 · Eight weeks — right shape, wrong clock. SPCX peaked at a $201.80 close on June 16 (inside the forecast band), then round-tripped: first close below the $135 offer on July 16, and $108.27 on August 5 after the first earnings report. Every checkpoint the forecast has reached is inside its band, but the decline it put in months 3–6 arrived in month 2. Full scoring and a revised forward call are in the update section.
• Jun 12 · Day one — hit. Predicted: opens up 10–20%, near $157, closes above the offer. Actual: opened $150.00 (+11.1%), closed $160.98 (+19.2%) — both inside the band. The miss: a midday spike to $176.49 (+31%) was more euphoric than called, briefly crossing the ~$175 bonus-unlock level (Step 4).

The answer

Probably not. From the $135 offer, SPCX has ~30% odds of beating the S&P 500 over three years, and a day-one buyer does worse. The expected path: a solid, not euphoric debut (85% odds it closes day one above the offer); a squeeze-prone summer on the ~4% float; a bend down in months 3–6 as staged lockups expand supply ~14x; an end to 2026 above the offer (median $145) but below the summer high. The way this breaks is the Musk premium: if SPCX trades as a belief asset rather than on fundamentals, the multiple can persist far longer than history allows. Part 1 gives the full forecast; Part 2 gives the evidence and the recent record behind it.

Written June 11 and left as written. Ten weeks on, after a 46% drawdown, a 35% rebound through the first lockup expiry and a partial round trip back, SPCX closed at $136.97 on August 21 — $1.97 above the offer and about two points behind the S&P 500 from it. "Probably not" is a three-year call and ten weeks decides nothing. See the August 6, August 13 and August 24 updates.

The reasons, in brief:

• Valuation dominates. ~97x sales while unprofitable is far past the worst historical bucket. IPOs above 40x sales averaged −59% market-adjusted over three years bought at the first close.Corrected Aug 13: the entry price was missing here. Measured from the offer, which is the price this bullet is about, the same bucket averaged −15% on 14 deals. The direction holds; the magnitude was overstated. See the factor table. The businesses can't grow into $1.75T by 2029; upside requires the market to keep paying for belief, not revenue.
• Scale, profitability, and valuation at listing predict returns. Founder control helps, but less. The framework's blind spot is the belief-asset re-rating (Arm +500%), which is exactly SPCX's bull case, so the upside percentiles are widened, not ignored.
• Falling below the offer is the norm. 56% of IPOs are under their offer price three years out. The damage concentrates in months 7–24, after lockups clear and hype meets reported numbers.
• Entry price matters more than the company. Large IPOs bought at the offer beat the market (+13% over 3 years); buying the first-day close loses (median −25.7% at 3 years; 10 of 11 recent deals negative from their day-one close).

The SPCX forecast, on the record

The deal: SpaceX prices after the close on June 11 and begins trading June 12 on Nasdaq as SPCX, 555.6M Class A shares at a fixed $135. The book closed 3.5–4x oversubscribed (~$250B in orders), but the gray market (unofficial venues where shares trade before listing) saw its premium collapse from ~60% in May to ~16%, about $157, by June 10.

The forecast is built in four steps: score the factors, test the price arithmetic, compare past mega-IPOs and backtest the framework against them, map the supply schedule. The trajectory and numbers at the end follow from them.

Step 1 · Score SPCX against the factors

Signal is direction; weight is how much that factor historically moves 3-year returns (effect sizes are in the Part 2 table). A positive signal with low weight barely moves the forecast; the deal is decided by the high-weight rows.

Step 2 · What the $1.75T price has to assume

Revenue cannot grow into this price in three years. Upside from $135 is a bet that the market keeps paying for Starship/Mars optionality, not for the businesses. At $135, SpaceX carries ~12.96B shares for a $1.75T valuation against $18B of 2025 revenue. Run the businesses forward and Starlink ($11.4B growing ~50%) reaches $55–60B by 2029; launch (~$6B, growing slower) adds $10–13B; call it ~$70B total. To merely hold $135 through 2029, the market must still pay ~25x sales. At 15x the share price is ~$81 (−40%); at 8x, top-decile for a mature growth company, ~$43 (−68%). This is why valuation dominates the factor score and the bear tail below is fat.

