Research report · AI · Valuations · Q3 2026

AI Valuations: Q3 2026

Most AI valuations quoted in 2026 are not multiples. Windsor Drake's 36-company public AI comparable set carries a median 15.0x EV/Revenue with a full range of 0.6x to 68.7x, computed from exchange closing prices of 21 August 2026 and SEC filings. Seven private primary rounds that published a contemporaneous run rate imply a median 27.1x on that run rate, 81% above the priced public median. Introduces The Denominator Test: a valuation is a multiple only when a transactable price meets an audited revenue figure.

Sector
AI
Focus
Valuations
Published
August 24, 2026
Length
25 slides
Reading time
13 minutes

Slide deck

25-slide deck. Desktop readers can page through the embedded viewer below. Mobile readers can open the direct PDF link.

Cover of AI Valuations: Q3 2026 slide deck Open slide deck PDF

Key findings

  • Composite AI EV/Revenue is 15.0x across 36 listed constituents, interquartile range 9.3x to 23.8x, full range 0.6x to 68.7x.
  • AI cloud and compute hosting clears the highest median at 33.2x, ahead of AI silicon at 23.9x, while AI application software clears 11.1x.
  • Seven AI primary financings since November 2025 that published a contemporaneous run rate imply a median 27.1x on that run rate, 81% above the priced public median of 15.0x.
  • Gross margin correlates with the log of EV/Revenue at r = 0.82 across the 16 constituents disclosing it: Palantir at 84.7% margin clears 68.7x, Super Micro at 10.8% clears 0.6x.
  • Contracted revenue reorders the AI cloud layer: Nebius clears 44.3x on trailing revenue and 1.50x on disclosed customer commitments, against CoreWeave at 10.3x and 0.75x.
  • 65 of the 66 AI-capability transactions in the Windsor Drake Exit Index were acquired by a strategic operator and one by a financial sponsor.
  • The largest AI acquisition on record, SpaceX's $60bn all-stock purchase of Anysphere closing 14 August 2026, filed no target financials, while IBM's $11.3bn cash acquisition of Confluent computes to 9.1x trailing revenue.
  • AI companies are about half of S&P 500 market capitalisation, up from about a quarter in 2022, and private credit outstanding to AI firms has gone from near zero to over $200 billion.

Methodology

Framework: The Denominator Test. All multiples are Windsor Drake's own computation from primary market data, not licensed from a vendor: enterprise value equals market capitalisation plus total debt less cash and short-term investments, using official Nasdaq and NYSE closing prices of 21 August 2026 and balance-sheet and trailing-twelve-month revenue from each company's most recently reported fiscal quarter. 36 listed constituents across six sub-segments, all named. Private valuations are classified as a priced multiple, a private round mark or a headline valuation, and every private figure carries its classification. Implied private marks divide a disclosed post-money valuation by a run rate the company published with the same round and are labelled as such.

Frequently asked questions

What multiple do AI companies trade at in 2026?

Windsor Drake's 36-company public AI comparable set carries a median 15.0x EV/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. The composite is wide enough that the layer of the stack a company occupies matters far more than the AI label attached to it.

Is a private AI valuation the same thing as a multiple?

No. A private round prices a share class; it rarely publishes an audited revenue figure. Where a company did publish a contemporaneous run rate, the implied mark can be computed: seven such rounds since November 2025 imply a median 27.1x on that run rate, against 15.0x for priced public companies. A tender or secondary price is a third category again, because no share class was priced by a primary financing.

Which AI sub-segment carries the highest multiple?

AI cloud and compute hosting, at a 33.2x median across six constituents, ahead of AI silicon and silicon IP at 23.9x across eight. Both sit well above AI application software at 11.1x. The ordering is the opposite of what gross margin alone would predict, because the cloud layer is priced against contracted revenue that trailing revenue has not yet caught.

Why does Palantir trade at 68.7x and C3.ai at 4.1x?

Margin structure and growth durability. Palantir reported 84.7% GAAP gross margin, 47.1% GAAP operating margin and 93% revenue growth in the quarter to 30 June 2026. C3.ai reported 30.9% gross margin for fiscal 2026 and revenue down 52.5% in its fourth quarter. Both are sold as enterprise AI.

What sets an AI company's gross margin?

Inference and inspection cost. The price of a given capability level has fallen sharply: Stanford HAI measured the cost of a GPT-3.5-equivalent query dropping from $20.00 to $0.07 per million tokens between November 2022 and October 2024. Frontier pricing has moved differently. Anthropic's top Opus tier fell from $15 to $5 per million input tokens over the year to August 2026 while a higher tier opened above it, and OpenAI's flagship input price rose from $1.25 to $2.00.

How many AI transactions disclose a valuation multiple?

Few. Fourteen of the 66 AI-capability transactions in the Windsor Drake Exit Index disclose consideration. The largest AI acquisition on record, SpaceX's $60 billion all-stock purchase of Anysphere, closed on 14 August 2026 with no target financials filed. Public comparables are the reproducible benchmark available to a seller.

Who is buying AI companies in 2026?

Operators. Sixty-five of the 66 AI-capability transactions in the Exit Index were acquired by a strategic buyer and one by a financial sponsor. Named acquirers in the trailing twelve months include Bloomberg, Stripe, Airwallex, SoFi, Coupa, Backbase, Duck Creek and OpenAI itself. IBM acquired Confluent for $11.3 billion in equity value, closing 17 March 2026, at roughly 9.1x fiscal 2025 revenue.

Is 2026 a good year to sell an AI company?

Enterprise AI spending is forecast at $2.60 trillion in 2026, up 47% (Gartner), and the listing window has reopened for scale issuers, with 104 US IPOs raising $145.8 billion year to date (Renaissance Capital). Official-sector bodies are simultaneously flagging concentration risk: the Bank of England puts AI companies at about half of S&P 500 market capitalisation. Whether it suits a specific asset depends on its margin structure and its readiness, not on the calendar.

Companies covered

Public and private companies referenced in this report.

NVIDIAAdvanced Micro DevicesBroadcomMarvell TechnologyAstera LabsCredo TechnologyArm HoldingsMonolithic Power SystemsTaiwan SemiconductorMicron TechnologyVertivSuper Micro ComputerCoreWeaveNebius GroupIRENApplied DigitalCipher MiningTeraWulfEquinixDigital RealtyOracleSnowflakeDatadogMongoDBElasticCloudflareGitLabBrazePalantirAppLovinSamsaraTempus AISoundHound AIInnodataBigBear.aiC3.aiOpenAIAnthropicxAIDatabricksAnysphereElevenLabsLovableSunoGroqCerebras SystemsTogether AIMistral AIFigure AIConfluentIBMSpaceXMeta PlatformsMicrosoftAlphabetAmazonScale AIBloombergStripeAirwallexRegologyCanoe IntelligenceMetronome

Download the deliverables

If you are evaluating a sale

Windsor Drake is a sell-side only M&A advisory firm for founder-led software, fintech, and technology companies with enterprise values between $5 million and $300 million. This research comes from the same desk that runs our sale processes. If you are weighing an exit in this market, a confidential valuation is the place to start: no obligation, senior attention, and a view grounded in the transactions this report tracks.

Cite this report

AI Valuations: Q3 2026. Windsor Drake Market Intelligence, 2026. windsordrake.com/market-intelligence/reports/ai-valuations-q3-2026. Windsor Drake Market Intelligence data is free to use with attribution to Windsor Drake (windsordrake.com).

From Windsor Drake