OpenAI is heading toward its IPO at roughly 34x revenue and Anthropic at roughly 21x, while the median public SaaS company trades at 8.5x EV/revenue in mid-2026. That's the short answer. The longer answer is that these aren't the same valuation framework wearing different multiples — they're two entirely different underwriting models.
SaaS investors have spent a decade converging on one formula: take ARR, adjust for net revenue retention and gross margin, apply the Rule of 40, and land on a multiple. AI IPO investors are throwing most of that out and pricing off two variables instead — how fast revenue is compounding right now, and how scarce the underlying compute or model capability is. Understanding which framework applies to which company is the difference between correctly pricing the next wave of AI IPOs and getting run over by them.
Figures from OpenAI and Anthropic IPO filing disclosures, TechCrunch, Yahoo Finance, and the Value Add VC SaaS Valuations dashboard, as of July 2026.
How AI Companies Are Pricing IPOs Differently Than SaaS
AI companies are pricing their IPOs on trailing revenue growth rate and infrastructure scarcity rather than the ARR-quality framework — net revenue retention, gross margin, Rule of 40 — that sets SaaS multiples. Foundation-model companies average roughly 37.5x revenue in 2026 versus a public SaaS median of 8.5x, a gap that reflects investors underwriting AI on a growth-and-scarcity model instead of a recurring-revenue-quality model.
| Valuation Input | AI IPO Framework | SaaS IPO Framework |
|---|---|---|
| Primary multiple driver | Revenue growth rate + compute scarcity | ARR quality (NRR, gross margin, Rule of 40) |
| Typical 2026 multiple | 21-38x revenue | 8.5x EV/revenue (median) |
| Gross margin tolerance | Negative-to-low margin accepted if growth is high | 75%+ gross margin expected for premium multiple |
| Retention metric weight | Secondary — usage growth matters more than logo retention | Primary — NRR above 120% drives multiple expansion |
| Comp set used by bankers | Other foundation labs, chip/infra names, hyperscalers | Public SaaS index, ARR-per-employee peers |
| Rule of 40 relevance | Mostly fails it, priced anyway | Central screen — 11.7x above 40%, 3.9x below 20% |
| Downside case priced in | Model commoditization, compute cost deflation | Churn, competitive displacement, seat compression |
Figures blended from OpenAI/Anthropic IPO disclosures, Aventis Advisors, L40, and the Value Add VC SaaS Valuations dashboard. Rule of 40 multiples per L40 2026 public software data.
The Numbers: OpenAI, Anthropic, and Cerebras Against SaaS Comps
OpenAI is the clearest data point. At an $852 billion valuation against roughly $25 billion in annualized revenue, the company is priced at approximately 34x revenue — a multiple that would be unthinkable for a SaaS company of any size, but is treated as reasonable for a foundation-model leader still compounding revenue at triple-digit annual growth rates.
Anthropic filed for its IPO at a $965 billion valuation on roughly $47 billion in annualized revenue — about 21x, noticeably lower than OpenAI's multiple despite Anthropic overtaking OpenAI on run-rate revenue in April 2026 (roughly $30 billion versus $25 billion at that snapshot). The lower multiple isn't a vote of less confidence; it's simple math — Anthropic's revenue base is larger relative to its valuation, which mechanically compresses the multiple even as the growth story stays strong.
Cerebras, 2026's largest AI-adjacent IPO, priced at roughly $40 billion (about $49 billion fully diluted) against $510 million in 2025 revenue — a 76% year-over-year increase. At the IPO price that's close to 78x trailing revenue; the stock then popped over 100% in early trading, pushing the effective multiple well past 100x. Compare that to CoreWeave, which went public earlier in 2026 at a price-to-sales ratio near 15x — still nowhere close to SaaS territory, but roughly a fifth of Cerebras's multiple, because CoreWeave's revenue base is larger and its backlog less concentrated in a handful of customers.
Set those next to the SaaS Valuations dashboard: the median public SaaS company trades at 8.5x EV/revenue as of June 2026, up 90% from the 4.5x trough in 2023 but still 47% below the 16x peak of 2021. Even the best SaaS companies — those clearing a Rule of 40 score above 40% — only reach a median of 11.7x EV/NTM revenue. Every AI company in this comparison is priced above that best-case SaaS ceiling, several of them by a factor of three or more.
Revenue Multiple: AI IPOs vs. SaaS Benchmarks (2026)
OpenAI/Anthropic IPO filings, TechCrunch, Yahoo Finance, L40, Value Add VC SaaS Valuations dashboard
Why the AI Valuation Framework Rewards Growth Over Retention
The SaaS framework exists because SaaS businesses are, structurally, annuities — a dollar of ARR this year is worth roughly the same next year unless it churns, so the entire valuation exercise is about proving the annuity is durable. Net revenue retention above 120%, gross margin above 75%, and a defensible moat against competitive displacement are the inputs that tell an investor the annuity won't shrink.