Step 3 · The mega-IPO comparables and the framework's track record returns from offer, approximate

The better reference class for a $1.75T listing is the largest-ever IPOs; the factor averages come from thousands of mostly small deals far outside this sample. Scoring the framework against each comparable, using only what was knowable on its listing day, it got 5 of 6 right on first-year entry risk and ~3 of 6 at three years. Line color shows the verdict:

Step 4 · The supply schedule est. tradable share of company

Supply is the highest-confidence input. Lockups (agreements barring insiders from selling for a set period) normally hold an IPO's tradable float steady until day 180, when supply roughly doubles. SPCX's float grows ~14x in six months, and the bonus tranche is price-triggered: a rally above ~$175 itself releases more stock. A third of the small starting float went to retail investors, three times the mega-cap norm. This drives the months 3–6 bend in the trajectory below.Flagged Aug 24: this table is a lockup schedule, and it was treated throughout as though it were the whole supply picture. It is not. It counts existing shares becoming sellable and never counts new shares being created. SpaceX issued about 391M Class A shares on August 14 to close the Cursor acquisition — an option signed April 21, 2026 and disclosed in the S-1, so knowable on this date. See the August 24 update.

The forecast frozen pre-trade · June 11, 2026

The four steps above point one way: strong demand and a tiny float early, then mounting supply against an unsupportable multiple. In sequence:

In numbers, the targets and probabilities below are subjective calibrated estimates built from the cohort base rates, adjusted up for the mega-IPO selection effect and the framework's right-tail blind spot (both Step 3), and down for the supply schedule (Step 4). P10/P90 are 10th/90th-percentile outcomes; the band should contain ~80% of what happens, and medians should be beaten about half the time.

Scoring the forecast at eight weeks

Through August 5 SPCX had traded 38 sessions, reported one quarter, and reached the first of its lockup cliffs. That is enough to score the near-term calls and to say what the frozen model got wrong about the mechanism. Everything in Part 1 above is untouched; this section sits beside it.

What actually happened milestone closes, not evenly spaced

The squeeze the forecast expected happened, harder and faster than called: SPCX closed at $201.80 on June 16, its third session, having touched $225.64 intraday that day — 67% above the offer at the high, 49% at the close. It closed well clear of the $175.50 performance level on the way.Corrected Aug 13: this sentence originally added "meeting the condition for the bonus tranche." It did not. The tranche needed five closes at or above $175.50 within the ten sessions ending at the earnings date, and SPCX spent that window between $108 and $135, so its 455.8M shares stayed locked. See the correction. Then it gave all of it back. It joined the Nasdaq-100 on July 7 into a falling tape, aborted the Starship 13 launch on July 16, closed below the offer for the first time that same session, and bottomed near $108 into the first earnings report.Corrected Aug 13: this sentence put "roughly $100B of market value in a day" on the abort. The abort session itself cost about $55B ($135.27 to $131.11); the ~$94B day was the session after, July 17 ($131.11 to $123.99). Q2 revenue beat by 13% and the stock rose 9.4% on the day — then fell 13.6% the next session when investors read the capex line.

The checkpoints, scored

What the frozen model got wrong about the business

The denominator is growing much faster than Step 2 assumed. That step ran $18B of 2025 revenue forward to ~$70B by 2029 and concluded the price could not be grown into. Q2 2026 revenue was $7.81B, up 92%, beating consensus by 13%: connectivity $4.3B (+66%, 12M Starlink subscribers, $1.66B of segment operating profit) and an AI segment — xAI, X, and cloud — at $2.56B, up 247%, which the frozen model did not include at all. SpaceX guided for the first time in its history and raised the full-year number. On a ~$38B 2026 revenue base, the $1.43T market cap is roughly 37x sales, not the 97x trailing multiple the forecast was built on. The valuation half of the bear case is materially weaker than it was in June.

And a bear driver replaced the old one. The frozen forecast's downside was multiple compression: a rich price grinding lower as supply arrived. The market is now pricing something the forecast never modeled — a funding gap. At the Q2 run rate, 2026 capex lands near $65B against roughly $38B of revenue and a negative bottom line. The $75B IPO raise covers about one year of that.Corrected Aug 24: the IPO was not the only raise. SpaceX priced a $25B multi-tranche bond on June 23, eleven days after listing and six weeks before this paragraph was written, and reported about $100.8B of cash as of June 19; roughly $20B of the bond repaid the March bridge loan for xAI and X. The run-rate arithmetic holds, but the liquidity against it was on the order of $80B, not one year of IPO money. See the correction. The three-year bear case is no longer mainly "the multiple has to come in"; it is "the buildout has to be financed," which means new equity, convertible paper, or debt, at prices the company does not control. This is the single largest structural change to the thesis since June 11.

