Everyone's Arguing SpaceX's Valuation. Run the Same Assumption on NVIDIA and the Real Joke Is the Method.

By The 🍌🐀 (The Banana Rat)

I borrowed the growth assumption used to justify SpaceX’s IPO price, pointed it at NVIDIA, and the answer it produced says more about the method than about either company.

Not investment advice, and not a price target. This is an applied-valuation exercise. I hold a long position in NVIDIA (NVDA) and no position in any other company named here, and may buy or sell NVDA at any time. As-of date: 2026-09-19, market data at the 2026-09-18 close. Share prices can fall. Do your own research. Full scope and disclosure at the end.

The experiment in short

Here’s the thing almost nobody does with a big long-horizon valuation number: take it apart. Do it to SpaceX’s, and the guesses, not the company, turn out to be the story.

In June 2026, SpaceX went public in the largest IPO ever recorded, at roughly $1.8 trillion [3]. Ahead of its June debut, research firm New Constructs ran the numbers: to justify that price, SpaceX would need about 50% revenue growth a year for a full decade, taking roughly $18.7 billion of 2025 revenue to about $1.1 trillion by 2035 [1][2]. Fortune couldn’t find a precedent for that kind of ramp [2].

So I ran the same assumption on NVIDIA. A deliberately simple model gives about $1,174 a share today, more than five times its $222.27 close [10][11].

That is not a price target, and here’s why. It is what those assumptions produce, nothing more. Keep SpaceX’s 50% growth and change only the multiple, the price the market pays for each dollar of sales, and the same model runs from $558 to $2,789 a share. Change only which already-reported year you start from and the answer moves about 40%. A number that bends that easily isn’t a verdict on a company. It’s a readout on the guesses behind it.

At first glance that reads as NVIDIA being wildly cheap next to SpaceX. The likelier reading, and the one the rest of this piece argues, is that the borrowed assumptions are doing far more work than the headline number admits.

The point in one line: a ten-year valuation tells you as much about the guesses behind it as about the company. Always ask to see the guesses before you trust the number.

Before the experiment: is this a fair test?

Two questions have to be settled before borrowing anyone’s yardstick: why the assumption is hard, and whether NVIDIA is a fair place to point it.

Why the last year is the hard part

Fortune zoomed in on the final year of the SpaceX scenario, when revenue would have to climb from about $718 billion at the end of 2034 to about $1.1 trillion a year later, and asked: “What’s the precedent for such a vertiginous ramp in just 12 months? It doesn’t exist.” [2]

Growth is measured in percentages, but customers pay in dollars. Fifty percent of $18.7 billion is about $9 billion of new sales in year one. Fifty percent of $718 billion is about $360 billion in year ten. Same rate, roughly forty times the dollars, and about a third of the decade’s entire revenue gain lands in that final year alone (my arithmetic, not Fortune’s).

For scale: NVIDIA, in the middle of the AI boom, added about $85 billion of revenue in fiscal 2026 [4], and the biggest one-year jump on record I can find, commodity trader Vitol in the 2022 oil spike, was about $226 billion [13]. SpaceX’s final year asks for well over half again as much, and not from a price spike. It needs customers buying more.

“SpaceX’s price doesn’t need one great year. It needs ten in a row, with the biggest one saved for last. That’s not a forecast. It’s a wish list with a due date.”
— The 🍌🐀

Why NVIDIA is a fair subject

A test is only fair if the subject is. Run SpaceX’s assumption on a struggling company and the absurd answer tells you about that company, not about the assumption. NVIDIA is the opposite kind of subject: revenue grew 65% in fiscal 2026 [4], and its latest quarter came in up 106% year over year [10]. For NVIDIA right now, 50% isn’t a stretch. It’s a slowdown.

The catch is the starting line. SpaceX begins the decade at $18.7 billion; NVIDIA begins at about $303 billion [10][12], sixteen times bigger. Same percentage, sixteen times the dollars, every single year. If the assumption still breaks here, the problem can’t be the company.

Key Findings

SpaceX’s $1.75 trillion price rests on a stack of assumptions that rarely get said out loud. The quickest way to see them is to lend those same assumptions to another company and watch what comes out the other end.

