Analysis
New research from Madrona Venture Capital, Andreessen Horowitz and MIT paints a consistent picture: annual recurring revenue in AI-native software is a far less durable metric than it was in the SaaS era that preceded it, TechCrunch reported.
The "Fast In, Fast Out" Problem
Madrona's research on enterprise AI ROI found that 77% of the enterprises it studied reevaluate their AI vendors every six months or on a rolling basis -- a sharp departure from the annual or multi-year contract cycles that made traditional SaaS revenue so valuable to underwrite. Madrona's own framing calls this a "fast in, fast out" dynamic, and it changes what an ARR number actually signals: in legacy SaaS, a dollar of ARR implied roughly a dollar of expected revenue over the following twelve months, discounted for churn. In AI-native software, that same dollar can be gone at the next quarterly review if a competitor ships a better model or a cheaper price, with none of the switching-cost friction that protected incumbent SaaS vendors for a decade.
The Pilot-to-Production Gap Is Real, But Improving
A separate MIT report found that fewer than half of enterprise AI pilots now reach full production -- a meaningful improvement from the 5% success rate MIT documented in 2025, but still a conversion rate that leaves more than half of enterprise AI initiatives stalled somewhere between proof-of-concept and paid deployment. A 150-person survey of enterprise IT professionals found 74% planning to expand AI budgets within 12 months regardless, meaning the spending intent is real even where individual pilots fail to convert -- enterprises are still committing net-new AI budget even as their track record of converting that budget into production usage remains weak.
Pricing Is Shifting Under the Same Pressure
A16z's own study of 50 technical AI buyers found more than half prefer outcome-based pricing -- paying for a specific result, like reports processed or support tickets closed -- over the token or seat-based pricing models that dominate today. a16z's own partners describe this as "economically valuable to both sides," since outcome pricing aligns vendor incentives with actual customer value delivered rather than raw usage volume. But outcome-based pricing is also structurally more volatile revenue than a flat per-seat subscription: it rises and falls with a customer's own business activity, which means AI vendor revenue increasingly inherits the cyclicality of its customers' businesses rather than sitting insulated from it the way traditional per-seat SaaS largely did.
What This Means for Valuations
IDC projects $4.25 trillion in total enterprise tech spending for 2026, confirming the addressable market genuinely is enormous -- the opportunity isn't in question. What's in question is whether the ARR multiples venture investors have applied to AI-native startups this year -- multiples Pulse has flagged as high as 400x revenue on some recent rounds -- properly discount for a revenue base that churns on a six-month cycle rather than an annual one. Traditional SaaS multiples were built on the assumption that ARR compounds predictably; if AI-era ARR is genuinely less sticky, the same multiple applied to a less durable revenue stream is a mispricing that shows up eventually, whether at the next down round or the first IPO that has to explain net revenue retention to public-market analysts who've never seen a six-month enterprise re-bid cycle before.
For founders, the practical takeaway from this research is structural, not cosmetic: contract terms, renewal cadence and customer concentration now matter as much as the top-line ARR number itself when it's time to raise the next round.