The revised forecast revision 1 · August 6, 2026 · from $108.27

Same construction as the original: subjective calibrated estimates, P10/P90 meant to contain ~80% of outcomes, medians meant to be beaten about half the time. The frozen June 11 numbers are unchanged above and both sets stay on the record. Three changes drive the revision — supply is arriving faster than scheduled, revenue is compounding faster than modeled, and financing risk is now explicit.

The lockup that didn't bite

Revision 1 went up on August 6, priced off the $108.27 close the day before, the morning 911.5M insider shares came free. That close turned out to be the bottom. In the five sessions since, SPCX has gone up 35%, closed back above the $135 offer, and passed the S&P 500 from the offer price. The supply overhang that both the frozen forecast and its revision leaned on arrived exactly on schedule and the market bought it.

The correction

The August 6 update said the bonus tranche triggered. It didn't. The 455.8M-share price-contingent block required five closes at or above $175.50 inside the ten sessions ending at the earnings date, not at any point after listing. SPCX traded between $108 and $135 through that window, so the block stayed locked, and the 911.5M shares that came free on August 6 were the base tranche on its own. The error was visible in the update's own numbers — 911.5M is exactly the 20%, with no 10% on top — and it was not caught. Both places are corrected in place above, with the original wording shown. The frozen Step 4 table is left as written, including its looser "if SPCX ≥ ~$175" phrasing, which is the sort of approximation that made the mistake easy.

Three more things the audit found

Every statistic in Part 2 was re-checked against Ritter's source tables this week. The core numbers hold exactly — the 56.1% below offer at three years, the −16.6% and −25.7% medians, the +38.5% mean, the profitability, scale, dual-class and VC splits, and the month-by-month underperformance windows are all reproduced to the decimal. Three things did not survive.

• The valuation penalty was quoted from the wrong entry price. The −59% for IPOs above 40x sales is measured from the first close. From the offer — which is where Part 1's whole case is framed — the same bucket averaged −15%, on 14 deals. The bear case survives; one of its biggest numbers was four times too big for the price it was applied to.
• The float evidence was missing, and it argues the other way. Large IPOs with a float of 10% or less have been the best bucket for offer-price buyers, +25% market-adjusted over three years. Step 1 scored SPCX's ~4% float as a negative. The reasoning there — expansion speed, not level — is still sound, but the frozen forecast never told the reader the level itself has historically been a positive.
• A $100B market-value loss was attached to the wrong session. The Starship 13 abort cost about $55B on the day; the ~$94B day was the one after.

All three are corrected inline where they appear, with the original wording shown. None of them changes a frozen target or probability, and none is quantified back into the forecast, because re-scoring your own call after the fact using errors you found yourself is how a scorecard stops meaning anything.

Revision 1, scored at one week frozen Aug 6 · from $108.27

One week is not a verdict on calls that run to 2029. It is a verdict on the two that a week can already reach, and a fair look at how the rest are travelling:

Why there is no revision 2

Re-cutting the fan a week after the last one, because the price moved, is the exact error described above. So revision 1's numbers stand and will be scored as written, including the two that already look wrong. For the reader's sake, and so this isn't a hedge: if I were re-cutting it today the year-end 2026 median would be near $130, not $105, with "above $135 at year-end" closer to 45% than 25%. That is an opinion, not a scored forecast, and the difference is the point. The next dated revision comes after Q3 earnings, or sooner if something structural happens rather than something merely large.

Two things did genuinely change this week, and neither is the price. The supply calendar is lighter than modelled — 455.8M shares still locked, and the biggest single unlock in the schedule absorbed on a day the stock rose. And the AI segment now has a public number attached to it: Musk told staff it passes everything else by September, and put SpaceX on a path to 10 GW of AI compute by the end of 2027, which he values at $300–500B of annual revenue. Q2's AI line was $2.56B against $5.25B for connectivity and launch combined, so September is close but not free. Whether any of that is worth $1.92T is the question the three-year call is actually about.

What still has to happen. Roughly 319M shares release around day 70 (August 21), with further tranches near days 90, 105, 120 and 135; about 28% of the insider block, on the order of 1.3B shares, follows Q3 earnings in late October or November; the day-180 backstop clears on December 8; Musk's 6.4B shares stay locked until June 12, 2027. At $146.15 the market cap is about $1.92T, or roughly 51x an estimated $38B of 2026 revenue — richer than the ~37x that revision 1 called a materially weaker bear case, and the capex line that caused the August 5 drop has not moved.