• The 4.21× multiple is my translation, built from their numbers. Take the $1.75 trillion price, grow it at the ~10% a year investors are assumed to want (Fortune’s estimate) for ten years to get $4.54 trillion, then divide by the $1.08 trillion of revenue their 50% growth produces in year ten. The result: about 4.21 times sales [1][2]. The ingredients are theirs; the recipe is mine.
• Run SpaceX’s assumption on NVIDIA and you get about $1,174 a share today. From $303 billion of trailing revenue at 50% a year you reach $17.47 trillion of yearly sales, or about $3,045 a share in year ten across 24.15 billion shares [10][11]. Bring that back to today’s dollars and you get $1,174. Here’s the shortcut hiding inside it: because the same growth rate and the same discount rate apply to both companies, they cancel out. $1,174 is simply NVIDIA priced at SpaceX’s ~94 times sales (derived below), against the ~18 times it actually trades at.
• Keep SpaceX’s 50% and one guess still moves the answer fivefold. The guess is the multiple: what buyers pay for each dollar of sales. At 2×, NVIDIA works out to $558 a share. At 10×, $2,789. One unstated guess, a fivefold spread.
• Even the starting point is a choice. Start from NVIDIA’s last full fiscal year ($215.9B of sales) and you get $837 a share. Start from its latest twelve months ($303.0B) and you get $1,174. Same company, same math, 40% apart.
• At this size, the real question is who pays. A $17.5-trillion-a-year NVIDIA needs customers spending on that scale. Today, the four biggest cloud builders (Amazon, Microsoft, Alphabet and Meta) plan about $725 billion of total capital spending for 2026 [7].

Put together: if 50% a year for a decade is what it takes to justify SpaceX at $1.75 trillion, that same growth assumption, carried across on my bridge, outputs a value more than five times NVIDIA’s price. The market clearly doesn’t believe that: it prices NVIDIA at $222.27. The yardstick is the problem. One quiet guess moves it fivefold, a single earnings report moved it 40%, and it assumes someone buys $17.5 trillion of NVIDIA products a year. A valuation that bends this easily isn’t telling you about the company. It’s telling you about the person holding the yardstick.

What the SpaceX number actually says

The facts first. In June 2026, SpaceX went public in the largest IPO ever recorded: 638.9 million shares at $135, including the full exercise of the underwriters’ over-allotment option, for net proceeds of about $85.7 billion and roughly $1.8 trillion of valuation at the offer price [3][15]. The stress test behind this piece came from David Trainer’s team at New Constructs, which ran a reverse discounted-cash-flow scenario ahead of the listing [1].

Three terms in plain English, because the jargon does real work here:

• Reverse DCF. A normal discounted-cash-flow model starts with forecasts and produces a value. A reverse DCF runs the projector backwards: it starts with the price the market is paying and solves for the performance that price requires. It doesn’t say what will happen. It says what would have to.
• CAGR. Compound annual growth rate, the single steady rate that gets you from here to there. Fifty percent a year for ten years means multiplying by 1.5 ten times, which is 57.7×, which is why these numbers get enormous so fast.
• Terminal multiple. What the market eventually pays for a dollar of a company’s annual revenue once it’s mature. A mature industrial might get 2×. A software business with a moat might get 8× or 10×.

New Constructs published the growth and margin: their scenario grows SpaceX’s revenue about 50% a year to 2035 with a 23% NOPAT (net operating profit after tax) margin [1]. Fortune added the return: “By Fortune’s estimates, he’s positing that investors will want a total annual return of around 10%” [2]. The authors weren’t being credulous. They were building a stress test and saying so.

Because New Constructs values profit rather than sales, I have to build a 4.21× revenue bridge to test their yardstick. Here is how the plumbing works. Their model is a proper DCF: it values the cash flows over the whole forecast period plus a terminal value, so today’s $1.75T is not simply a year-ten value sitting a decade away. To translate their exercise into a simple revenue-and-multiple model, I do something they didn’t, and I’m telling you before I do it: treat the entire $1.75T as if it compounds to a year-ten value at the roughly 10% return Fortune inferred, then express that value as a multiple of year-ten revenue. That produces 4.21×. It is not New Constructs’ terminal multiple, and it does not reproduce their DCF. It’s the bridge for the experiment below.

Exhibit 1 — The SpaceX assumption, translated into a simple model

Source: built from [1] and [2] with tools/moonshot-model.py. One sales multiple for every company ignores that some businesses keep far more of each sales dollar as profit than others, which is exactly why Exhibit 3 tests a range of multiples instead of trusting this one.

Notice what just happened. Even translating one valuation method into another required an assumption. That’s the whole article in miniature.

“Every valuation is a Jenga tower of guesses: growth rate, profit margin, discount rate, exit multiple. SpaceX’s trillion-dollar tower stands tall right up until you pull one block. Go ahead, pull one. The tower was never the company. It was the guesses.”
— The 🍌🐀

Actionable Takeaway: when you meet a long-horizon valuation number, ask which assumptions the author defended and which they inherited or bridged. Then ask what happens if you swap each one.

