The Moat and the Multiple
In November 2021, Adobe traded just under $700. At the July 15, 2026 close, $224.56. In between, the company did not shrink. It grew.
Revenue per share rose from about $33 to about $56. Earnings per share rose from $10.02 to $16.70, up two-thirds. The share count fell from 481 million to 427 million by fiscal 2025, the last full year on file; the difference was retired through repurchase. Operating margin held where it started, in the mid-thirties. By every line an owner reads, the business is larger and more profitable than it was at the top.
The price fell by two-thirds anyway. Every dollar of the decline came out of the number the market would pay for a dollar of earnings: roughly seventy times profit at the top, roughly thirteen at the July close. Falls of that kind, all multiple while the earnings rise, are uncommon.
The record
The business first, in its own numbers, with no price attached; a quotation is not a fact about the company. Suppose the exchange had closed in November 2021 and reopened on July 15. An owner reading only the filings would conclude his company had done splendidly, because it had. Everything that went wrong lives in the one number the filings do not carry.
In 2016 Adobe did $5.9 billion of revenue and earned $2.32 a share; last year, $23.8 billion and $16.70 (the record). Gross margin has sat between 85% and 89% throughout. Operating margin widened from 25.5% to 36.6%. The return on the capital the business employs has run at a median of 33% across the record and cleared 15% in nine years of ten. Last year $7.1 billion of net income became $9.9 billion of owner earnings; the cash runs ahead of the accounting.
Diluted shares fell from 504 million to 427 million. Across the years where the filings report a count, $45.7 billion went to buy in 128 million shares, an average of $357.36 each; the fiscal 2025 count is assembled from the four quarterly purchase tables its filings carry, since the company stopped tagging an annual figure that year. The price paid ran from $103.37 in 2016 to $548.61 in 2021, and the heaviest year was fiscal 2025, when $11.3 billion went out at an average of $366.27. The year before, $9.5 billion went at $542.86. The stock closed at $224.56 on July 15. A repurchase is a purchase, and it answers to the same arithmetic as any other; the price paid is as much a part of the record as the shares received.
The window that matters is narrower: the November 2021 peak to the July close. Over that stretch, revenue per share rose about 70%, earnings per share 67%, margins held, and the share count fell another 11% through fiscal 2025, standing at 403 million by the quarter ended May 2026. Nothing in the operating record explains a two-thirds fall in the price.
The mechanism
Of the two terms in a price, only the earnings are audited. The multiple is set by the market alone, and the whole of Adobe’s decline lives there. It compounds two judgments: the price of money, and duration, meaning how many years of earnings still lie ahead and how surely they arrive.
Duration has a plainer name. A moat is nothing but the power of a business to go on earning after competition arrives to take the earnings away, and what a multiple prices, once the price of money has taken its share, is that power. The moat is the claim, the multiple its price, and the price is revised daily whether or not the claim has changed.
At seventy times, much of what the buyer paid was for years past 2030 that nobody, including Adobe’s own managers, can describe. A multiple that high is faith that the moat has no visible end.
Part of the seventy, though, was never about Adobe. In late 2021 money was nearly free, and free money makes far-off earnings look nearly as good as present ones, so the whole growth-software shelf wore multiples it could not have worn at any other price of money. Then money stopped being free, and the group’s multiples came in together, Adobe’s with the rest. How much of the fall was the price of money and how much was doubt about the moat cannot be split exactly; no such instrument exists. But the shelf itself is a rough control. What a name lost beyond the group, after its earnings had risen and rates had done their work, is the market’s answer on its moat alone. When that part keeps compressing, the market is saying it no longer knows how long the good years last.
Revenue in the quarter ended May 29, 2026 came in at $6.62 billion, ahead of the company’s own target of $6.43 to $6.48 billion, and the subscriptions built on its AI products passed $500 million by the company’s own count, tripled in a year. The price did not follow.
The market, Graham said, votes in the short run and weighs in the long run. Nearly five years is a long count of votes.
