ChainCatcher reported that FundaAI, in a research report on enterprise AI applications, said enterprise AI budgets are still expanding, but the trajectory begins to diverge in the second half of 2026 and in 2027.
The report said guidance from a large U.S. telecom operator showed AI spending rising from a January base of 100 to about 190 in December, with another 40%–50% year-over-year increase expected in 2027. By contrast, a large European automaker A has increased spending by only 10%–15% so far this year, and its guidance for next year is broadly flat.
It also said incremental spending is moving away from paid seats and toward API and token consumption, along with production workflows. For the telecom operator mentioned in the report, the mix between subscriptions and API has shifted from about 50%/50% to 40%/60%, and may move to 35%/65%. A mid-to-large biopharma company A, meanwhile, adjusted from 80%/20% to about 70%/30%.
On open-source adoption, the report said take-up is uneven. In active use cases, adoption can account for 30%–40%, but the spending share is smaller because unit pricing is lower. Experts cited in the report estimated that open-source inference can be about 40%–70% cheaper than closed-source frontier models. On a full-cost basis, that gap narrows to 20%–40%. Model routing, caching, and context compression can still cut API spending by another 20%–30%.
On the production side, AI budgets are increasingly being built bottom-up around workflow ROI. The telecom operator’s typical production ROI was put at about 1.5x to 2x, with a payback period of 6 to 18 months. Mature use cases can reach 3x to 5x.
The report added that the next wave of spending will be tied to agents, software modernization, network operations, productized workflows, and longer-cycle business processes. At the same time, engineering capacity, process redesign, governance, and data readiness are becoming tighter constraints than funding.