Point the same growth assumption at NVIDIA

Now carry the settings across: 50% growth, ten years, the 4.21× bridge, and the ~10% return. Whatever comes out is not a claim about NVIDIA. It is a readout on the yardstick. One thing to keep straight from here on: “the same yardstick” means their growth assumption carried across on my bridge, not a rerun of their own model. They value profit; I use a multiple of sales.

The experiment in four plain steps

1. Start with what NVIDIA actually sold. Over its latest twelve months, that’s about $303 billion of revenue [10][12].
2. Grow it 50% a year for ten years. That becomes $17.47 trillion of revenue a year by year ten.
3. Turn revenue into a share price. Multiply by a terminal multiple, the price the market pays for each dollar of a mature company’s sales (I use the 4.21× bridge from Exhibit 1), then divide by the number of shares. That gives about $3,045 a share in year ten.
4. Bring it back to today, just as the SpaceX test does. Their analysis asks what future profits are worth today, against a $1.75 trillion price [1]. So I discount NVIDIA’s year-ten price back at the same ~10% a year Fortune estimates they assumed [2]. (They discount every year’s profit; I discount only the year-ten price, a simplification.) That gives a present-value equivalent of about $1,174 a share, against a real closing price of $222.27 [10][11].

Here’s the shortcut hiding inside those four steps. Because the same 50% growth and the same ~10% rate are applied to both companies, they cancel out. What’s left is simple: the $1.75 trillion SpaceX test values SpaceX at about 94 times its 2025 sales [1][2]. Give NVIDIA that same multiple on its $303 billion and you get $1,174 a share. NVIDIA’s actual price is about 18 times its sales [10][11]. So the whole moonshot boils down to one question: should NVIDIA be priced like SpaceX?

“SpaceX gets priced at about 94 times its sales for promises. NVIDIA gets about 18 times for results it has already reported. One of those multiples is paying for a story, and it isn’t the one with 75% gross margins.”
— The 🍌🐀

What $1,174 does and doesn’t mean

Think of the model like guessing what a house will sell for in ten years. You have to pick how fast the neighbourhood grows, what buyers will pay per square foot at the end, and how big the house is today. The answer is only as good as those guesses. The SpaceX test has already chosen the growth rate: 50% a year. Keep that fixed and change just one other guess at a time:

• The price per square foot (the multiple). This is what the market pays for each dollar of NVIDIA’s sales in year ten. At the 4.21× implied by SpaceX’s own price, the answer is $1,174 a share. Drop it to 2× and you get $558; raise it to 10× and you get $2,789. When someone hands you a single number, they have hidden a fivefold spread behind the curtain.
• How big the house is today (the starting year). On August 26, NVIDIA reported a new quarter, so its starting revenue grew. That’s like measuring the house after an extension instead of before. The answer jumped about 40%, from $837 a share (last full fiscal year) to $1,174 (latest twelve months), while the growth rate, multiple and discount rate stayed the same.
• What the market pays for SpaceX today (the real price per square foot). The stress test used a $1.75 trillion value, but SpaceX now trades at about $2.0 trillion on the 13.18 billion shares its latest 10-Q reports [6][15], roughly 108 times its 2025 sales [1][2]. Apply that same multiple to NVIDIA’s $303 billion and the model outputs about $1,350 a share, a model output and not a target, against a real price of $222.27. In other words, investors are already paying more for SpaceX’s moonshot than the stress test said it needed.

The summary: three guesses, three different answers, all on the same company and the same 50% growth: $558, $1,174, $1,350, $2,789.

The numbers this experiment starts from

NVIDIA’s last full fiscal year was $215.9 billion of revenue, up 65% [4]. Then, on August 26, it reported a quarter of $96.2 billion, up 106% from a year earlier and ahead of its own $91 billion forecast [5], with Data Center (the AI chip business) at $89.0 billion, a 75% gross margin and a next-quarter forecast of about $108 billion [10].

So I use the freshest number: the last four quarters added together ($57.0B + $68.1B + $81.6B + $96.2B) come to about $303 billion [4][5][10][12]. SpaceX starts at $18.7 billion, so the same growth assumption is working on a base about 16 times bigger. That’s why the answers get so big, so fast.

Exhibit 2 — SpaceX’s growth assumption, carried across to NVIDIA

Sources: [1][2] for the assumptions; [10][12] for the revenue base; [11] for the 24.15B share count and the $222.27 close; [15] for SpaceX’s 13.18B shares; [6] for SpaceX’s $152.71 close. Generated by tools/moonshot-model.py --base-rev 302.97 --price 222.27 --shares 24.15 (SpaceX per-share rows computed by hand from [15] and [6]).

