Enterprise AI Has Become a Line Item, Not an Experiment
Business customers now account for the majority of revenue at the largest model developers, and the fastest-growing products are the unglamorous ones: coding, support and document work.
The most reliable indicator of whether a technology has arrived is not adoption; it is renewal. By that measure, the enterprise phase of artificial intelligence is well underway.
Business customers account for roughly 80% of revenue at one of the two largest model developers, with more than 300,000 firms using its tools. At the other, enterprise sales have grown to more than 40% of the mix and are approaching parity with consumer subscriptions.
What companies are actually buying
The applications generating revenue are narrower and more boring than the demonstrations that generate attention.
Software development is the largest category by a wide margin. One coding product reached close to $1 billion in annualized revenue within a year of launch. The economics are easy to justify: developers are expensive, and a tool that measurably increases output pays for itself.
Customer support is second, where deflecting routine inquiries has a directly measurable cost saving.
Document-heavy work — contract review, claims processing, compliance checks — is third, concentrated in financial services, insurance and health care.
The distribution problem
Selling to large enterprises is not a product problem; it is a channel problem. Model developers have responded by building partner networks aimed at the systems integrators and consultancies that decide what large organizations actually deploy, with one committing an initial $100 million this year for training, technical support and joint marketing.
That is an unglamorous investment, and it is the one that most reliably separates enterprise software companies that scale from those that stall.
The measurement question
The uncomfortable finding across corporate surveys is how few deployments have produced a quantified return. Companies are spending, and spending is rising, and a substantial share of projects remain in pilot or in a state where the benefit is described qualitatively.
That gap is what the market is really debating when it argues about AI capital spending. If the productivity shows up in enterprise financial statements over the next two years, the infrastructure buildout is justified. If it does not, a great deal of spending will have been pulled forward against a return that arrives later — or somewhere else.
