Software multiples in the agentic era

The Bessemer Cloud Index sits at roughly 8.0x revenue in early 2026 (BVP Cloud Index, 2026). The number is doing too much work. It is a single multiple averaged across a category that has quietly split in three. One sub-cohort is repricing seat-based software for a margin that no longer exists. A second is repricing on outcomes and is still being judged on classic SaaS metrics it no longer matches. A third is mispriced as a moat business when the underlying defensibility belongs to the workflow upstream of it. The consensus underwriting still assumes the three things that made SaaS multiples coherent in the first place. Durable recurring revenue. Low cost to serve. Gross margin near the 80 to 90 point cloud ceiling. Agentic AI tests all three at once.

Start with recurring revenue. The seat-based contract was the unit that made SaaS revenue durable. The buyer purchased predictable access, the vendor recognised it ratably, and net retention compounded. Agentic AI undoes the unit. Roughly 65% of SaaS vendors have layered AI consumption meters on top of seat pricing, producing an immediate lift in average contract value that is real and a recurrence quality that is not the same as the pre-AI baseline (Monetizely, 2026). Seat-based pricing share fell from 21% to 15% of vendor base in 12 months. Hybrid surged from 27% to 41%. Gartner forecasts at least 40% of enterprise SaaS spend will be usage-, agent-, or outcome-based by 2030 (Monetizely citing Gartner, 2026). The investor implication is that a portion of every reported ARR line is now variable revenue with a different cohort behaviour from the seat baseline that drove the multiple in the first place. Treating it as the same recurring revenue is the first mispricing.

Move to cost to serve. The SaaS multiple assumed near-zero marginal cost per additional user. Inference cost destroys that assumption. ICONIQ’s January 2026 snapshot placed average AI product gross margin at 52%, up from 45% in 2025 and 41% in 2024 (ICONIQ, 2026). Bessemer’s State of AI 2025 placed LLM-native gross margin around 65% (Bessemer, 2025). Both are well below the 80 to 90 point cloud ceiling. The SaaS Magazine analysis of AI cost-of-goods compression is concrete: bolting an AI assistant onto an 80-dollar seat adds roughly 15 dollars of variable cost and moves gross margin from 80% to 65% on that seat (SaaS Magazine, 2026). Agentic architectures, where a single user action triggers a sequence of model calls, multiply inference cost 5 to 20 times per user action (SaaS Magazine, 2026). Every dollar of AI revenue carries variable cost that classic SaaS multiples have not historically priced. For every million dollars in 2026 AI revenue, roughly 230,000 dollars exits as inference cost before headcount (The SaaS CFO, 2026). The framework needs a gross-margin floor explicit in the underwriting, not implicit.

Then the moat. The classic SaaS multiple combined recurring revenue with a defensibility assumption that the product owned a workflow the buyer would not replace. Agentic AI tests the workflow assumption directly. HarbourVest’s read on the great reset names the three moat categories that survive intact: data moats, compliance infrastructure, and systems of record (HarbourVest, 2026). Houlihan Lokey’s Q1 2026 vertical software report reaches the same view from the M&A side, with vertical AI that owns proprietary workflow data reshaping competitive moats faster than horizontal AI is doing anything visible (Houlihan Lokey, 2026). Morningstar’s framework anchors on data scale, switching cost engineering, and distribution into bought-and-paid-for accounts (Morningstar, 2026). The investor inference is that a meaningful share of the existing public software cap stack is in the fourth bucket: workflow exposure to AI substitution that the current multiple does not reflect.

What the framework should evolve to. Three sub-cohorts now need to be priced separately, not averaged.

Premium-multiple agentic AI. Vertical workflow ownership, proprietary data that compounds with usage, outcome-based pricing that passes inference cost through, gross margin trending toward the Bessemer 60% Shooting Star band (Bessemer, 2025). The underwriting case is recurring outcome revenue rather than recurring seat revenue, with the moat sitting in the data and the workflow rather than the application.

Fairly-priced hybrid SaaS. Seat-based product with disciplined consumption-meter layering, gross margin stabilising near the 65% AI-software band, hybrid pricing now contractual at renewal, NRR in the 110 to 120 range with consumption tail-wind (Monetizely, 2026). The underwriting case is steady multiple in the 4 to 6 times revenue band that nature.com-style high-NRR cohorts have historically held (Nate Lind on Bessemer multiples, 2026).

Mispriced legacy SaaS. Seat-based product with no consumption discipline, gross margin compressed by AI bolt-ons that have not been repriced, workflow exposure to agentic substitution, and multiple still anchored to the 8x Cloud Index average. The underwriting case is rerating to the 1 to 3 times revenue band where flat-growth, high-churn names already trade (Nate Lind, 2026).

The Bessemer Supernova-versus-Shooting-Star segmentation, 25% margins with experimental pricing on one side and 60% margins after refinement on the other, is the same distribution viewed at the company level (Bessemer, 2025). A multiple that prices the average has been wrong at both ends for at least four quarters.

The new operating metrics investors should price on. Inference efficiency, in revenue per inference dollar, is now a unit-economic disclosure that should sit alongside gross margin (The SaaS CFO, 2026). Outcome-priced revenue as a share of total revenue is the variable that distinguishes durable recurring outcome from temporary consumption lift. Workflow ownership measured at the moat layer that HarbourVest names is the variable that separates premium from rerating candidates. Gross margin on the AI line stripped of cross-allocations is the variable that closes the gap between disclosure and underwriting.

The downside case, stated honestly. If buyers continue to underprice outcomes and overprice access, hybrid SaaS holds its multiple longer than this note implies. If inference costs fall meaningfully through model-efficiency gains, the 52% ICONIQ floor moves up and the cohort splits compress. The view changes if the gross margin of disclosed AI revenue across the index moves above 70% in two consecutive quarters. Until that print, the framework above is the better underwriting frame.

The diligence checklist. Disclose AI revenue, AI gross margin, and inference efficiency ratio. Identify the moat category, data, compliance, system of record, or vertical workflow. Price outcome revenue and seat revenue separately. Hold the Cloud Index average at arm’s length when underwriting individual names. The companies that hold premium multiples in 2027 will not be the longest AI roadmaps. They will be the ones whose AI line, stripped, prints inside the Shooting Star band.

Sources

• Bessemer Venture Partners, BVP Nasdaq Emerging Cloud Index, 2026
• Bessemer Venture Partners, “The State of AI 2025”
• ICONIQ, “State of AI Bi-Annual Snapshot 2026”
• SaaS Magazine, “The AI COGS Problem: SaaS Gross Margin Compression 2026”
• The SaaS CFO, “Your AI Feature Is Quietly Destroying Your Gross Margin”, 2026
• The SaaS CFO, “How to Calculate the Inference Efficiency Ratio”, 2026
• Monetizely, “The 2026 Guide to SaaS, AI, and Agentic Pricing Models”
• HarbourVest, “The Software Industry’s Great Reset and the New Moat That Matters”
• Houlihan Lokey, “AI in Vertical Software, Q1 2026”
• Morningstar, “What AI Means for Software Companies’ Moats”
• Nate Lind, “SaaS Valuation Multiples in 2026: What the Bessemer Cloud Index Means for Your Exit”