Enterprise AI buying shifts gears

Enterprise buyers are no longer treating artificial intelligence models like a speed test; they are buying for audit trails, vendor stability and exit options. (futurumgroup.com) Futurum Group said April 12 that OpenAI’s GPT-5.3 Instant Mini pushes faster responses and lower-latency use cases just as more companies move generative artificial intelligence into production. Its survey of 838 decision-makers found 67% of organizations already run generative artificial intelligence models in production and 75% plan to increase budgets. (futurumgroup.com) The same firm said OpenAI will shut down Sora’s web and app experiences on April 26, 2026, and retire the Sora application programming interface on September 24, 2026. OpenAI’s help center and developer deprecation page list those same dates and tell users to export content before the app closes. (futurumgroup.com) (help.openai.com) (developers.openai.com) That puts procurement teams on a different checklist than they used in 2024, when model rankings and demo quality dominated many buying conversations. OpenAI now sells enterprise privacy controls, retention settings, single sign-on and a compliance platform for logs, which shows how governance features have moved into the core product pitch. (openai.com) (help.openai.com) Anthropic is making the same argument from the other direction: reliability, controls and partner support. Its trust center lists SOC 2 Type II, ISO 27001 and ISO 42001 materials, and the company said in March it is putting $100 million into a Claude Partner Network to help enterprises move deployments from pilot to production. (trust.anthropic.com) (anthropic.com) PYMNTS reported April 12 that Anthropic is gaining on OpenAI in enterprise adoption as businesses widen their supplier lists. Anthropic has also been building distribution through service firms, including a November 2025 deal with Cognizant to make Claude available to as many as 350,000 employees. (pymnts.com) (anthropic.com) The services layer is expanding alongside the model race. Tata Consultancy Services said it has already made 25,000 campus offers in India, while its employee learning intensity rose 25% year over year as the company tied hiring and training to artificial intelligence demand. (telecom.economictimes.indiatimes.com) OpenAI still leads in Futurum’s survey, with GPT adoption at 61%, and its enterprise materials stress that customer data is not used to train models by default. Anthropic describes itself as focused on “reliable, interpretable, and steerable” systems, a pitch aimed at the same buyers now asking how to monitor and control model behavior after deployment. (futurumgroup.com) (openai.com) (anthropic.com) The result is a market where faster answers still matter, but a faster model or a discontinued product can now trigger the same review as any other supplier risk. In enterprise artificial intelligence, the buying question is shifting from “How good is the demo?” to “What breaks if this vendor changes course?” (futurumgroup.com 1) (futurumgroup.com 2)