Adobe doesn't sell software anymore. It sells compute by the sip.
That is the only line in its latest earnings that deserves a trader's attention. Not the user growth. Not the "AI momentum" language the investor deck leans on. The mechanical fact buried underneath every headline: Adobe has converted generative AI from a feature into a metered commodity, priced per unit of inference and capped by an account-level credit ledger.
Firefly generations draw from that ledger. Vector operations draw from it. Video renders draw from it. Exhaust the allotment and the meter keeps running — you pay.
When a company stops selling a capability and starts selling the meter that measures it, the product has changed category. Adobe is no longer a creative tools business with an AI feature. It is a compute reseller with a creative interface.
I have watched this exact transition before. Not in SaaS. In crypto.
Anyone who ran a liquidity mining position in 2020 recognized the shape of it immediately. The protocol subsidizes activity with emissions, watches TVL inflate, then discovers that the moment the subsidy stops, the users evaporate. Adobe is running the inverse experiment — instead of paying users to consume compute, it charges them for each unit. Same fundamental question underneath both models: what does one unit of execution actually cost, and who absorbs the loss when the volume spikes?
Most analysts are reading Adobe's answer as a revenue story. They are wrong to. It is a cost story wearing revenue clothes.
Context first.
Adobe sits at roughly a $200 billion market capitalization with annual revenue near $20 billion and research-and-development spending above 20% of that figure. It owns the file-format chokepoints of the creative industry — PSD, AI, PDF — the same way a settlement layer owns a chain. Its installed base is the highest-concentration professional creative population on earth. When Adobe changes a pricing mechanic, the entire creative software sector reprices within eighteen months. That leverage is the reason this earnings cycle matters beyond one balance sheet.
The company's AI strategy has moved through three phases. Phase one was capability demonstration — Firefly as a showcase, priced into the subscription to reduce churn. Phase two was integration — Firefly embedded across Photoshop, Illustrator, Premiere, InDesign, so that using the AI became the default path rather than a separate product. Phase three, which is what this earnings cycle confirms, is monetization through metering. Generative Credits are the instrument of that phase.
The mechanism is worth dissecting because it is the template every AI business will copy. A credit is not a software license. A license is a right to use a tool indefinitely. A credit is a consumable — a unit of computation that is spent and gone. Adobe has quietly migrated from selling rights to selling consumables. That is the difference between a subscription business and a utility business, and the market has not fully priced the valuation implications of that shift.
Here is where the crypto parallel stops being metaphorical and becomes load-bearing.
Every serious AI token project I have audited over the past three years runs on the same underlying tension: token emissions look like revenue, but they are actually a liability denominated in compute. A project mints tokens to pay for GPU time, or subsidizes users to generate content, and books the gross activity as growth. On-chain, the ledger is transparent — you can watch the emission schedule against the actual compute bill, and the spread between them is the real solvency metric. Adobe's generative credits give traditional SaaS the same visibility, and the picture is uncomfortable.
Inference is the new gas fee. It is variable, it scales with usage, and it is invisible until it isn't. Adobe's generative AI carries a marginal cost per generation that a static software license never had. Every time a user hits "generate," the company pays for GPU cycles, electricity, and cooling. The subscription price was built for software margins — 80%+ gross. The credit price has to be built for compute margins, which are structurally lower and far more volatile.
That gap is the entire story. If Adobe prices credits to recover inference cost, users feel nickel-and-dimed and migrate to flat-rate competitors. If it prices credits below cost to preserve the growth narrative, gross margin erodes and the earnings quality deteriorates quarter by quarter. There is no clean exit from that wedge. The company is trying to have a utility business with a software business's margin profile, and those two things do not coexist indefinitely.
I saw the same wedge in DeFi's liquidity mining era. Protocols subsidized TVL to manufacture growth metrics, then faced an impossible choice: cut emissions and lose the numbers, or sustain emissions and bleed treasury. Most chose to sustain until the treasury was empty. The ones that survived were the ones that found genuine, non-subsidized demand before the subsidy ended. Adobe's clock is now running on that same question — is there demand for AI generation when it is priced at true cost, or only when it is bundled into a subscription that hides the meter?
