The Fade Line

Everyone argues about Palantir’s multiple. The multiple is arguing about one number. Tonight, that number gets a new reading.

Palantir reports second-quarter earnings tonight, after the close. By tomorrow morning the argument will be everywhere it always is: the stock is “obviously” too expensive at fifty-seven times revenue, or the doubters have “missed the point” for the fifth straight year. Both camps quote the same multiple. Neither one usually says what the multiple actually asks for.

A price is not an opinion; it is an invoice. This note runs the valuation machine backwards — instead of guessing what Palantir is worth, we compute the revenue future its $292 billion enterprise value already charges you for, and then ask how plausible that future is. The answer compresses the entire bull–bear fight into a single number we call the fade rate: the fraction of this year’s growth that survives into next year. Everything else — the multiple, the AI narrative, tonight’s print — is commentary on that number.

A $292 billion demand letter

Start with what you are buying. At $125.24 (10:30 ET this morning, 3 Aug 2026), Palantir’s 2.40 billion shares are worth $300.2B. Net of about $8.0B in cash, the market is paying $292.2B for the business itself. That business produced $5.2B of revenue over the last four reported quarters — so the price is 56× trailing sales, roughly 132× trailing earnings. For scale: the most expensive of its high-growth software peers trades below 40× sales.

A multiple that high is not a mystery; it is arithmetic waiting to be done. Suppose you buy today and hold for ten years, and suppose that in 2036 the market pays a still-generous 10× sales for the business. Then the revenue Palantir must deliver is fixed by two choices only: the return you demand along the way, and that exit multiple. The whole grid:

Read the gold cells. To earn an ordinary 10% a year, Palantir must grow revenue 30.7% every year for a decade — $5.2B today becoming $75.8B by 2036, at which point it would still need a buyer at ten times sales.

The cash machine is real — and it barely moves the answer

To be clear about what the skeptics get wrong: Palantir is a genuinely exceptional business, and the numbers in this note assume it stays one. Trailing GAAP net margin is 43.7% — among the highest of any large software company — and reported free cash flow of $2.7B is 51.5% of revenue (flattered by roughly 14% of revenue paid in stock rather than cash; we model a 40% steady-state cash margin instead). Fiscal 2025 revenue split $2.4B government / $2.1B commercial, and the government franchise — two decades of security clearances, accreditations, and embedded deployments — is about as close to a moat as enterprise software gets. Return on invested capital runs north of 20% (22.3% trailing).

Here is the uncomfortable part. Run a full discounted-cash-flow model backwards — every future year’s cash flow valued at a 10% discount rate, a terminal value beyond year ten — and ask what constant growth rate makes the math equal $292.2B. The answer is 33.1% a year. Now torture the margin assumption: at a 35% cash margin the required growth is 35.0%; at 45% it is 31.4%. A ten-point swing in profitability — the entire distance between “very good” and “best ever” — changes the demanded growth by about three points.

This is not a margin-priced stock. It is a growth-priced stock. Every dollar of the debate belongs on one question: how fast does 80% growth decay?

85% doesn’t fade politely — or does it?

What makes Palantir a hard case — and this is the part the “it’s obviously a bubble” crowd skips — is that the growth is not slowing. It is accelerating, and has been for three straight years: +24% in 2022, +17% in 2023, +29% in 2024, +56% in 2025, and +85% in the most recent reported quarter. Tonight’s consensus asks for +80% more. Nobody sane extrapolates 80% for a decade. The live question is the fade rate — each year, what fraction of the growth rate survives?

Solve for the fade rate at which the discounted cash flows are worth exactly today’s price and you get ×0.78 — growth may lose at most about 22% of itself each year. That path takes Palantir to $80B of revenue by 2036 (31% a year for ten years), still growing 8% in year ten. Slower decay than that and today’s buyer wins; faster and the buyer loses. The entire disagreement about a $300 billion company balances on that one decimal.

Honesty requires the chart’s left half be read both ways. Bears will note that Palantir has already faded once — growth collapsed from the 40s to 17% by 2023 under exactly the “law of large numbers” logic bears use today — and bulls will note what happened next: the AI platform cycle re-accelerated it to 85%. A company that has already broken one fade curve cannot be assumed to obey the next one. That cuts in both directions: fade curves are not destiny, and 2023 proves growth can halve in two years when a narrative pauses.

The club it is being asked to join

How rare is a ~30% compound decade from a $5B revenue base? The modern reference class is short. From comparable starting revenue, Amazon compounded roughly 29% a year from 2004 to 2014 (about $6.9B to $89B); Google roughly 28% from 2005 to 2015 (about $6.1B to $75B); Meta roughly 29% from 2014 to 2024 (about $12.5B to $165B); and Nvidia — the outlier the bulls lean on — has compounded faster than that since 2016 on the same AI cycle Palantir is riding. That is approximately the complete list of US large-caps that have done what Palantir’s price now requires of it. The base rate is not zero. It is also not a coin flip — it is a handful of the greatest business runs in market history.