The supply that wasn't on the calendar

The August 13 update's headline was that the largest unlock in the schedule arrived and the stock went up. Nine sessions later most of that is back. SPCX closed at $136.97 on August 21, 6.3% below the $146.15 that update was written off, and the second unlock did what the first one didn't: 319M shares came free on August 20, the stock fell as much as 6% intraday to $131.86, closed down 4.1%, and spent a session back under the offer. From the $135 offer SPCX is now +1.5% against +3.8% for SPY since June 11 — about two points behind the market, having been 3.6 points ahead nine sessions ago.

That is the price, and the price is not the point. What this stretch turned up is that the supply model on this page has been counting the wrong thing.

The day-70 tranche was the one the last update placed on August 21; it came on the 20th, which is inside the day-or-two spread the sources carried. Its behavior is the interesting part. The August 6 unlock was nearly three times larger and the stock rose 6%; this one was a third the size and cost 4%. The difference is not the share count. It is that the first tranche landed on a stock 20% below its offer with a capex scare freshly priced in, and the second landed after a 35% run.

Issuance is not on the lockup schedule

On August 14 SpaceX closed its acquisition of Anysphere, the parent of the AI coding tool Cursor, in an all-stock deal at an implied $60B, and folded it into a new SpaceXAI division. The consideration was 389,289,254 Class A shares for Cursor equity plus 1,752,426 for vested RSUs: about 391M new shares, roughly 5% of the Class A base and about 3% of shares outstanding. It is the largest startup acquisition on record.

Step 4 called supply "the highest-confidence input," and it was — for the thing it measured. It measured lockups: shares that already exist becoming sellable on a schedule printed in the prospectus. It never measured issuance. A lockup expiry moves stock from one pocket to another and leaves the denominator alone; an all-stock acquisition adds to the denominator. For a page whose every target is a per-share number, the second is the more damaging of the two, and it is the one with no calendar.

And it was knowable. SpaceX signed the Anysphere option on April 21, 2026 — seven weeks before the forecast was frozen — and it was disclosed in the S-1: the right to buy Cursor for $60B in Class A shares, or to walk away for a $1.5B termination fee plus an $8.5B deferred-services fee, $10B in all. The deal was then announced publicly on June 16, five days after the freeze and seven weeks before revision 1, which does not mention it either. This is not information that arrived; it is information that was sitting in the filing the rest of Step 4 was built from.

Then on August 19 Bloomberg reported that SpaceX had approached Cognition, the AI coding company behind Devin, about a takeover. CEO Scott Wu answered publicly that Cognition is "not for sale and we haven't been talking," and said the two are discussing a compute arrangement instead. Cognition was valued at $26B in May and is reported to be raising at $40B or more. Nothing happened, and nothing about the approach is confirmed by the company. What is worth recording is that it was the second large AI acquisition attempt in nine weeks, five days after the first one closed. If issuing stock for AI assets is a policy rather than an episode, no lockup table can bound the share count, and the honest version of Step 4 has a row in it that says unknown.

The financing case was measured against one of two raises

Revision 1's headline structural change was a funding gap: at a ~$65B annual capex run rate, "the $75B IPO raise covers about one year of that," so the buildout has to be financed at prices the company does not control. The run-rate arithmetic holds. The funding side of it was wrong, and not marginally.

SpaceX priced a $25B bond on June 23 — five tranches running from 2031 to 2056, drawing roughly $90B of orders — eleven days after listing and six weeks before revision 1 was written. The accompanying disclosure put cash and equivalents at about $100.8B as of June 19. Around $20B of the proceeds repaid the bridge loan taken in March to absorb xAI and X. Net of that, the company went into the second half with something on the order of $80B of liquidity against a ~$65B annual capex run rate — tight, and nothing like one year of IPO money.

That does not delete the risk. A business spending $65B a year with negative free cash flow still has to return to the market, and revision 1's 65% on "raises ≥$10B of equity or convertible debt before June 2027" stands unaltered. Neither of the two financings on the record resolves it: the June bond was straight debt, not equity or converts, and the $60B of stock issued for Cursor was acquisition consideration rather than a raise. Both calls stay open as written. What changed is that the single largest structural claim revision 1 made was described from an incomplete balance sheet, and the sentence is corrected in place with the original wording shown.

One data point against the launch moat

On August 19 the Chinese company LandSpace recovered the first stage of its Zhuque-3 on an onshore pad, on the rocket's second flight — the first time any Chinese company has landed an orbital-class booster. Zhuque-3 lifts about 40,350 lb to low Earth orbit against Falcon 9's 50,265, it is a smaller vehicle, and it is nowhere near Falcon's cadence, which stands at 93 flights so far this year. It changes nothing about 2026 revenue and no forecast on this page moves on it.

