- A 34-AI-startup cohort is pulling in close to $80 billion in annual revenue, up 112% in six months, per analysis by The Information.
- Anthropic and OpenAI capture 89% of that revenue between them.
- Anthropic recently passed OpenAI in revenue, driven by AI coding tools.
- Perplexity, ElevenLabs, and Cognition have each crossed $500 million in annual revenue.
What Happened
A 34-AI-startup cohort is pulling in close to $80 billion in annual revenue — up 112% in six months — but Anthropic and OpenAI capture 89% of that revenue between them, The Decoder reported on Monday, citing The Information. Anthropic recently passed OpenAI in revenue, largely driven by AI coding tools — consistent with Ramp’s AI Index released last week.
Why It Matters
The 89% concentration data point crystallises a debate that has been escalating across the AI venture-capital community: whether the bulk of AI value accrues to model makers or to application-layer companies. Sequoia Capital and other tier-one investors have publicly argued the model-maker thesis. The Information’s revenue concentration figure supports that view at the current snapshot, though the framing comes with significant caveats.
The caveats matter. The two leaders also burn through more than $30 billion a year combined, mostly on training costs. Anthropic shares revenue with Amazon and Google. OpenAI has to hand 20% to Microsoft through 2030. Some of the headline revenue numbers therefore overstate net economic value flowing to the model makers.
Technical Details
The 34-startup cohort The Information analysed represents the most commercially active AI startups. Beyond Anthropic and OpenAI, three companies — Perplexity, ElevenLabs, and Cognition — have each crossed $500 million in annual revenue. The remaining 29 in the cohort collectively account for roughly 11% of total revenue. Aggregate revenue growth of 112% over six months implies a roughly trailing-12-month doubling rate for the category.
The cash-burn figure — $30 billion+ for the top two combined — is concentrated in training-cluster operating costs and the multi-year buildout of frontier-tier model capability. Inference costs scale with usage but are typically priced above marginal cost on enterprise contracts.
Who’s Affected
Anthropic and OpenAI investors gain an external data point validating the revenue-concentration thesis. Application-layer AI startups face a sharper question about their structural margin profile and competitive moat. Perplexity, ElevenLabs, and Cognition cross a meaningful psychological threshold by reaching $500 million ARR. Venture-capital firms that bet on horizontal model makers (Sequoia, Founders Fund, Andreessen Horowitz on the OpenAI cap table) gain validation; firms that bet primarily on application-layer companies face renewed scrutiny. The broader implication is that AI VC strategy is bifurcating into model-maker bets and a long tail of application-layer plays with thinner margins.
What’s Next
Expect quarterly updates from The Information and other AI-revenue trackers to refine the 34-startup cohort and concentration figures. Anthropic and OpenAI’s anticipated public listings (or further private rounds) will produce additional disclosure on the revenue and cost economics. Mid-tier application-layer companies — Notion, Replit, Cursor, Glean, Lindy, Decagon — face the question of whether they can scale into the model-maker revenue band or settle into a margin-constrained application tier.
