Adobe

On 31 July we published a report on Adobe at $250.41. It said the shares had fallen sixty-four percent while revenue rose sixty, that the decline was a re-rating of durability rather than a deterioration in earnings, and that nothing in the accounts was breaking. We told you to accumulate, to add toward $200, and — the sentence we cared most about — to size the position small, because the ten-year picture was genuinely unknowable.

Since then Adobe named its next chief executive, lost the executive who ran three quarters of its revenue, and reported a quarter. The share price today is $250.50. Nine cents higher. That flatness is the most misleading number in this report, because underneath it the composition of the thesis changed a great deal. A verdict that cannot be checked is a slogan, so here is the ledger first.

One wrong, one half-wrong, six right. The one we got wrong is the one that mattered most, because it was the sentence the whole verdict rested on. And the half-wrong one is more instructive than the wrong one, so it gets its own part.

The dials move accordingly: Moat 7 → 6, Management 6 → 5, Growth 6 → 5. Financial strength and valuation are unchanged at 9. The global score falls from 7.4 to 6.8. The verdict, as you will see, does not change — but the reason for it does, and a reason that changes is worth more of your attention than a conclusion that does not.

In July we set ourselves a test, in public, and I want to hold us to it properly rather than quietly move the goalposts. The sentence was: "The tell will come in about four quarters: if free users keep compounding while paid revenue keeps sliding, it was weakness."

Free users compounded. Creative freemium reached 100 million, up more than seventy percent, from fifty million a year ago; total monthly actives crossed one billion for the first time. And paid revenue did not slide. It accelerated, from 12.0% to 12.7% to 12.9%. By the test we wrote, Adobe passed.

The test was badly chosen. Revenue at a subscription company is a rear-view mirror. It is the recognition of contracts signed in prior periods, released ratably into the income statement, and it will keep rising for several quarters after the selling stops. The number that tells you what was sold is the order book — remaining performance obligations, the contracted revenue not yet billed — and net new recurring revenue. Those are the forward indicators, and both broke.

Read the bottom row twice. Adobe added the same amount of new recurring revenue in the third quarter as in the first — after a quarter in which it added fifty million free creative users and crossed a billion monthly actives. On the call, an Evercore analyst put net new ARR down thirty-six to thirty-seven percent year on year and asked management to explain it. Nobody disputed the number. The incoming chief executive answered that Adobe deliberately routes a portion of its traffic through the free funnel before converting it, and that the goal is to "prove value first."

That may well be the right strategy. It is certainly a deliberate one. But notice what Adobe did with its guidance on the same day: it raised full-year revenue and earnings, and left the recurring-revenue growth target at 10.2%, exactly where it was in June. It raised what is already banked and held what is still to be sold. Managements do that when the near term is safe and the far term is not yet theirs to promise.

So here is the corrected test, and we will be held to this one. Adobe has told us the payoff arrives in 2027 — Narayen's own words, "that will play out, I think, over 2027." The thing to watch is not revenue and not user counts. It is whether RPO growth returns to double digits and whether net new ARR stops shrinking year on year. The interim chief financial officer said RPO should "step up in our fourth quarter, and then remain fairly flat over the next three quarters." That is a specific, checkable promise from the company. We will check it in December, and again in March.

Everything above is easier to see than to read. The first panel is the divergence: what Adobe billed against what Adobe has contracted. The second is the recurring revenue added each quarter, with the portion that was bought rather than sold shaded separately. The third is the funnel and the fraction of it that pays. The fourth is the part almost nobody has connected, and it is the reason we changed our mind about what the risk actually is.

The fourth panel is the one to sit with. Adobe's gross margin fell 1.2 points — and Adobe's Firefly now serves other companies' models: Google's Gemini and Nano Banana, OpenAI's GPT-Image-2.5, Black Forest Labs' FLUX, Runway, Kling, Luma, Topaz, ElevenLabs. Every one of those generations is inference Adobe pays for and, in the free funnel, does not charge for. That is not a moat collapsing. It is a cost of goods appearing where a software company never had one.

