Enterprise AI moving into procurement
Enterprise AI is shifting from pilots to line‑item spending: OpenAI says enterprise now accounts for more than 40% of its revenue as customers buy agentic workflows rather than isolated chat tools. Vendors are competing on computing capacity, controls and price — not just model claims — and smaller players can still win by packaging clear workflows, as Perplexity’s recent ARR milestone shows. (decrypt.co) (bloomberg.com) (techstory.in)
A year ago, a lot of companies treated artificial intelligence like a lab demo: one team got a chatbot, another team ran a pilot, and finance could still call it an experiment. This week, OpenAI said enterprise customers now make up more than 40% of its revenue, which means the software is moving into actual budget lines instead of side projects. (openai.com) (decrypt.co) The product companies are buying is changing too. OpenAI said customers are shifting toward “agentic workflows,” which means software that does a chain of tasks like pulling data, drafting an answer, and taking an action, instead of just returning one chat response. (openai.com) (decrypt.co) That changes who signs the check. A chatbot can live inside one department, but a tool that touches contracts, support tickets, code, or internal records usually has to clear procurement, security, and information technology before it spreads across a company. (openai.com 1) (openai.com 2) Once procurement gets involved, the sales pitch gets less glamorous. OpenAI’s enterprise offers single sign-on, system for cross-domain identity management, usage controls, and a compliance platform for logs, while Anthropic is pushing its own admin controls, spend caps, and single sign-on for business plans. (help.openai.com 1) (help.openai.com 2) (anthropic.com) (support.claude.com) The competition is also moving down to raw capacity. Bloomberg reported on April 9 that OpenAI told investors it has a computing advantage over Anthropic, which is another way of saying the company thinks it can serve more demand without slowing down or turning customers away at busy times. (bloomberg.com) Price starts to matter more in that world than benchmark bragging rights. OpenAI has also been emphasizing cost controls for application programming interface customers, because a chief information officer buying thousands of seats cares about predictable bills more than a flashy demo. (openai.com 1) (openai.com 2) This is why smaller players still have room. Perplexity says more than 20,000 organizations use its enterprise product, and its enterprise pitch is not “we built the biggest model” but “we give your staff a secure answer engine with citations, file access, and multiple models in one place.” (perplexity.ai) (perplexity.ai) That packaging is turning into real money. TechStory reported that Perplexity’s annual recurring revenue passed $450 million in March 2026, up about 50% from the prior month, after it introduced “Computer,” an agent that can handle tasks like shopping, email sorting, and social media summaries. (techstory.in) So the market is starting to look less like a science fair and more like ordinary enterprise software. The winners still need strong models, but they also need enough chips, enough controls, and a simple enough workflow that a procurement team can compare them the way it compares payroll software or cloud storage. (openai.com) (perplexity.ai))