It earns a line because of where Step 2's arithmetic rests. That step ran launch forward at $10–13B of growth by 2029 on the assumption that the launch business is a moat that holds while Starlink compounds. Reusable booster recovery was a SpaceX-only capability for ten years. As of August 19 it is not, and the terminal-value half of the bull case is the half that assumption feeds.

What the street is willing to say

Two weeks of published targets, in the order they landed:

Low to high is a factor of six on a $1.9T company ten weeks public. That spread is not a modeling disagreement, it is a disagreement about what business this is. Bernstein's 2031 case needs about ten Starship flights a day; Musk's own stated target is close to one a day by the end of 2027, so the number requires a further tenfold increase on a cadence that does not exist yet. Morningstar's $62 assumes essentially none of it. Both are defensible and they cannot both be nearly right.

Revision 1, scored at two weeks frozen Aug 6 · from $108.27

Nothing here resolves. Two weeks moves a stock and settles nothing that runs to 2029, and the row that flipped — ahead of the market, then behind it — flipped on 6% of price, which is what a two-point spread over ten weeks is worth.

Still no revision 2

The August 13 test was that the next revision comes after Q3 earnings, "or sooner if something structural happens rather than something merely large." The Cursor close qualifies as structural on its face: it permanently changed the share count and it changed what the company is. It is still not a reason to re-cut, for one reason. Nothing about it arrived this week. The option was in the S-1 on June 11 and the deal was public on June 16; the frozen forecast could have priced it and revision 1 could have priced it, and neither did. Re-forecasting because I finally read something that was always there is not new information reaching the model — it is my own error, and the rule on this page is that self-found errors are corrected in place and never re-scored into the call. Q3 earnings, still.

One more thing worth saying plainly, because it cuts against me. The off-the-record opinion published on August 13 — that a re-cut year-end median would be near $130 rather than revision 1's $105 — is currently the closest number on this page, at $136.97. That is worth nothing at all. It was published as an opinion precisely so it could not be mistaken for a scored forecast, and a nine-session lead on a four-month call is noise. The reason to keep the two categories apart is exactly so that when the unscored guess looks good, it does not get to count.

What predicts IPO returns

The average IPO underperforms other firms of the same size by about 3.6% per year for five years after listing (Jay Ritter's dataset: 9,253 US IPOs, 1980–2024), and most of that damage sits in small listings. For a $75B deal, what matters is how the factors below separate winners from losers among large IPOs. They are the same factors scored against SPCX in Part 1, measured as 3-year market-adjusted returns, meaning the IPO's return minus the market's over the same window.

IPO-specific factors, ranked by evidence

Does founder ownership predict better returns?

Yes, with caveats, and it is the factor a SpaceX buyer leans on most. Founder-CEO firms among large US companies earned +8.3%/year benchmark-adjusted over 1993–2002 (+4.4%/year after controls, so not purely a tech-sector artifact). Dual-class IPOs, the classic founder-control structure, returned +29.5% over 3 years vs +18.0% for single-class IPOs; among tech IPOs, dual-class beat the market by 13.8% while single-class lagged by 15.4%.

How often IPOs fall, and when

Falling below the offer price is the norm, not the exception (9,195 US operating-company IPOs, 1975–2021, Ritter): 56.1% trade below offer three years later (57.1% at five, over a third losing more than half), and the median three-year return is −16.6% from the offer, or −25.7% from the first-day close that a day-one buyer actually pays; rare huge winners pull the mean up to +38.5%. Small listings drag these base rates down and large deals clear them more often, but the shape holds at every size.

When the declines happen

The underperformance is back-loaded: IPOs trade roughly market-like for six months, and the damage concentrates in months 7–24 (vs size-matched firms, 1980–2024).

The recent record, 2023–2026

Big deals systematically beat the long tail of small listings that make up most of every cohort. Every dot below is a real deal: all 482 operating-company US IPOs from 2023 through June 2026 with a reported deal size and return (stockanalysis.com), excluding SPACs and 27 micro-listings with manipulation-pattern spikes, the amount raised against the return from offer. Green is the 199 deals that raised $100M or more; black is the 283 smaller listings. The few winners above +500% sit at the top edge, with true returns in the tooltips:

The pattern, four years running

Every cohort since 2023 repeats the same shape (Renaissance Capital cohort data: deals with ≥$50M market cap, ex-SPACs):

