How the Latest AI Leap Is Disrupting Business Models

Full analysis

Two stories in this episode point the same direction, and if you run a software business or buy software, they are the ones that matter. A developer used Claude Opus to rebuild Photoshop as a free substitute, and Anthropic told pre-IPO investors it expects to go from $16.8B in Q3 revenue to a $110B annualized run rate by year-end. One says the cost of copying software just collapsed. The other says the company selling the copying tool is about to triple. Everything else here (CoreWeave's bonds, ASML's supply crunch, the OpenAI spy) is real but orthogonal.

The thing actually being decided for the reader: how much of your software spend, and your software moat, survives a world where a model can reconstruct a product's functionality cheaply. That is hard to undo once customers taste a free clone, and nothing in this episode sets a hard deadline except the next model release.

The Skeptic. The Photoshop clone went viral and dropped Adobe 3%. Three percent. The market looked at "free Photoshop built by AI" and shrugged off 97% of Adobe's value. That tells you the clone is a toy, not a business. A clean-room rebuild documents features; it does not ship color management, 30 years of file-format edge cases, enterprise admin, or the support contract that is the actual product. Chamath Palihapitiya saying "IP and software was rendered worthless two days ago" is a podcast line, not a P&L. Bank of America hired 4,000 people and spent hundreds of millions patching holes a frontier model exposed. Capable models raise your costs before they lower anyone's.

The Researcher. Watch the names, because the episode is sloppy with them. "Fable 5.5," "Astra," "Opus 5.5," "Mythos," "Muse," "Genie" are a mix of real products, rumored releases, and transcription guesses. Anthropic's numbers deserve the same scrutiny. $16.8B is gross revenue before Anthropic's share goes to Amazon and other compute partners, and the 85% gross margin and 11% operating margin are "adjusted," meaning they exclude training costs and stock comp. Strip training out of an AI lab's margin and you have measured everything except the expensive part. The $110B target rests on one unshipped model. That is a forecast, not a result.

The Open-Source Advocate. The decompile-and-rebuild story is the open-weight thesis arriving by the back door. If a model can read a product's behavior and reconstruct a working substitute, the model itself is the great equalizer, and it does not have to be Claude. Qwen, Llama, and Mistral are close enough on code that the clone economy does not depend on any single lab. That cuts against the Anthropic bull case: if copying software becomes a commodity capability, the margin lives in distribution and trust, not in whoever's model did the decompiling this week. The tool that destroys software moats has no moat itself.

The Compute Pragmatist. ASML is the real ceiling and it got buried under the gaming talk. ASML makes the only machines that print leading-edge AI chips, it cannot meet demand for at least two years, and customers are reserving capacity into 2029 and 2039. That is a physical limit no amount of model capability routes around. Now put CoreWeave next to it: borrowing at 9% to earn 5%, bonds at 80 cents, the CEO selling stock every week since April while telling staff the market "misunderstands" the business. The constraint on AI is not ideas. It is lithography machines and the debt stack of the people renting you GPUs.

The Enterprise Buyer. If you sign software contracts, read the Adobe story correctly. The clone proves that standalone tools with high prices and thin switching costs are exposed. The product you cannot clone is the one wired into your workflows, your audit logs, your SSO, your compliance posture, and a support line that answers at 3 AM. Bending Spoons is the clearer read than the Photoshop stunt: Berkshire put in $330M to back a company that buys tired software (AOL, Vimeo, Evernote, Airtable) and uses AI to cut maintenance cost. That is the durable play. Not "AI kills software," but "AI makes old software cheap to run."

Where they split. The Skeptic and the Open-Source Advocate disagree on whether the Photoshop clone means anything: a toy that moved a stock 3%, or the first visible crack in software pricing. They are both right about different time horizons. The Researcher and whoever is bullish on Anthropic's IPO disagree on the $110B number: a forecast on one unshipped model versus the fastest revenue ramp in software history. And the Compute Pragmatist quietly undercuts everyone, because ASML's two-year wall and CoreWeave's 9%-cost-5%-return math say the binding constraint is physical supply, which neither the clone bulls nor the Anthropic bulls have priced.

What it hinges on. One belief: does model capability actually reconstruct working, shippable software, or only a demo that impresses on Twitter and dies on contact with enterprise requirements? If you run a software business, the thing to test is boring and specific. Take your three highest-margin, lowest-integration products and have a model try to clone the top ten customer workflows. If it gets past a demo, your pricing is exposed and you reprice before a free clone does it for you. If it stalls on the integration, support, and compliance layer, you have your answer and your moat. Either way you learn it before your customers do.

Prediction: Adobe's Creative Cloud subscription revenue, reported in its Q1 fiscal 2027 earnings in March 2027, will be higher than the prior-year quarter.

Confidence: Medium. Clean-room clones hit the easy layer; the moat is everything above it.

Why: A developer cloned Photoshop's core features with Claude Opus and Adobe stock fell only 3%, which means the market judged the clone a demo rather than a substitute for color management, file-format support, enterprise admin, and the support contract that is what most Creative Cloud seats actually pay for. Chamath Palihapitiya's "software IP is worthless" line is the loud version of a claim the price action already rejected. The way this prediction fails is if a funded competitor ships a real, supported clone and enterprises switch inside two quarters. The clean-room technique alone cannot get there because it reproduces behavior, and enterprises buy operations, compliance, and account relationships on top of behavior. Standalone consumer tools are exposed; a deeply integrated subscription base does not churn on a viral clone in one quarter.

Revisit by 2027-03-31: We're right if Adobe reports year-over-year growth in Creative Cloud subscription revenue in its Q1 FY2027 results. We're wrong if that line comes in flat or below the prior-year quarter.