Adobe, in its own words
The fair way to hear both cases is to let Adobe argue them itself, in its most recent annual report, under the signatures of people who can be sued for lying. It argues both.
Here is the toll-bridge case. Firefly, its generative engine, is “commercially safe… trained on data Adobe has the rights to use”; every output carries a Content Credentials “nutrition label”; enterprise and teams customers “are eligible to obtain an intellectual property indemnification” for AI content. That is the assurance a big company’s lawyers require before generative art can be put to commercial use. If the moat is real, it lives there: not in the pixels, which anyone’s model can now make, but in the indemnity, the permission to use the work without being sued. The stakes are not a product line. Digital Media is about 74% of revenue, and Firefly lives inside it.
Here is the wasting-asset case, from the same document, in the risk factors, where a company must name what could go wrong. Its AI innovations may not “be successful, adopted or monetizable”; “consumer reception and monetization pathways are uncertain”; it is investing in AI “without assurance that our customers and users will adopt them,” and it warns the cost “could adversely impact our margins.” Adobe now lists “AI inferencing costs” by name among the costs of delivering its own software.
Two facts sit outside the boilerplate. First, the slowdown is real and partly hidden. Adobe’s guidance calls for recurring revenue to grow about 10%, but roughly $480 million of that comes with the Semrush acquisition; strip it out and the organic rate is closer to 8%. Second, and cutting the other way, Adobe has done the one thing a toll bridge can do when a faster road opens: charge the road. Firefly is now a multi-model product. Adobe sells access to its rivals’ engines, Google’s and OpenAI’s among them, inside its own workflow, and meters them by the credit. It turned the disruptors into suppliers.
Which is the truer picture of 2030? The filing does not say, because it cannot know, and neither can we.
Two histories
These falls end in two very different places.
Sometimes the market was wrong, and the fall was the opportunity. Apple in 2013 traded near nine times earnings on a story that the iPhone had peaked and the magic had died with its founder; the fear was wrong, and the earnings compounded for another decade. Meta in 2022 fell to nine times on “the metaverse is a bonfire and TikTok is eating them”; the business was never in the danger the price implied. A new technology has killed some franchises and spared others, and the volume of the fear says nothing about which. The automated teller machine was supposed to end the bank teller; instead the count of tellers roughly doubled between 1970 and 2010, because the machines made branches cheap enough that banks opened far more of them. A moat rooted in work people still need done can absorb a new technology, and sometimes swallows it.
And sometimes the market was right, the moat was real, and it melted anyway. Encyclopædia Britannica was the finest object in its category for two centuries. Einstein and Freud were among its contributors. It sold door to door to parents who wanted the best for their children, some 117,000 sets a year at the 1990 peak, near $650 million in revenue. Then Microsoft pressed a lesser encyclopedia onto a disc that cost about a dollar-fifty to make and bundled it free with a new computer. The moat did not narrow. The format it guarded stopped being worth anything, and Britannica’s revenue fell by half in three years.
The public version of that story, the one with a stock chart, is Kodak. Kodak held eighty percent of American film, invented the digital camera in its own laboratory in 1975, and then spent nearly four decades being killed by it. The market trailed the melt the entire time: Kodak still traded above $20 in 2007, three years after digital cameras outsold film and more than thirty years after its own laboratory built the thing that was killing it; by 2012 it was at sixty-five cents, in bankruptcy. By every statistical test it was cheap the whole way down. Part of the reason a market runs late on a melt is not in the numbers. The holder who has once pronounced the stock cheap finds every lower price an improvement in his argument.
Even the happy endings ran long. Cisco was the most valuable company on earth in 2000, at two hundred times earnings. Earnings per share went on to rise sevenfold, and the stock still spent the next quarter-century below its peak; the error was the price, and the multiple took twenty-five years to forgive it. Microsoft tripled its earnings between 2000 and 2013 while its stock went nowhere, a lost decade of pure multiple compression from a bubble start. When the multiple came back, the toll bridge was still standing, thirteen years on.