AI foundation-model revenue doesn't behave like an annuity yet. It's growing so fast — Anthropic went from roughly $1 billion to $30 billion in annualized revenue in about fifteen months — that retention is close to a non-issue; the more relevant question is whether the growth rate itself is sustainable and whether the company can get access to enough compute to keep serving it. That's a supply-constrained framework, closer to how investors price a scarce commodity producer than how they price a software subscription business.
This is also why AI-native SaaS — companies that wrap foundation models into vertical software, trading at 25-30x per recent multiples research — sits between the two extremes. They inherit some of AI's growth premium but still get underwritten partly on retention and margin, because their revenue is closer to a recurring subscription than a foundation lab's API consumption.
What the AI IPO Valuation Framework Means for Investors and Founders
For LPs and public-market investors, the practical implication is that comping an AI IPO against a SaaS index will systematically underprice it while growth is still triple-digit, and will systematically overprice it the moment growth decelerates toward SaaS-like rates without SaaS-like retention economics to defend the multiple. The 2021-2023 SaaS correction — a 72% multiple compression — is the closest precedent for what happens when a growth-premium framework meets a slowdown, and it's worth watching for signs of the same pattern in AI once frontier-model growth rates normalize.
For founders building AI-native companies, the framework question is existential to your fundraising strategy. A company that can credibly claim foundation-model-style growth and compute scarcity gets priced at 21-38x. A company that's really a SaaS business with an AI feature gets priced — correctly — closer to the 8.5x-11.7x SaaS range, regardless of how many times "AI" appears in the deck. Investors in 2026 have gotten sharper at telling the difference, which is also why AI-native SaaS companies with genuine model integration are commanding 1.5-3x the multiple of AI-feature bolt-ons per recent private-market data.
It's also worth tracking through the Tech IPO dashboard and the AI Valuations dashboard — as more foundation labs and AI infrastructure names complete their IPOs through the rest of 2026, the sample size for this framework grows, and the gap between AI and SaaS multiples will either hold, widen, or start compressing toward each other. Right now, with OpenAI at 34x and Anthropic at 21x against a SaaS median of 8.5x, the gap is still wide open.
How Bankers Are Actually Building AI IPO Comp Sets in 2026
Talk to bankers running these books and the comp set for an AI IPO looks nothing like a SaaS S-1's comparable-companies page. Instead of pulling ARR-per-employee and NRR figures from ten public software peers, they're building a three-tier stack: other foundation labs (OpenAI, Anthropic, xAI/SpaceX post-merger at a combined $1.25 trillion valuation) for growth-rate benchmarking, AI infrastructure names (Cerebras, CoreWeave, Nvidia at roughly 25x sales) for scarcity-premium benchmarking, and — only as a downside sanity check — the SaaS index to show what happens if growth normalizes faster than expected.
That third tier matters more than it looks. Underwriters aren't ignoring SaaS multiples; they're using the 8.5x median as the floor in a bear case, not the base case. The base case still assumes AI revenue keeps compounding at 100%+ annually for another 12-24 months before growth decelerates toward a rate SaaS-style metrics could actually evaluate. Anthropic's jump from roughly $1 billion to $30 billion in annualized revenue in fifteen months is the data point underwriters point to when justifying why the SaaS floor doesn't apply yet.
The risk in that approach is the same risk every growth-premium framework has carried since the dot-com era: it works exactly until growth decelerates faster than the multiple compresses. SaaS investors learned that lesson between 2021 and 2023, when the median public multiple fell from 16x to 4.5x — a 72% compression — as growth rates that had justified premium pricing reverted to the mean within about eighteen months. AI IPO investors in 2026 are making a bet, implicitly, that either the deceleration takes longer to arrive or that AI's higher gross-margin ceiling on inference gives it more room to absorb a slowdown than SaaS had.
Bottom line: AI companies are being valued on a growth-and-scarcity framework that has almost nothing to do with the ARR-quality framework that sets SaaS multiples — which is why OpenAI at 34x, Anthropic at 21x, and Cerebras at 78x-plus all clear the ceiling of even the best public SaaS comps at 11.7x. That gap will close eventually; the only open question is how much of it closes through AI multiples falling versus SaaS multiples catching back up. Track both sides of that convergence on Value Add VC.
Get VC data most people never see — free.
Weekly benchmarks, valuations, and fund data. No spam, unsubscribe anytime.