Note what membership bought those companies’ shareholders: all four entered their decade at far lower multiples of sales than 56×. Palantir is being asked to match their execution while starting from a price that already banks most of it.

The exit door is narrower than the entrance

Every path in this note ends with someone in 2036 paying a multiple for a bigger, slower Palantir. What do markets actually pay for excellent software companies after the hypergrowth years? The comp set answers:

ServiceNow is the destination that should concentrate the mind: a magnificent, ~24%-growth, high-margin platform — at 9× sales. CrowdStrike, the most expensive name in the set, earns 39× while growing 26%. Our base case hands 2036-Palantir a 10× exit — richer than ServiceNow today, justified only if it is still growing near-double-digits at ~$80B scale. The “conservative exit multiple” in most Palantir bull math is quietly the most generous line in the model.

Tonight is not an earnings report. It is a fade-rate observation.

Consensus for the quarter: revenue about $1.81B (+80% year over year), EPS $0.33. Palantir’s recent record against the bar is close to perfect — five beats and a meet in the last six quarters, with EPS beats averaging over 20%. And yet the stock fell on three of those six perfect prints, including a 12% drop in May 2025, an 8% drop last November, and 7% this May:

That pattern is this note’s thesis in miniature: the market is not grading Palantir against consensus; it is grading the print against the demanded path. A beat that implies a faster fade is a miss. Specifically, tonight maps to the fade line as follows: a revenue print near $1.86B (+85%, no deceleration at all) is evidence for the ×0.85-and-slower-fade world; the consensus $1.81B (+80%) is a first tick down the curve, consistent with everything from ×0.85 to ×0.78; anything under ~$1.78B (+77% or less) is the first data point that decays faster than the breakeven line can tolerate. The full-year guide matters more than the quarter — consensus sits at $7.73B for 2026, and a raise through ~$7.9B would be the strongest slow-fade evidence available tonight.

The options market has already voted on the volatility, if not the direction: at ±11.4%, tonight’s implied move is priced almost exactly at the six-print historical average of ±10.9% — roughly ±$34 billion of market value riding on the reading. Wall Street analysts, for what it is worth, sit far up the bull side of the line: the consensus price target is $171 (median $182) — 36% above the market, even after the stock’s 49% slide from last November’s $207.52 high to June’s $106.37 low and back to $125 today. The street, in other words, already assumes the slow fade. The market, at $125, is priced a notch below the street and a notch above the math.

Where we put the line: fair value $110

We weight three fades. The bear case (30%) is ordinary software gravity — the 2023 deceleration revisited without a second AI wave, decaying to GDP-plus by year ten. The base case (45%) is the glide path already embedded in the 23-analyst consensus through 2028, then normal maturation. The bull case (25%) is the AI-platform supercycle persisting — commercial and international compounding hard enough that Palantir is still growing ~19% in 2036, having joined the Amazon/Google/Meta club outright.

Probability-weighted fair value: $112 per share, about 10% below tonight’s $125.24 — with an enormous, honest range ($63 to $195) that is itself the finding: this is a security whose value is almost entirely a bet on one behavioral parameter of one income-statement line. The math is also acutely rate-sensitive — the base case alone swings from $121 at a 9% discount rate to $73 at 12% — so the fade line moves with the bond market even if Palantir executes perfectly.

What would move our line? Toward the bulls: two consecutive quarters holding 80%+ growth (fade ≥ ×0.9 evidence), a 2026 guide above $7.9B, or commercial revenue overtaking government while both accelerate. Toward the bears: any quarter printing below +70%, a flat guide, or US government budget friction — the segment that is 54% of revenue answers to appropriations cycles no company controls. The risks outside the model are the standard ones for a story stock at 56× sales, all amplified by the multiple itself: single-narrative concentration, ~2%/yr dilution from stock compensation, key-person dependence on an idiosyncratic CEO, and the demonstrated capacity of this exact share register to cut the price in half in seven months on no fundamental break, as it did from November to June.

The knife-edge, restated once, in plain language: at $125, you are not paying for what Palantir is. You are paying for eighty-cents-on-the-dollar persistence of the fastest growth any $5 billion-revenue company is currently producing, for ten consecutive years, ending in a $80 billion company that must still command a premium multiple. It might happen — it has happened, four times in the modern record. Price it accordingly, and watch tonight’s year-over-year number, not the EPS beat.