And it is not Adobe's problem alone, which is the most useful fact in this report. Figma grew revenue forty-eight percent in its June quarter and posted a negative thirty-two percent GAAP operating margin, with cost of revenue up 117%; its shares fell fifteen percent on the print and are down about fifty-seven percent over a year. Canva cut its own revenue growth expectation by roughly a third, to about twenty percent, explicitly citing the cost of frontier models — and its own backers marked the company down seventeen percent, from $42bn to $34.9bn. The entire creative-software category is discovering simultaneously that generative features cost real money to serve. Adobe is the only one of the three still earning an 88% gross margin and a 35% operating margin while it happens.

The largest single move in the shares during these seven weeks had nothing to do with earnings. On 3 September, after the close, Adobe named Anil Chakravarthy as President and Chief Executive effective 1 December, with Shantanu Narayen becoming Executive Chair after eighteen years. The next day the shares fell 6.73%, touching −7.6% intraday. By contrast, the Q3 earnings print on 10 September cost the stock almost nothing: from that day's close to 16 September the shares are up 0.67%. The market was not frightened by the quarter. It was frightened by the appointment.

Why. Chakravarthy came up through Customer Experience Orchestration and worldwide field operations — the enterprise marketing business, not Creative Cloud. In the same week, David Wadhwani, who ran the creativity and productivity business for about five years, roughly three quarters of Adobe's revenue, and who was widely assumed to be the internal front-runner, announced he was leaving. Narayen thanked him on the call for "over a decade of partnership." Jefferies, which had expected Wadhwani to be the "rational choice," now expects further departures and organisational change.

Set that beside a chief financial officer who has been interim since 15 June, when Dan Durn left for Marvell. Adobe enters its fiscal 2027 planning cycle with a new chief executive who has not run the creative franchise, no permanent finance chief, and no president for the business that pays for everything — while executing the most consequential strategic decision in its history, which is how much of itself to give away.

In July we called this a leadership vacuum and marked management down to 6. That was, if anything, generous. It goes to 5. Two honest counterweights: Chakravarthy is the man who built Adobe's enterprise business into a genuine second leg, and a board that wanted continuity in the creative franchise had the obvious candidate available and chose not to use him. That is a signal about where Adobe thinks its future revenue comes from, and it is not Photoshop.

Here is the thing we did not see clearly enough in July, and it is worth stating plainly because it changes the shape of the risk rather than its size.

A traditional software company has almost no cost of goods. Once Photoshop is written, the ten-millionth copy costs nothing to deliver. That is where an 89% gross margin comes from, and it is why we treated that number as evidence of an intact moat. But a generative feature is not a copy of a program. It is a computation, performed on demand, on hardware somebody has to pay for, every single time a user presses the button. The more successful the feature, the larger the bill.

Adobe has now made two decisions that push that bill up at once. It gave away the entry point — a hundred million free creative users, generating at Adobe's expense. And it turned Firefly into an interface layer over other people's models, paying Google, OpenAI, Black Forest Labs, Runway, Kling, Luma and Topaz for inference it then serves, often free, inside its own product. Both are defensible. Neither is free. The gross margin fell 1.2 points for the first time in eight quarters, and the non-GAAP operating margin fell 232 basis points. Adobe guided to that compression — around 44% for Q4 and 45% for the year — so it is planned rather than a miss. Planned is not the same as costless.

Two events in this window show the strategic exposure that comes with the interface-layer position, and they cut in opposite directions.

On 6 August Adobe launched a single plugin inside ChatGPT carrying more than seventy tools from Express, Photoshop, Firefly, Premiere, Acrobat, Lightroom, Illustrator, InDesign and Adobe Stock — free, with no Creative Cloud subscription required. Distribution to the largest new consumer surface in software, at the cost of teaching people they do not need the subscription.

Then on 10 September, the same day as the earnings print, OpenAI restricted what Adobe may advertise inside ChatGPT: Adobe lost the ability to promote Acrobat Studio and the Firefly image generator, and ChatGPT stopped returning outbound links for queries about photo editors and image generators. Video campaigns were left alone — an asymmetric policy that maps exactly onto where OpenAI sells its own. An Adobe senior director confirmed it. That is the interface-layer bargain in one day: you are welcome inside the platform right up to the point where you compete with it.