• Large deals carry everything. The $100M+ slice beat its full cohort at every year-end (+27% in 2023, +29% in 2024, +18–21% in 2025), a premium of roughly 20–30 points. The median listing loses money; only 38.3% of 2025's IPOs ended the year above offer.
• Year-end averages flatter; seasoning reveals. Each cohort finishes its listing year roughly flat (+4%, ≈0%, +1–2%), then decays as lockups clear: 2023's flat year-end became a −30.9% three-year hold (−88.3% vs the market), and 2025's large deals that were above offer at year-end flipped negative through H1 2026.
• Supply grows into weakening returns. Issuance climbed every year ($19.4B, $29.6B, $44.0B raised, and 2026 listings running +17% ahead year-over-year) while the Renaissance IPO Index fell −8% in Q1 2026 vs the S&P's −4%, the hot-market profile that historically precedes the worst cohorts.
• Winners cluster in the cycle's hot theme (AI infrastructure and stablecoins now); fintech/BNPL and crypto cluster among the losers. All of it matches the factors above: big, profitable, reasonably priced, bought at the offer.

The pop fades: day one vs today 2025–26 deals with verified day-one closes

Buying at the day-one close lost money in 10 of these 11 deals (median −45%), the live demonstration of the finding above that the pop is "captured by IPO allocants, not aftermarket buyers." Each deal's day-one pop, next to what a buyer at that day-one close has made since:

The scorecard

The answer, once more: probably not. SPCX should clear the $135 offer on day one, run hot through the summer, bend down as insider stock unlocks, and trail the S&P 500 over years 1–3. The forecast was frozen on June 11, 2026, before the first trade. The page's presentation has been edited since, but every frozen target and probability is unchanged, and the git history plus an archive.org snapshot are the audit trail. Score it on hit rates across all the intervals, not on any single miss.

Standing at revision 1 (August 6, 2026): six near-term calls hit, one missed on duration, one right in direction but two months early, one immaterial, two still open. Every checkpoint reached landed inside its band; every error was one of pace, and all in the same direction — the model was slow. The revised numbers in the update are added beside the frozen ones, never over them, and each revision is dated so the drift is visible. Both sets get scored.

Standing at August 13, 2026: the first lockup expiry passed without the decline that both sets of numbers were built around, and SPCX closed at $146.15 on August 12, back above the offer and 3.6 points ahead of the S&P 500 from it. Revision 1 is one week old and already wrong on its year-end threshold; it stands unaltered anyway, because a forecast re-cut every time the price moves cannot be scored at all. One factual error in the August 6 text — that the price-contingent unlock had triggered — is corrected in place with the original wording shown. No revision 2 until Q3 earnings.

Standing at August 24, 2026: SPCX closed at $136.97 on August 21, $1.97 above the offer and about two points behind the S&P 500 from it. The second unlock cost 4.1% where the first, three times its size, had gained 6%. The substantive finding is not the price: Step 4 tracked lockups and never tracked issuance, and SpaceX issued ~391M new Class A shares on August 14 to close the Cursor acquisition — an option disclosed in the S-1 before the forecast was frozen. A fourth correction lands in the same place as the first three: revision 1's funding gap was measured against the IPO raise alone and ignored a $25B bond sold on June 23. Both are corrected in place; no frozen or revised number is re-scored on the strength of an error found here, and there is still no revision 2 before Q3 earnings.