Go straight to the last row, and read its label first: both figures are hypothetical model outputs, not forecasts and not expected returns, and both rest on assumptions that would have to hold for a decade. At SpaceX’s $152.71 close, its own moonshot scenario implies about 8.5% a year, below the ~10% return that same scenario assumes investors want. Carry the same growth assumption across to NVIDIA, the company I own, and the same bridge prints 29.9%.

The rest of the table explains why. SpaceX, measured by its own moonshot, comes out at about $133 a share, almost exactly its $135 IPO price, because the test was built backwards from that price. Running it forwards lands you where you started, so treat it as a receipt, not a discovery. Yet SpaceX trades at $152.71 [6], above even that. The same growth assumption, carried across to a company measured on reported sales, outputs a number five times its price.

One more thing the table is careful about. $3,045 is what a share would be worth in 2036 if everything went to plan; $1,174 is what that 2036 figure is worth in today’s dollars, on the same if. Money you get later is worth less than money you get now: at 10% a year, $100 promised ten years from now is worth only about $39 today, because $39 invested at 10% grows to $100 over that time. Comparing $3,045 with today’s $222.27 is the most common mistake in this kind of analysis. It skips the discounting step, and without that step it isn’t a valuation at all.

The multiple matters too

Everything so far holds SpaceX’s 50% fixed. Now vary growth and the multiple at once, because the multiple is the guess nobody argues about.

Exhibit 3 — NVIDIA’s value per share today, by growth rate and year-10 multiple

Each cell is what one NVIDIA share would be worth today (present-value equivalent) under that pair of guesses; the bold cell uses both of SpaceX’s assumptions. Actual price: $222.27. Generated by tools/moonshot-model.py. Read the bottom row: at the SpaceX growth assumption, the answer runs $558 to $2,789 depending only on the multiple, a 5× spread. Read the 4.21× column: growth from 20% to 50% runs $126 to $1,174, about 9×. Growth wins the range contest. The surprise is how much the supposedly boring multiple still moves things. And look at the corners: 20% growth at 10× ($299) sits closer to 50% growth at 2× ($558) than either sits to the headline.

Now run it backwards. Instead of picking a growth rate and getting a value, start with the price NVIDIA actually trades at, $222.27, and ask what growth would justify it. At the 4.21× bridge, the answer is 27.0% a year for a decade. At a 10× multiple it’s 16.5%; at 2× it’s 36.8% [11]. Same price, same company, three different stories, and the only thing that changed is a guess the speaker probably never mentioned.

That’s why “the market is pricing 27% growth” is half a sentence. The honest follow-up isn’t “do you believe that growth?” It’s “at what multiple?”

🍌🐀 Read: The 🍌🐀 has watched people argue about growth rates for thirty years and has almost never watched anyone argue about the exit multiple. Watch the multiple, not just the growth rate.

Now turn that question on SpaceX itself. Its AI business, the part its market pitch leans on, lost $6.4 billion from operations on $3.2 billion of revenue in 2025 [14]. Yet the SpaceX yardstick hands that money-losing AI business and a 75%-margin platform [10] the very same multiple. That’s the tell. A yardstick that can’t tell those two businesses apart isn’t measuring the business. The right-hand columns of Exhibit 3 show what happens when the multiple reflects the business instead. The point isn’t which column is right. It’s that SpaceX’s number never had to ask. (I mapped how those layers stack up at the edge in The Edge-LLM Stack: Arm, NVIDIA, and QNX.)

Actionable Takeaway: never accept a single-cell answer from a two-dimensional model. If someone hands you one number, ask for the grid. If they can’t produce it, they haven’t done the sensitivity; they’ve done the arithmetic once and rounded up their confidence.

🧮 Try it yourself: pick the growth rate and multiple you actually believe from Exhibit 3, and see what the model says. Tell me which cell you believe: find me on X @TheBananaRat or Bluesky.

For the record, I’m not a neutral observer of big NVIDIA numbers. In 2024 I argued NVIDIA could eclipse the most lucrative empire in history, the Dutch East India Company at its peak. That’s exactly why I want the grid and not the cell: the bull in me needs the discipline more than anyone.

The assumption nobody names: which year you start from

Fiscal 2026’s $215.9 billion is a real, audited number [4]. Nobody made it up. But treat a moving baseline as a fixed “today” and you can swing the answer 40% without touching a single growth assumption, and nobody has to lie to do it.

That’s what makes this dial the sneakiest of the three. It doesn’t look like an assumption. It looks like a fact. Compare the reported bases, apples to apples:

Exhibit 4 — NVIDIA on SpaceX’s growth assumption, three starting points

Held the same in every row: 50% growth, the 4.21× multiple, the ~10% return and 24.15B shares. All three starting revenues are reported results, not estimates. Generated by tools/moonshot-model.py --base-rev <base> --price 222.27 --shares 24.15.