The creative-industry impact is where this gets granular, and where I can speak from direct observation. I have run automated content pipelines for trading dashboards and marketing assets. The tasks Firefly now performs — background removal, batch color correction, automated vector tracing, template generation — used to consume entry-level design labor. That work is gone, and it is not coming back. My estimate from watching comparable automation roll through trading desks is that entry-level creative roles face displacement in the 50% range, while mid-to-senior designers are enhanced rather than replaced — call it 30 to 40% efficiency gain, with the core creative judgment still human-dominated. The visible labor-market impact lands within twelve to eighteen months, not five years.
This is the same pattern that hit junior data analysts when I automated my own reporting stack. The tool does not fire the junior. The tool makes the senior three times faster, and suddenly nobody needs to hire the junior. The displacement is a second-order effect of a productivity gain, and it is invisible in any single quarter's earnings print. That is precisely why the market misses it.
The competitive geometry makes the margin problem worse, not better. Adobe is the incumbent defending a subscription fortress against three distinct attacker classes. Canva attacks on ease-of-use, wrapping AI in a drag-and-drop interface that requires no professional training. Midjourney attacks on raw generative quality, where Adobe has no aesthetic advantage. And the open-source ecosystem — Stable Diffusion and its derivatives — attacks on price, which is to say free, which is to say zero marginal cost for anyone willing to run their own inference.
That last threat is the one Adobe cannot answer with a credit system. You cannot meter a commodity that your competitors give away. If a user can spin up an open-weight model locally and pay electricity instead of credits, Adobe's entire monetization mechanic loses its grip on the price-sensitive segment — which is the majority of the market by volume, if not by revenue.
And this is exactly where the crypto-and-AI convergence gets interesting for anyone holding DePIN tokens or decentralized compute allocations. The open-source inference layer is being subsidized right now by token emissions across a dozen decentralized GPU networks. Those networks are racing to drive the marginal cost of inference toward zero, funded by token holders who believe compute demand will eventually exceed the emission liability. Adobe's credit pricing is the ceiling those networks are trying to undercut. If decentralized inference becomes reliable enough for production workflows, Adobe's meter becomes a premium tax on convenience rather than a defensible moat.
The copyright overhang complicates everything downstream. Adobe's models are trained substantially on its own licensed Stock library, which it frames as a compliance advantage. That framing is a bet. If current litigation establishes that training on licensed-but-not-consented assets is still exposure, Adobe's training-data advantage becomes a training-data liability, and the cost of retraining an entire model family lands directly on the gross margin already squeezed by inference. I audited the Celsius balance sheet in 2022 by comparing on-chain reserves against off-chain promises. The lesson transferred cleanly: the market prices the promise until the ledger forces the correction. Adobe's ledger is its inference bill, and it is not on the front of the investor deck.
So here is the contrarian read on this earnings cycle.
Every analyst covering Adobe will focus on adoption — how many users touched Firefly, how many credits were consumed, how engagement grew. Those are activity metrics, and activity metrics have been weaponized by every subsidized ecosystem I have ever traded against. The number that actually matters is buried and rarely disclosed directly: the ratio of credit consumption to incremental paid conversion, measured against the disclosed or implied inference cost. If credits are consumed heavily by users who never upgrade or pay overage, Adobe has built an engagement engine that burns compute without converting it. That is a subsidy dressed as a product.
Watch for the free-tier conversion rate specifically. Adobe's foundational Firefly features lowered the barrier to entry by design — which is free user acquisition, until the free allotment runs out and the conversion rate turns out to be single-digit. The optimistic case, the one the deck prefers, is that Adobe becomes Office 365 with Copilot: seat expansion plus price expansion. The pessimistic case — the one the ledger supports until proven otherwise — is that Adobe has trained an entire market to expect free generation and is now trying to sell them a meter they already associate with zero cost.
I shorted Celsius because the on-chain reserves and the off-chain promises could not both be true. I am not shorting Adobe — the fortress is real and the cash flow is enormous — but I am treating its AI revenue as suspect until the credit ledger balances against the compute bill. The pattern of a subsidized metric masquerading as organic demand is one I have paid tuition to recognize.
The forward-looking tell is not in this earnings cycle at all. It arrives when Adobe either opens Firefly's API to third-party developers with transparent per-call pricing — confirming it is running a genuine utility model — or quietly reabsorbs the credits into flat subscription tiers, admitting the meter was a marketing construct. The first path means Adobe accepts lower margins in exchange for a real platform. The second means the AI monetization story was never a story.
A compute business and a software business cannot wear the same valuation. Watch which mask Adobe keeps when the credits stop being novel — and watch whether the decentralized inference networks undercut the meter before the meter ever learns to defend itself.
