Enterprise AI shifts to economics

Enterprise buyers are starting to treat generative artificial intelligence like any other software purchase: speed, latency, and cost now weigh more than raw model size. (futurumgroup.com) Futurum Group said OpenAI’s GPT-5.3 Instant Mini launch points to that shift, as companies move from pilots into production and care more about response time and cost per task. Futurum’s first-half 2026 survey of 838 decision-makers found 67% of organizations already run generative artificial intelligence in production, 75% expect to raise budgets, and OpenAI leads enterprise adoption at 61%. (futurumgroup.com) OpenAI described GPT-5.3 Instant as a faster update to ChatGPT’s most-used model, and its developer documentation positions the lower-cost “mini” line for “cost sensitive, low latency, high volume workloads.” OpenAI’s current pricing page also lists sharp price gaps between flagship and mini models, reinforcing why procurement teams compare token costs as closely as model quality. (openai.com) (developers.openai.com) (openai.com) That buying logic is getting support from finance chiefs, not just engineering teams. Bain & Company said Monday, April 13, 2026, that 42% of chief financial officers expect artificial intelligence investment to rise by more than 30% within two years, based on a global survey of more than 100 chief financial officers. (bain.com) Bain said 56% of senior finance executives are already increasing enterprise-wide artificial intelligence investment by more than 15% this year, and 83% expect increases above 15% over the next two years. The same Bain research says finance departments are moving from approving artificial intelligence budgets to using the tools inside forecasting, reporting, and other internal workflows. (bain.com 1) (bain.com 2) The backdrop is a market that has changed quickly since 2025. OpenAI released GPT-5.3 Instant in March 2026 as a smoother, faster ChatGPT model, then released GPT-5.4 on March 5, 2026, as its more capable flagship model for professional work. (openai.com 1) (openai.com 2) That split mirrors how enterprises now shop for models. The largest model still matters for complex coding, research, and multi-step work, but lower-latency models are better suited to customer support, internal assistants, document handling, and other high-volume jobs where a few seconds and a few cents multiply fast. (futurumgroup.com) (developers.openai.com) Futurum said vendors now have to prove “workflow fit” alongside performance, and Bain’s finance survey points in the same direction: companies are preparing to spend more, but they want deployments that plug into existing systems and show measurable returns. The result is a procurement market that looks less like a race for the smartest model and more like a contest over unit economics. (futurumgroup.com) (bain.com)