What this shows. Nothing about NVIDIA’s future changes between these rows. Only the starting point does: which already-reported revenue you call “today.” That one choice moves the answer from $837 to $1,174, a 40% jump, and it changes the “growth the market is pricing” from 31.4% to 27.0% without anyone changing their view of the company. Nothing was learned about NVIDIA’s next decade on August 26. The model got a new starting point, and the starting point is an assumption too.

Actionable Takeaway: any long-horizon number should come with its discounting and its starting point stated. A future price without a discount rate is half a calculation, and a base year you didn’t pick is an assumption you inherited, and for a company growing this fast it can move the answer by more than the whole argument is about.

What the SpaceX assumption would need: who pays?

This is where the SpaceX assumption stops being a number and becomes a real-world claim, and where it falls apart. The SpaceX analysts flagged it themselves: at 50% a year, SpaceX’s 2035 revenue of $1.1 trillion would be about 1.5 times Amazon’s sales today and bigger than the economy of Taiwan [1]. For a company that sold $18.7 billion last year, that’s already a stretch. Run the same assumption on a company selling at real scale and it demands $17.5 trillion of NVIDIA products a year. NVIDIA didn’t produce that number. The SpaceX assumption did.

Now line it up against the biggest AI budgets on earth. The four giant cloud companies (Amazon, Microsoft, Alphabet and Meta) plan about $725 billion of capital spending in 2026, and roughly $5.3 trillion across 2025 to 2030 on Goldman Sachs’ numbers [7]. That covers everything they build: land, buildings, power, cooling, networking and every chip from every supplier.

Hold that number still and read these three:

• SpaceX needs $1.1 trillion in a single year. That is roughly a fifth of what the entire cloud oligopoly plans to spend on all their data centres, power and chips over six years.
• Its biggest disclosed AI deal is $15 billion a year (Anthropic, $1.25 billion a month, [14]). Reaching $1.1 trillion takes about 70 of those, every year, forever.
• NVIDIA’s single year would be $17.5 trillion. More than three times that entire six-year budget. Hand NVIDIA every dollar of that six-year $5.3 trillion and you have covered about 30% of one year.

SpaceX’s number only looks survivable because it starts small: $18.7 billion multiplied 58 times is “only” $1.1 trillion. But the 50% has no brakes. It doesn’t care how big a company already is, or how many customers exist to pay. Put it on a company already selling at scale and you see what it was quietly assuming for SpaceX all along: that growth never slows and buyers never run out.

“A growth rate is just a promise with the details left out. Read the fine print on SpaceX’s 50% and it’s ten years of customers who haven’t shown up yet.”
— The 🍌🐀

And the two moonshots aren’t independent. SpaceX’s own prospectus puts its addressable market at $28.5 trillion, of which about 93% is AI [14], so SpaceX needs the same customers NVIDIA does. It is also itself a major NVIDIA customer: in 2025 its AI segment spent $12.7 billion on capital expenditure, more than its Space ($3.8 billion) and Connectivity ($4.2 billion) segments combined, and the capacity it rents to Anthropic alone includes roughly 325,000 NVIDIA GPUs [14].

And it cannot easily stop being one. The stress test also assumes SpaceX keeps about 23 cents of every sales dollar as profit [1]. For its AI business to do that rather than pass the money to its supplier, SpaceX would have to build chips and software good enough to replace NVIDIA’s, and keep them good enough as NVIDIA’s own products move on. Its prospectus says it has so far agreed with Tesla on no more than a “general framework” for Terafab, its planned chip venture, with timelines, milestones and spending still to be negotiated [14]. My judgment, with moderate confidence: that gap does not close inside this decade. A funded programme with a published schedule would change my mind. I own NVIDIA, and none of this says its shares are cheap.

NVIDIA’s bet is the opposite one, sell the same tools to everyone, which is the argument in the Wintel piece.

One last check of scale, using the biggest yardstick there is. US nominal GDP was running at about $32.49 trillion a year in Q2 2026 [8]. NVIDIA at $17.47 trillion of sales would be about 54% of today’s US economy, and still about 36% in 2036 even if the economy grows 4% a year (my arithmetic, not an official forecast). NVIDIA sells worldwide, so this is only a sense of scale, the same kind of check Fortune ran on SpaceX [2]. But it makes the point plainly.

That’s the thesis in one picture. The assumption baked into SpaceX’s price only looks reasonable because SpaceX starts small. Give it to any company that’s already big and it produces something larger than a third of the US economy. When a yardstick gives an answer like that, the problem isn’t the company being measured. It’s the yardstick, and by that stress test’s own math, SpaceX’s price needs it.

