Valuation Lab · Benchmark
Windsor Drake’s 36-company public AI comparable set carried a median 15.0x EV/trailing revenue as of 21 August 2026, with an interquartile range of 9.3x to 23.8x and a full range of 0.6x to 68.7x. AI cloud and compute hosting cleared 33.2x, AI silicon 23.9x and AI application software 11.1x. Seven private primary rounds since November 2025 imply a median 27.1x on unaudited run-rate revenue, 81% above the public median. Those are two different data classes: post-money funding marks are shown in their own view and are never averaged with trading multiples.
This is the evergreen page for the series. The dated Q3 2026 report keeps its historical scope and is linked below; this explorer will carry each subsequent quarter’s observations as they are verified. Margin structure, not the AI label, explains the dispersion: Palantir at 84.7% gross margin and 93% growth trades at 68.7x, C3.ai at 30.9% gross margin with revenue down 52.5% trades at 4.1x, and Super Micro at 10.8% gross margin trades at 0.6x.
The interactive explorer loads here. The default view is published below so the numbers are readable without JavaScript.
Public AI comparable set, median EV/revenue: 15.0x. EV / trailing revenue, 36-company public AI comparable set as of 21 Aug 2026 (Windsor Drake Market Intelligence, AI Valuations: Q3 2026 (published 24 Aug 2026; public set as of 21 Aug 2026) (retrieved 2026-09-22)). Compute hosting clears 33.2x, silicon 23.9x and application software 11.1x; gross margin and growth durability, not the AI label, separate Palantir at 68.7x from Super Micro at 0.6x.
| Segment or constituent | EV/revenue | n |
|---|---|---|
| Whole set: interquartile range (n=36) | 9.3x-23.8x (median 15.0x) | 36 |
| AI cloud and compute hosting (n=6) | 33.2x | 6 |
| AI silicon and silicon IP | 23.9x | |
| AI application software | 11.1x | |
| Palantir (84.7% gross margin, 93% growth) | 68.7x | |
| C3.ai (30.9% gross margin, revenue down 52.5%) | 4.1x | |
| Super Micro (10.8% gross margin) | 0.6x |
Public trading observations are enterprise value over trailing revenue for listed companies, segmented into application software, infrastructure and compute, and silicon. Private funding observations are post-money equity valuations divided by whatever run-rate revenue the round disclosed; they carry no audited denominator and reflect option value on future scale, so they sit in a separate view with their own label. Disclosed control transactions are rarer still: fourteen of the 66 AI-capability deals in the Windsor Drake Exit Index disclosed consideration, and IBM’s $11.3B purchase of Confluent computes to 9.1x trailing revenue while SpaceX’s $60B all-stock purchase of Anysphere disclosed no revenue at all.
Application software clears the lowest AI median (11.1x) because inference sits in its cost of revenue and compresses gross margin. Infrastructure and developer-facing segments carry higher forward multiples in the broader software market too: DevOps at 9.0x, data infrastructure at 5.5x and developer tools at 5.4x EV/NTM revenue in September 2026 (Multiples.vc), against 2.2x for horizontal SaaS. A founder building AI application software should model inference cost through gross margin in the SaaS calculator’s AI preset rather than assume a category premium.
Sixty-five of the 66 AI-capability transactions in the Exit Index had a strategic buyer. Sponsors are largely absent at current marks, which means process design for an AI company is about strategic fit, data assets and margin durability rather than leverage capacity.
If you are weighing a sale and want a senior advisor to test these numbers against live buyer appetite, the first conversation is confidential and without obligation.
A median 15.0x EV/trailing revenue across Windsor Drake’s 36-company public set as of 21 August 2026, with an interquartile range of 9.3x to 23.8x; compute hosting 33.2x, silicon 23.9x, application software 11.1x.
No. Private rounds rarely publish audited revenue. The seven rounds observed imply a median 27.1x on run rate, but that is a funding mark, not a trading or transaction multiple, and it is shown separately.
AI cloud and compute hosting at a 33.2x median across six constituents, ahead of AI silicon and silicon IP at 23.9x.
Through cash flow. Use the SaaS calculator with the AI application preset (lower gross margin from inference cost) and the DCF calculator; do not add an AI premium to a SaaS multiple.
Three data classes are kept separate throughout: public trading observations, disclosed control transactions and Windsor Drake analyst ranges; funding-round marks form a fourth class where present. Medians are stated with geography, period, sample size and basis where the publisher provides them; subgroups below three observations are suppressed; judgment ranges are never presented as quartiles. Negative or zero EBITDA makes EV/EBITDA not meaningful. The benchmark snapshot is versioned (2026.09.22-1) and the chart, table, download and written summary reconcile to it. Cite as: Windsor Drake, “AI valuation multiples” (2026.09.22-1), https://windsordrake.com/ai-valuation-multiples/.
Related: AI Valuations: Q3 2026 (dated report) · DCF calculator · SaaS valuation multiples · All valuation tools · Download the benchmark dataset (JSON)