Selling Intelligence Was the Warm-Up For AI. Owning Vertical Markets May Be The Play

The most profitable move in artificial intelligence may no longer be selling intelligence. It is selling a model to a company, wiring it into that company’s most valuable work, learning the business from the inside, and then shipping the product that competes with it.

In a previous piece for Inc42, I argued that the black box was losing the boardroom, and that enterprise AI built on the customer’s own terms was the opening worth chasing. The months since have made the argument more literal than expected. The frontier labs are now moving into their customers’ markets, and they are doing it in plain sight.

The Pattern Has Hardened Into A Playbook

Read the moves of the past year together and a sequence emerges. Sell the model to the world’s most valuable businesses. Help wire it into their most sensitive and highest-margin work. Map how the business runs from the inside. Then turn those learnings into a product and stand up as the customer’s competitor.

Each step, taken alone, reads as partnership. Assembled, they read as strategy.

Of course, a lot still depends on execution. Just because a model company knows how a customer runs its business does not mean it can run that business better than the customer. However, the labs would say this is simply where the technology leads, and that is part of the truth.

It is also the most dependable way to convert a supplier relationship into a market entrant position.

Figma is the clearest illustration. It partnered with Anthropic on AI design tooling, the kind of collaboration that looks like validation for a newly public company. Then Mike Krieger, Anthropic’s chief product officer, resigned from Figma’s board, and three days later Anthropic launched Claude Design, a standalone tool aimed squarely at Figma.

The stock fell around 7% on the day, and the activist fund Findell Capital has since pushed the board to review the entire Anthropic relationship.

Asked about it at a closed event, Figma’s Dylan Field reached for the exact phrase OpenAI’s board once used to remove Sam Altman, saying the lab had not been “consistently candid.” The partner had become the competitor inside a single news cycle.

Harvey is a more practical example, because the evidence sits on Anthropic’s own website, which still features the legal startup as a customer success story for contract analysis, due diligence and litigation. In May, Anthropic launched Claude for Legal and began selling those same workflows directly to law firms.

Harvey carries an $11 Bn valuation, and the model that powered the challenger now stands beside it as a rival.

The pattern repeats wherever a workflow is worth owning. Intercom built its support agents on OpenAI’s models, and OpenAI now sells Presence, its own voice and chat support agent. Abridge and Ambience built clinical documentation on OpenAI’s API, and OpenAI now sells ChatGPT for Healthcare to hospitals, with a free tier for individual clinicians and documentation built in.

Benchling embedded Claude into biotech R&D across more than 1,300 companies, and Anthropic now sells its own research workbench, Claude Science.

The same logic is reaching financial services, where Claude connects to market-data providers and runs credit memos and know-your-customer checks inside banks. In each case the supplier watched the customer prove the market, then walked into it.

The strength of this incentive also shows up clearly in the earliest partnerships within the industry. Microsoft owns roughly 27% of OpenAI. It had bought GitHub in 2018 for $7.5 Bn. OpenAI is now building a jobs platform for professional hiring and, as per reporting in The Information, a code repository, moves that would sit against LinkedIn and GitHub, both Microsoft properties.

This is not a claim that anyone has been denied. It is a measure of how far the pull toward the customer’s market reaches, when it can tug even against an early strategic investor’s own businesses.

Pharma Is Where the Trap Is Sharpest

Drug discovery is the case that should give every board pause. Novo Nordisk and Bristol Myers Squibb route revenue and workflow data into Claude, treating Anthropic as a supplier. In May, BMS placed Claude before more than 30,000 employees as what it called a shared intelligence platform across research, clinical development and manufacturing.

On 30 June, Anthropic launched Claude Science and, in the same breath, announced it would run its own preclinical drug discovery programmes.

The customers, through their unchecked reliance on the model, are funding what they treat as a vendor but is in fact a competitor lurking in the shadows, and are helping to sharpen the very capability that may one day contest their pipelines.

The market read it at once. On the day, Schrödinger fell as much as 8.3%, Recursion 3.3% and IQVIA more than 2.3%. Anthropic frames the work around neglected diseases that large pharma overlooks, and that framing is fair as far as it goes. It does not change the structural fact that the vendor has walked onto the customer’s field.

The Economics Make Restraint Implausible

Follow the money and the behaviour explains itself. Anthropic’s revenue run rate reached about $65 Bn by the end of July, up from roughly $1 Bn in December 2024, at a $965 Bn valuation, with a filing already lodged for a listing that could come this autumn. OpenAI has roughly doubled to about $40 Bn in revenue.

Eight of the Fortune 10 are already Claude customers, a sign of how much of the real economy now runs through a single supplier. Selling tokens is a commodity business with thin and contested margins. Selling the finished application into a market with real returns on capital is where the value pools.

A company carrying a near-trillion-dollar valuation into public markets cannot credibly promise to leave the richest layer of the stack to its customers forever. None of this requires bad faith. The incentive is obvious, which is exactly what makes the move more likely.

The AI Narrative Is Getting Harder To Reconcile

This is where the public story and the operating record pull apart, and founders are reading one against the other.

The labs speak the language of safety, abundance and responsible stewardship, and in the same year they have entered design, law, support, healthcare and drug discovery. The mood at the frontier can run further still. 

On the All-In podcast in August, investor Gavin Baker said he had been told by people he trusts that Dario Amodei has floated the idea that Anthropic “might be the only private company in the world” at some point, with only governments beside it.

Anthropic rejected the account outright, with one of its researchers calling it false and Amodei rebutting it directly. Mark Zuckerberg has since published an essay making the opposite case, for open models and a distribution of power rather than its concentration.

This episode matters not because it is settled, but because it is not. The claims are second-hand, disputed and changing week to week, and that instability is itself the point. 

Nobody, inside these companies or outside them, can tell you today how this resolves. Of course, this narrative is great support and justification for the enterprise value of an IPO this large. What is not in doubt is the direction of the incentive and the exposure it creates for everyone who builds on top.

Alex Karp of Palantir has named the mechanism most plainly, arguing that enterprises are “paying for tokens that create no value” while surrendering their alpha, the proprietary edge that makes a business defensible.

Michael Burry put it more bluntly, remarking that Anthropic was eating Palantir’s lunch. The narratives will keep shifting. The opportunity arithmetic underneath them will not.

Own Your Alpha

For India the exposure is doubled. The country’s software firms and its 1,900-plus global capability centres are among the heaviest consumers of frontier APIs, and its services industry wires client workflows through those same models, sending client alpha upstream with every integration. The lesson from the cases above is that being a large and loyal customer is no shield. It may simply be the reconnaissance.

The opening is the same size as the risk, and it is the one I argued for last time, only more urgent now. Vertical AI built on domain depth the labs cannot see from the outside. Private and on-premise deployment of open models. Harnesses that keep proprietary data inside the building.

Open-weight models isolated from the internet and run for a single enterprise. Intercom has already moved its agent partly onto its own in-house open-weights model rather than depend on a supplier that competes with it, and that is the direction of travel.

India’s sovereign-model efforts and its vertical software builders should read this moment as the start of a new market preference forming.

Every Indian founder and CTO now faces the same reckoning before the next integration. It is to know precisely what alpha they are sending upstream, and whether they can protect it before it returns as somebody else’s product. The labs may or may not become the last companies standing. The firms that decide early to own their alpha will not have to find out.

The post Selling Intelligence Was the Warm-Up For AI. Owning Vertical Markets May Be The Play appeared first on Inc42 Media.