⚠️ Risks & the counter-case

No model survives contact with reality untouched, and this one has soft spots. Some weaken the comparison itself. Others are real risks for NVIDIA shareholders, and since I own the stock, they belong here in plain sight.

• I used a rough tool, on purpose. The SpaceX analysts valued profit, assuming SpaceX keeps about 23 cents of every sales dollar as after-tax operating profit [1]. I used a multiple of sales instead. A dollar of sales isn’t worth the same at every company: NVIDIA keeps far more of each dollar, and a business that keeps more usually earns a higher multiple. So applying SpaceX’s multiple to NVIDIA probably undersells NVIDIA rather than flatters it, which is exactly why Exhibit 3 shows a range instead of one number.
• Nobody has grown 50% a year for a decade at this size. Fortune couldn’t find a precedent for even the final year of SpaceX’s scenario [2], and SpaceX starts at $18.7 billion. For NVIDIA, starting at $303 billion, the same 50% means about $150 billion of new sales in year one and about $5.8 trillion in year ten (my arithmetic). NVIDIA is growing faster than 50% today [10], so year one isn’t the problem; years five and beyond are. In any realistic case growth slows well before year ten, and that’s exactly what the SpaceX assumption never allows.
• The real bear case isn’t slower growth. It’s a lower multiple. Analysts call it multiple compression: investors decide to pay less for each dollar of sales even while the sales keep coming. A supplier selling $17 trillion a year to a handful of giant buyers starts to look more like a utility, and utilities don’t get premium multiples. Exhibit 3’s left column shows the effect: even with the full 50% growth, a 2× multiple gives $558 a share, and with 20% to 25% growth it gives $60 to $90, well below today’s $222.27.
• The customers are becoming competitors, and that cuts against SpaceX too. Several of NVIDIA’s biggest buyers now design their own chips, so its platform has to hold up even if its largest customers partly leave. But SpaceX is one of those customers, and its own AI revenue faces the same risk from the other side: the Anthropic agreement, its biggest disclosed AI contract, can be ended by either party on 90 days’ notice after the first three months [14].
• The share count isn’t a constant, and I treated it as one. Per-share figures divide by 24.15 billion shares [11], but NVIDIA returned about $26 billion to shareholders in Q2 alone, with about $99 billion of buyback authorization left [10]. A shrinking share count would push per-share outcomes higher than shown. The model ignores buybacks, dilution, capital structure, taxes and cyclicality. It’s a revenue-and-multiple exercise, nothing more.

What the risks add up to. None of them saves the SpaceX assumption: three make the moonshot harder, and the other two are limits of my simple model that, if anything, undersell NVIDIA. And yes, NVIDIA has a real bear case: pick a low multiple and slower growth and the model lands below today’s price.

What would change my mind. On the SpaceX side, the 50% path needs roughly $28 billion of revenue in 2026 and $42 billion in 2027 (my arithmetic); every year it falls short, the gap compounds, and its AI segment still lost $6.4 billion from operations on $3.2 billion of revenue in 2025 [14]. On the NVIDIA side, the warning sign is quarter-to-quarter growth stalling while margins fall, or the big cloud companies cutting their spending plans. Neither is happening yet: growth is slowing, not stalling (about +20%, +20%, then +18%, with next quarter’s forecast implying about +12%), and that forecast is still up about 89% from a year earlier [10][12].

“Wall Street didn’t price a rocket company. It priced a wish and called it a growth rate. Hand that same wish to NVIDIA and it asks for more than a third of the US economy. When the yardstick needs a miracle to work, stop measuring the company and start measuring the yardstick.”
— The 🍌🐀

None of this says NVIDIA can’t get there. It says the number you get depends on settings that usually arrive uninspected. Start from an older fiscal year and you’re 40% off before you begin. Borrow someone else’s multiple, one built for a rocket company, and the answer swings fivefold without anyone saying so. The same grid has cells well below today’s price, and share prices can fall; this one is no exception.

This isn’t a price target. It’s arithmetic on assumptions, and the assumptions are the point.

The 🍌🐀 has spoken.

Subscribe for the next recalculation: the model gets re-run on NVIDIA’s Q3 FY2027 print, with every assumption on the page again.