In these histories the multiple ended up where the moat was; the only variable was how long the trip took, and it was rarely short. A multiple is the market’s guess at which kind of business it is holding, and a move from seventy times to thirteen is that guess crossing from one column toward the other. Whether the guess is right is another matter. Adobe sits somewhere inside these histories, and no number on its scorecard, or anyone’s, says where.
The peers
Adobe’s peers took the same fall at the same time, and the market priced the group as a single thing. The records are nothing alike.
Side by side on owner economics, the spread is wide. Intuit is nearest to Adobe, a 26.1% operating margin and a median 22% return on capital. ServiceNow reports a 13.7% operating margin while collecting $5.4 billion of cash from operations on $13.3 billion of revenue. Salesforce has worked its operating margin to about 20% on revenue near $42 billion, a repair recent enough that the market may still be pricing the old company. The screen files a fortress, a cash machine, and a repair job under the one word software.
“Software got cheap, buy the group” is the lazy conclusion. The compression was sector-wide, but the answer to melting or misunderstood comes name by name, from reading each one.
The pass
At the July 15 close Adobe stands at thirteen times the earnings it just grew by two-thirds, and at the May 29, 2026 balance sheet it held $5.6 billion of cash and investments against $6.6 billion of debt, the Semrush purchase having tipped the balance; the whole of the debt could still be retired out of a single year’s owner earnings. I do not know whether this is Apple in 2013 or Kodak in 2007. Neither does anyone who says otherwise with confidence; a verdict is what the seller of verdicts is paid to produce.
Both cases are on file. The indemnity can be read in the annual report. No document anywhere records how long the moat will last. The slowdown is real, and the acquisition masks part of it. The toll-taking is real too, and may prove temporary. And the melt case carries a standing warning: Kodak’s price trailed the facts for years, and the multiple that was already low is what reassured its holders on the way down.
The repurchase record belongs in this file. The buyer with the most information about the moat paid $357.36 a share on average across the record, and $542.86 as recently as 2024, against the July 15 close of $224.56.
A margin of safety is the difference between two numbers, and here one of the numbers is missing. What Adobe is worth turns almost entirely on how long the moat holds, and that is the question no one can answer. No value can be appraised here, so no margin can be computed; thirteen times, standing alone, is only a number. What any price demands of the future can be computed; no arithmetic reaches whether the future will oblige. Graham drew the line with a lawyer’s care: an operation that cannot be carried through analysis to a conclusion is not an investment, whatever the buyer calls it. A purchase here is a speculation on the life of the moat. Nothing forbids a speculation. Graham asked only that it be called one.
Munger kept three baskets: in, out, and too hard. Most honest work ends in the third, cheap and great things included, whenever the deciding question cannot be answered at any price. Here the deciding question is whether a two-century-old kind of moat survives a real technological turn. Buffett took the discipline from Ted Williams, who carved the strike zone into seventy-seven cells, knew he hit .400 in the middle and .230 at the low outside corner, and let the corner pitch go by even though it was a strike. In investing there are no called strikes; a man can stand at the plate all season and lose nothing until he swings and misses. Adobe at thirteen times, with a moat whose far side no one can see, is the pitch at the low outside corner. It goes in the too-hard pile. Not “avoid,” which is a verdict. Too hard, which is a confession.
One test remains, from the other side: if the melt had already begun, Adobe’s record as it stands would look very much like this one. A presentable record is what the early years of a melt look like. Britannica’s revenue held until the disc shipped, and Kodak looked cheap on the figures for a decade while the franchise died underneath.
What would move it out of the pile is not the price; thirteen times would be a fair multiple for a wonderful business and no bargain at all for a melting one, and no multiple settles a question about quality. The answer will arrive in the filings, one quarter at a time: whether recurring revenue still grows once the bought revenue is stripped out; whether Firefly’s credits earn money; whether pricing holds per seat or slides toward metered use if agents take over the work seats used to do; and whether any of it stays checkable at all, now that Adobe has folded its reporting into a single segment and some of the needed numbers are no longer published. Until the filings answer, my bat stays on my shoulder.
This note's conclusion is entered, verbatim and permanently, in the Ledger.