Methodology & caveats

• Prices are intraday snapshots from June 10–11, 2026; exact percentages may drift a few points with quote timing.
• The 38.3% above-offer figure and year-end cohort averages are Renaissance's December 2025 snapshot, not June 2026. With Figma and Bullish since below offer, today's share is likely lower.
• Figures marked "well-established" or "established" were verified against the cited primary studies; "canonical" items are standard findings from the IPO literature (Ritter 1991 and Ritter's data files) reported as published. All are long-period averages and may not hold for any given year or stock.
• Cohort counts, raises, and year-end and large-deal averages are Renaissance Capital's (deals ≥$50M market cap, ex-SPACs); "listings" counts (stockanalysis.com) include SPACs and micro-caps, and the two should not be mixed. Ritter's 2023 cohort average (−30.9% over 3 years) uses his stricter operating-company set with returns through Dec 31, 2025; the market-adjusted shortfall partly reflects the S&P rally over the same window.
• The scatter shows 482 US IPOs from 2023 through June 11, 2026 with a reported deal size and return on stockanalysis.com's year pages (Deal Size and Return From IPO Price columns), pulled June 12, 2026. 321 SPACs are excluded, identified by name ("Acquisition", "Merger", and similar) or by the $10.00 blank-check unit price combined with a sponsor-style name, with operating companies that price at $10.00 (such as Newsmax and U.S. GoldMining) kept; this matches the report's other statistics, which are ex-SPAC, since a SPAC at IPO trades at trust value and says nothing about operating-company listings. 67 listings lacking a return or deal size are omitted (these counts include SPACs, unlike Renaissance's); nine large deals with missing deal sizes (including ARM, Astera Labs, Reddit, CAVA, Klaviyo, Tempus, Kaspi.kz, Karman, and Metsera) are included with gross proceeds from contemporaneous reports. A further 27 listings that raised $30M or less yet showed returns of +150% to +4,221% are excluded as suspected manipulation: that profile matches the micro-cap pump-and-dump pattern regulators have repeatedly flagged in small IPOs, not genuine investable returns. Returns are June 12 quotes and can differ a few points from the article's June 10–11 snapshots. The $100M+ split uses the reported deal size directly. Returns above +500% (2 deals) are drawn at the +500% edge with true values in tooltips. SPCX, priced June 11 with no trading history, is excluded.
• The fall-frequency statistics (56.1% and 57.1% below offer at years 3 and 5) cover 1975–2021 IPOs, measured from the offer; from the first-day close the medians are worse (−25.7% and −32.0%).
• Share counts are not consistent across the page and cannot be made so without altering frozen text. Step 2 backs out ~12.96B shares from the widely reported $1.75T valuation at the $135 offer; the market caps in the update sections use the ~13.2B shares outstanding that data providers report, which at $135 would be about $1.78T. The gap is under 2% and moves no conclusion, but it means per-share figures derived from a market cap on this page can differ by that much depending on which section they came from. Similarly, the $75B raise is the base deal of 555.6M shares at $135; with the over-allotment exercised, 638.9M shares, proceeds were about $86B, and contemporaneous reports use both numbers.
• The factor table in Part 2 is calculated from the first closing price, as Ritter's tables are. Part 1 scored SPCX against those figures while framing its forecast from the $135 offer, where the same tables give materially different numbers — most sharply for valuation (−15% rather than −59% above 40x sales, on 14 deals rather than 46). Where that conflation changed a stated magnitude it is corrected inline with the original wording shown; the frozen targets and probabilities are unaltered, and the effect on them is not quantified here because doing so would amount to re-forecasting after the fact.
• The SpaceX section was written June 11, 2026, before the first trade, and is left unrevised as a test of the framework. Deal terms are from the S-1 and contemporaneous reports; the ~97x P/S is the $1.75T valuation over $18B of 2025 revenue. Gray-market prices (Hyperliquid) are thin and indicative only, and historical cohort averages may not describe a one-of-one mega-cap listing.
• Mega-IPO comparable returns are approximate (from the offer, split-adjusted, rounded to ~5%, ex-dividends) and are reference points for shape, not precise figures; the price-to-sales chart in Step 2 uses the multiples stated in the comparables table plus approximate listing multiples and 3-year returns, rounded, for about twenty other prominent large US listings, with pre-revenue Rivian drawn at the axis edge. The backtest signals are retrospective scorings and carry hindsight risk despite using only at-the-time observables.
• Day-one closes behind the pop-fades chart are from contemporaneous press reports; BitGo, SailPoint, and Venture Global were reported only as "settled near" levels, so their returns are imprecise by a few points. Some sizable 2026 deals (Madison Air, Fervo, INNIO, BX Digital Infrastructure) lack a verified day-one close and are excluded. "Since day-1 close" returns use the June 10–11 snapshots.
• The SPCX targets and probabilities are the author's subjective calibrated estimates, not a fitted model, and are published to be scored. The 2029 revenue scenarios assume no major new revenue lines (Starship commercial cargo at scale would change them).