Methodology

Scope & disclosure (full). This is an applied-valuation exercise, not a forecast and not investment advice. It borrows the growth assumption a published analysis used to stress-test SpaceX’s IPO valuation and applies it to NVIDIA, to show what numbers like this actually depend on. No price target is offered here. Every figure below is the output of an assumption stated on the page, and changing the assumption changes the figure. As-of date: 2026-09-19; market data at the 2026-09-18 close. Key factual claims were checked against the cited primary sources on that date. Conflict-of-interest disclosure: the author holds a long position in NVIDIA (NVDA) and no position in any other company named here, and may buy or sell NVDA at any time. Do your own research.

What was examined. NVIDIA’s fiscal 2026 full-year release (2026-02-25), Q3 FY2026 release, Q1 FY2027 release (2026-05-20) and Q2 FY2027 release (2026-08-26), all primary; NVIDIA’s 10-Q for the quarter ended 2026-07-26 for the share count; the New Constructs SpaceX analysis (2026-05-21) and Fortune’s coverage of it (2026-06-06); SpaceX’s IPO pricing announcement (2026-06-11) and final prospectus (Form 424B4, 2026-06-12); BEA/FRED national accounts; Vitol’s 2022 review; market data for NVDA and SPCX at the 2026-09-18 close.

Fact versus estimate. Base revenue, segment figures, share count, reported growth rates and share prices are facts, each sourced and dated below. Everything downstream of the 50% growth assumption is arithmetic on assumptions and is not a forecast of anything.

The one methodological liberty. New Constructs’ model is a DCF on NOPAT that values the forecast-period cash flows plus a terminal value, with a 23% margin and 50% growth [1]. This article does not reproduce it. To translate the exercise into a revenue-and-multiple model, I add an assumption they did not make: treat the entire $1.75T as compounding to a year-ten value at the roughly 10% annual return Fortune inferred [2], then divide by year-ten revenue: $1.75T × 1.10¹⁰ ÷ $1.078T = 4.21×. That figure is my bridge, not their terminal multiple. A sales multiple also ignores margin structure, which is why Exhibit 3 varies it from 2× to 10×.

What the model ignores. Buybacks and dilution (share count held at 24.15B), capital structure, taxes, currency, cyclicality within the decade, and any GAAP/non-GAAP distinctions.

Reproducibility. Every exhibit is generated by tools/moonshot-model.py. --base-rev 302.97 --price 222.27 --shares 24.15 reproduces Exhibits 1 to 3 (SpaceX per-share rows in Exhibit 2 computed by hand from [15] and [6]); varying --base-rev reproduces Exhibit 4. Substitute your own growth, multiple, discount rate or base and you get your own version of this article, which is rather the point.