And on 16 September, one day before this report, Canva launched ProSuite — Affinity, Cavalry, Flourish and Leonardo bundled and positioned explicitly as a Creative Cloud alternative, with the core Affinity apps free and native InDesign import. The attack on the entry point is now coming from a company that has decided the entry point should cost nothing at all.

In July we wrote that when a management team reduces the granularity of a contested metric, the burden of proof moves to management. That trend continued, and it is worth a short section because it is the reason parts of this report are less precise than we would like.

Gone in Q1 FY2026: Digital Media, Digital Experience and Publishing & Advertising were collapsed into a single reportable segment. Digital Media net new ARR — the number the entire bull-bear argument used to run on — is no longer published, and will not be in the 10-Q either. Adobe now reports two customer groups: Business Professionals & Consumers ($1.91B, +16%) and Creative & Marketing Professionals ($4.65B, +13%).

Gone in FY2026: "AI-influenced ARR," the figure Adobe used to lead with — more than $5 billion exiting fiscal 2025, over a third of the book — has quietly stopped being updated. In its place is AI-first ARR: more than $650 million, growing more than 150%. To Adobe's credit, the new number is the stricter and more honest of the two. It is also 2.4% of the company after three years of the most intense product push in its history, and the growth rate halved from "3× year over year" in Q2 to "more than 150%" in Q3.

Not yet available: the Q3 10-Q had not been filed when this was written. Everything here comes from the press release, the prepared remarks and the call.

And two numbers from our own data feed that we will not repeat to you, because they are wrong or misleading:

First, our feed shows 430 million diluted shares for the year-ago quarter. Adobe's own release says 424 million. We use Adobe's, which makes the reduction 6.8% rather than 8.1%. Second, our feed shows total debt of $6.76bn and net debt of $2.40bn; the borrowings in Adobe's release sum to $6.36bn against $5.64bn of cash and short-term investments, i.e. net debt of about $724 million. The difference is lease liabilities, which our provider folds into debt. Both versions are immaterial against $10.9bn of annual free cash flow, but a reader deserves to know which number they are being given. Third, the headline P/E of about 14× on our page is the GAAP trailing figure; the multiple everyone quotes in the market is on non-GAAP earnings and is closer to 10× this year. Neither is wrong. They are not the same number.

The single most important line is the one that grew 3.1%. Revenue rose 12.9% and GAAP profit rose 3.1%; the gap is margin, and the gap between GAAP profit at +3.1% and GAAP earnings per share at +10.5% is the buyback. Roughly half of the non-GAAP earnings growth this quarter came from there being fewer shares, not from there being more profit. That is not a criticism — retiring 6.8% of a company at nine times free cash flow is an excellent use of money, and we said so in July. It is simply a different engine from the one a July reader was underwriting.

Two things are worth saying about this table, one dull and one interesting.

The dull one: on every cash-based measure Adobe is cheaper than it was in July, because free cash flow rose while the price did not. Eleven percent of your money back in cash every year, from a business with an 88% gross margin, no meaningful debt and a quarter of its own market value still authorised for repurchase, is not a normal thing to be offered.

The interesting one is the dispersion. Between 11 and 15 September, after the same quarter, from the same data: Goldman Sachs rates it Sell with a $200 target. Bernstein rates it Buy with a $373 target. That is one house saying the shares are worth twenty percent less than today and another saying they are worth forty-nine percent more, on identical information. Morgan Stanley is at $240 Underweight; JPMorgan cut to $315 but stayed Overweight; RBC is at $315 Outperform; BofA at $220 Underperform. Not one firm changed its rating in the seven weeks we are covering — every single action was a price-target revision with the opinion left where it was. When the professionals cannot agree within a factor of nearly two and none of them will move their published opinion, you are not looking at a valuation dispute. You are looking at a disagreement about whether a business model survives, expressed in the only vocabulary the sell side has.

That, incidentally, is precisely what we said in July, and it remains the honest description of what you are buying: a magnificent set of accounts attached to a genuinely unknowable decade.