• The August 6, 2026 update uses closes through August 5 ($108.27). Only independently corroborated prices are plotted: the $135 offer, the $160.98 day-one close, the $201.80 high close and $225.64 intraday high (both June 16), the $131.11 first close below the offer (July 16), $108.37 (July 31), $125.33 (August 4, earnings day) and $108.27 (August 5). Intermediate daily closes were inconsistent across sources and are omitted rather than smoothed, so the x-axis of the round-trip chart is ordered by event, not evenly spaced in time. The market comparison uses SPY as the S&P 500 proxy: $737.76 on June 11 to $769.79 on August 5, +4.3%.
• The ~37x revised sales multiple is the $1.43T market cap at the August 5 close over an estimated ~$38B of 2026 revenue, which is a third-party full-year estimate rather than company guidance; SpaceX guided for the first time with Q2 results and raised the number, but did not publish a figure this page could verify. Segment figures, capex, and the loss per share are from the Q2 2026 report of August 4. The 2026 capex estimate near $65B annualizes first-half spending of $28.5B plus guidance and is the author's arithmetic, not a company forecast.
• Revision 1's targets and probabilities are subjective calibrated estimates on the same basis as the frozen set. They are added beside the June 11 numbers, which are unaltered; both remain on the record and both will be scored. Later revisions will be dated and appended the same way.
• The August 13, 2026 update uses closes through August 12. Unlike the June–August milestone chart, the unlock-week chart plots eight consecutive daily closes from stockanalysis.com's SPCX history (Aug 3 $114.53, Aug 4 $125.33, Aug 5 $108.27, Aug 6 $114.92, Aug 7 $133.11, Aug 10 $138.74, Aug 11 $133.29, Aug 12 $146.15); the daily percentage moves implied by those closes match the figures reported independently for August 5 (−13.6%), 6 (+6.1%), 7 (+15.8%) and 12 (+9.7%). The market comparison uses SPY, $737.76 on June 11 to $772.49 on August 12, +4.7%. The ~$1.92T market cap is the August 12 close over ~13.2B shares outstanding, and ~51x sales divides it by the same third-party ~$38B 2026 revenue estimate used above, not company guidance.
• The correction to the August 6 update is sourced to the lockup terms as reported before the event: the 455.8M-share tranche required SPCX to hold $175.50, 30% above the offer, for five of the ten trading days through the earnings date, and was reported unmet on July 16 with the stock at $134.91. That 911.5M shares — the 20% base tranche exactly, with no 10% addition — came free on August 6 corroborates it. Original wording is preserved inline rather than deleted. Insider selling after the unlock is not disclosed in real time; nothing here claims to know how much stock actually changed hands, only the price and volume.
• Remaining unlock dates and share counts are from contemporaneous reports of the prospectus schedule and differ by a day or two between sources (the day-70 tranche is variously given as August 12, 20 or 21); day 70 from the June 12 debut is August 21, and the counts are approximate. It came on August 20. The ~28% Q3-earnings tranche depends on an earnings date SpaceX has not yet set; reported dates for the remaining calendar tranches are September 9 and 24 and October 9 and 24, with the day-180 backstop on December 8 and Musk's block on June 12, 2027.
• The August 24, 2026 update uses closes through August 21. The eight consecutive daily closes plotted and tabulated are from stockanalysis.com's SPCX history (Aug 12 $146.15, Aug 13 $141.29, Aug 14 $140.00, Aug 17 $146.23, Aug 18 $143.34, Aug 19 $139.65, Aug 20 $134.00, Aug 21 $136.97), with volumes from the same source; the −4.05% on August 20 and the $131.86 intraday low that day are corroborated independently. The market comparison uses SPY on the basis established above, $737.76 on June 11 to $765.72 on August 21, +3.8%.
• Cursor consideration is 389,289,254 Class A shares for Anysphere equity plus 1,752,426 for vested RSUs, ~391.0M in total, per the closing disclosure; assumed unvested RSUs and options are additional and not counted here. "About 5% of the Class A base and 3% of shares outstanding" uses the ~7.70B Class A and ~13.6B total figures data providers report post-close. The ~$1.86T market cap and ~49x sales continue the page's ~13.2B pre-deal share basis plus the 391M new shares, over the same third-party ~$38B 2026 revenue estimate, which predates the close and excludes SpaceXAI; on trailing revenue of ~$23.0B the multiple is ~81x. The $60B is the implied deal value, not cash paid — the transaction is all-stock.
• The financing correction is sourced to contemporaneous reports of the June 23, 2026 bond pricing ($7.0B 5.350% 2031, $6.0B 5.650% 2033, $6.0B 5.875% 2036, $2.5B 6.600% 2046, $3.5B 6.650% 2056) and the ~$100.8B cash figure disclosed as of June 19, 2026, with ~$20B of proceeds earmarked to repay the March bridge loan for xAI and X. The "~$80B of liquidity" is the author's arithmetic from those two figures and is a point-in-time balance, not a runway calculation: it takes no account of operating cash generation, the second-half draw since June 19, or the terms of the remaining debt.
• The Cognition approach is a Bloomberg report that SpaceX has not confirmed and that Cognition's CEO publicly denied; it is recorded as a reported approach, not a transaction, and nothing on this page is forecast from it. Analyst targets are as published on the dates given and are not endorsements; the consensus figures (average $216, high $450, low $75, 28 buy / 2 sell) are third-party aggregations as of August 21 and vary by provider. The Zhuque-3 payload and Falcon 9 comparison figures are as reported; the 93 Falcon 9 flights are the 2026 year-to-date count as of August 21.