Sources

[1] David Trainer / New Constructs, “SpaceX (SPCX): Going Boldly Where No One Has Gone Before,” 2026-05-21 (scenario: ~50% revenue CAGR through 2035, 23% NOPAT margin; DCF of forecast-period cash flows plus terminal value) — https://www.newconstructs.com/spacex-spcx-going-boldly-where-no-one-has-gone-before/ ; methodology: “How New Constructs’ Discounted Cash Flow Model Works” (quoted: “We use the weighted-average cost of capital (WACC) to discount cash flows in each company model”; the reverse DCF solves for the growth needed “to produce a stock price equal to the current stock price”) — https://www.newconstructs.com/how-new-constructs-discounted-cash-flow-model-works/ · [2] Fortune, “SpaceX needs to grow at a rate no company has ever achieved to justify a $1.75 trillion valuation,” 2026-06-06. Quoted: “By Fortune’s estimates, he’s positing that investors will want a total annual return of around 10%”; “In this scenario, SpaceX’s revenues would jump from year-end 2034 to the close of 2035 from $718 billion to $1.1 trillion”; “What’s the precedent for such a vertiginous ramp in just 12 months? It doesn’t exist.” — https://fortune.com/2026/06/06/spacex-ipo-stock-price-valuation/ · [3] Space Exploration Technologies Corp., “Announces Pricing of Initial Public Offering,” 2026-06-11 (primary) — https://content.spacex.com/cms-assets/FINAL_Documents%20and%20Updates/SpaceX_PricingAnnouncement.pdf ; TechCrunch, “SpaceX officially prices shares at $135 in the largest IPO ever,” 2026-06-11; ~13.18B shares outstanding per SpaceX Form 10-Q (quarter ended 2026-06-30) — https://www.sec.gov/Archives/edgar/data/1181412/000162828026052535/spcx-20260630.htm · [4] NVIDIA, “NVIDIA Announces Financial Results for Fourth Quarter and Fiscal 2026,” 2026-02-25 (primary) — https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2026 · [5] NVIDIA, “NVIDIA Announces Financial Results for First Quarter Fiscal 2027,” 2026-05-20 (primary; Q1 FY27 revenue $81.6B) — https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2027 · [6] SPCX market data, close of 2026-09-18 ($152.71; market capitalization ~$2.0–2.1T); CNBC, “SpaceX stock rebounds, closing above $135 IPO price for first time in weeks,” 2026-08-10 — https://www.cnbc.com/2026/08/10/spacex-spcx-stock-ipo-price.html · [7] CNBC, “Tech AI spending approaches $700 billion in 2026,” 2026-02-06 — https://www.cnbc.com/2026/02/06/google-microsoft-meta-amazon-ai-cash.html ; CNBC, “AI boom: Big Tech capital expenditures now seen topping $1 trillion in 2027,” 2026-04-30 — https://www.cnbc.com/2026/04/30/ai-boom-big-tech-capital-expenditures-now-seen-topping-1-trillion-in-2027-.html ; Yahoo Finance on Goldman Sachs’ cumulative FY2025–2030 estimate (~$5.3T), 2026-06-03 — https://finance.yahoo.com/sectors/technology/articles/ai-spending-4-tech-giants-160343367.html ; ValueAddVC capex compilation (big-four 2026 ~$725B; post-Q2 tracking ~$720–745B) — https://valueaddvc.com/blog/big-tech-ai-capex-in-2025-microsoft-google-meta-amazon-and-the-spending-race · [8] U.S. Bureau of Economic Analysis via FRED, US nominal GDP $32,486.1B (Q2 2026, SAAR, second estimate) — https://fred.stlouisfed.org/series/GDP · [9] NVIDIA, “NVIDIA Announces Financial Results for First Quarter Fiscal 2026,” 2025-05 (primary; $44.1B, used in the TTM-through-Q1 comparative) — https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2026 · [10] NVIDIA, “NVIDIA Announces Financial Results for Second Quarter Fiscal 2027,” 2026-08-26 (primary; revenue $96.2B, +106% YoY; Data Center $89.0B; gross margin 75.0%; Q3 outlook $108.0B ±2%, gross margin 74.0%; ~$26.0B returned to shareholders, ~$99.0B authorization remaining) — https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027 · [11] NVDA close of $222.27 on 2026-09-18 (Yahoo Finance; StockAnalysis agrees); shares outstanding 24,147M at 2026-07-26 per NVIDIA’s Form 10-Q — https://www.sec.gov/Archives/edgar/data/1045810/000104581026000075/nvda-20260726.htm · [12] NVIDIA, “NVIDIA Announces Financial Results for Third Quarter Fiscal 2026” (primary; Q3 FY26 revenue $57.0B, used in the TTM sum with Q4 FY26 $68.1B [4], Q1 FY27 $81.6B [5] and Q2 FY27 $96.2B [10] = $302.97B) — https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-third-quarter-fiscal-2026 · [13] Vitol, “Vitol 2022 volumes and review,” 2023-03-20 (primary; “Turnover of $505bn in 2022 (2021 $279bn)”, i.e. about +$226B in one year, driven by the 2022 energy-price spike) — https://www.vitol.com/vitol-2022-volumes-and-review/ · [14] Space Exploration Technologies Corp., final prospectus (Form 424B4), SEC, filed 2026-06-12 (primary; xAI Merger closed 2026-02-02; 2025 capital expenditures: Space $3,832M, Connectivity $4,178M, AI $12,727M; shares outstanding immediately after the offering, before the over-allotment: 7,380,196,910 Class A + 5,695,668,265 Class B (the post-greenshoe count is in [15]); AI segment 2025: revenue $3,201M, loss from operations $(6,355)M; Anthropic compute agreements: ~325,000 NVIDIA GPUs, $1.25 billion per month through May 2029, terminable by either party on 90 days’ notice after the initial three-month period; “Our Market Opportunity”: quantifiable TAM $28.5 trillion = Space $370B + Connectivity $1.6T + AI $26.5T, excluding China and Russia; Terafab (with Tesla and Intel): “we have agreed with Tesla on a general framework for the future development of Terafab”, specific projects subject to separate agreements incl. timelines, milestones and capital expenditures) — https://www.sec.gov/Archives/edgar/data/1181412/000162828026042639/spaceexplorationtechnologi.htm · [15] Space Exploration Technologies Corp., Form 10-Q for the quarter ended 2026-06-30, SEC, filed 2026-08-04 (primary; IPO completed 2026-06-15: 638,888,888 Class A shares at $135.00 “including the full exercise of the underwriters’ over-allotment option”, net proceeds $85,675M after $575M of underwriting commissions and offering costs; shares outstanding as of 2026-07-28: 7,696,293,669 Class A + 5,485,486,276 Class B = 13.18B) — https://www.sec.gov/Archives/edgar/data/1181412/000162828026052535/spcx-20260630.htm