Enterprises reevaluate AI vendors every six months, putting ARR at risk

AI budgets grow, loyalty does not

Enterprises are on pace to spend $4.25 trillion on technology in 2026, and almost all of that growth is being driven by AI, according to IDC projections reported by TechCrunch. Yet the same spending wave is producing something the software business has not really seen before: enterprise revenue that stays insecure even after a product has been adopted.

Survey data from the venture capital firm Madrona, which polled 150 enterprise IT professionals, underlines the split. According to TechCrunch, 74% of respondents plan to expand their AI budgets over the next 12 months, and the rest intend to hold spending flat. Even so, fewer than half of the AI pilots those enterprises run ever graduate into full production.

Madrona frames that as an improvement rather than a failure. A widely cited MIT study reported last year concluded that 95% of enterprise AI projects failed to deliver a return on investment, so a sub-50% production rate, while still a low bar, marks real progress.

The multi-year contract is losing its power

The report's most consequential finding concerns what happens after a deployment succeeds. Some 77% of enterprises reevaluate their AI vendors every six months or on a continuous rolling basis. Madrona describes a "fast in, fast out" dynamic that breaks with traditional enterprise SaaS, where multi-year contracts and high switching costs gave vendors a durable moat of inertia. In enterprise AI, the firm argues, switching costs are lower and the reassessment cadence is relentless.

That has direct implications for the annual recurring revenue figures AI startups report. As TechCrunch notes, enterprise trial budgets fueled the AI boom of 2025, and 2026 was expected to be the year large customers settled into long-term commitments. Those contracts are what have allowed some startups to claim extraordinary growth, including cases of going from zero to $10 million in three months. For the first time, revenue remains at risk even after a product clears the pilot stage and is rolled out.

Pricing is part of the friction

Part of the problem is that many AI vendors have not settled on a pricing model that matches enterprise expectations. Separate research from Andreessen Horowitz, based on a survey of 50 technical AI buyers, found that more than half of them want AI fees tied to the work produced or to other outcomes, rather than to usage measures such as tokens consumed.

Charging by tokens is effectively a SaaS-era approach, suited to products where demand is predictable: once a company knows it needs email, HR software or cloud storage, cost simply scales with headcount or data volume. AI value is harder to map onto raw consumption. a16z partners Tugce Erten and Sarah Wang argue that pricing anchored to recognizable work, such as the number of reports processed, tickets closed or leads generated, is what lets a vendor prove its worth and gives the buyer a measurable return.

Why it matters

For SaaS incumbents and AI startups alike, these findings invert a foundational assumption of enterprise software: that once a deal is signed, the revenue is largely banked. AI has opened an era of continuous experimentation. Enterprises are more willing to try new vendors, which is an opportunity for startups, but no contract now guarantees durable income. Vendors will effectively have to re-earn their revenue every six months, and headline ARR and growth rates should be read with that churn risk in mind. Buyers, meanwhile, gain leverage to push for outcome-based pricing and to keep switching if results disappoint. Whether enterprises eventually revert to their historical habit of long-term technology commitments, TechCrunch notes, remains an open question.