The person who runs acquisitions for Salesforce has a message for the founders of artificial-intelligence companies valued in the tens of billions: there is no queue of buyers behind you.
John Somorjai said the exit routes for AI startups valued above $10 billion are unusually narrow. Public listings have been rare, strategic buyers are few, and the valuations set in private rounds are difficult to realise in either channel. The gap between what a company’s last round implied and what a buyer will pay has become the central problem of the sector’s capital structure.
His reasoning is not abstract. Salesforce has been one of the most active acquirers in enterprise software for a decade, and it looks at more companies than it buys. Somorjai described the diligence tests that stop a deal: security vulnerabilities, customer churn that is higher than the pitch deck claims, and the absence of a distribution channel that can carry the product to buyers who are not already looking for it.
Each of those is a way of asking the same question, which is whether the company sells something a customer cannot easily replace. A model wrapped in an interface is replaceable. The buyer can rebuild it, or wait for the platform vendor to ship the same feature, and the second option is usually cheaper. That is the specific risk that a large valuation makes intolerable, because a $10 billion price requires a business that survives being copied.
The distribution test is the one that eliminates most candidates. Enterprise software buyers do not shop. They renew, they expand, and they adopt what their existing vendors put in front of them. A startup without a channel has to build a sales organisation from scratch, and the cost of that organisation is measured in years and in quota-carrying headcount, not in a marketing budget.
Churn is the test that is hardest to argue with. AI products have shown impressive trial rates and unimpressive retention, because the first month is a demonstration and the twelfth month is a workflow. A tool that a team tries and abandons has negative value to an acquirer, since the acquirer inherits the support cost and the reputation.
The pricing question cuts across all of it. Somorjai described a shift in software from charging per seat to charging per result, citing the example of 99 cents for each question successfully resolved. That sounds like a small change in a contract. It is a large change in a business, because it moves the risk of failure from the customer to the vendor and makes the vendor’s revenue depend on whether the product actually works.
Per-seat pricing paid for access. Per-result pricing pays for output, which means a company that cannot measure its output cannot bill for it, and a company whose output is inconsistent will watch its revenue move with its error rate. Procurement departments have wanted this arrangement for years. Vendors have resisted it for the same length of time, for obvious reasons.
For the companies caught in the middle, the implication is that the next round is not the answer. A startup that raises at a high valuation with per-seat economics and thin retention is borrowing time against a buyer’s diligence, and the diligence will arrive eventually. Somorjai’s view is that a reckoning is coming for the companies whose product is a thin layer over someone else’s model.
That view is not universal, and the counter-argument deserves a hearing. Every platform shift produces a wave of companies that look thin and turn out to own a workflow, a data set or a regulatory position that the platform vendor cannot replicate. The vendors themselves have an interest in an ecosystem, because customers buy outcomes and the vendors sell components.
But the counter-argument is a case-by-case claim, and the market has been pricing it as a category. When the ten largest private AI companies are valued in aggregate at a number that assumes all of them become platforms, the arithmetic requires buyers who do not exist. Only a handful of companies in the world can write a cheque of that size, and several of them are the same companies being valued.
The result is a queue of companies with strong technology, no exit and a valuation that prevents a down round. Some will take a lower price and call it strategic. Some will merge with each other, which solves nothing but changes the story. Some will keep raising from investors who need the mark to hold.
Salesforce is in an unusual position to watch this, since it buys companies and sells to the same customers they sell to. Its own interest is in a market where prices fall, which is worth remembering when reading its deal chief’s assessment of prices. A buyer who says few assets are worth the asking price is also a buyer negotiating.
What Somorjai’s comments describe is a sector that has been financed on the assumption of an exit and has not yet produced the exits. The technology is real. The revenue is real at some companies and aspirational at others. The valuations were set in a market where the difference did not matter, and the buyers who will decide what happens next are saying that it does.
