How Does Cursor Make Money: $20-$200 Subscriptions, Token Pricing, and the Business Model Breakdown

$4 billion in annualized revenue is what Cursor's parent Anysphere hit by June 2026, up from roughly $1 billion a year earlier, with $2.6 billion now coming from enterprise contracts. That's the short answer. The longer answer is a token-metered subscription model that charges a markup on every AI model call while quietly nudging usage toward Cursor's own cheaper in-house model.

Cursor isn't a simple SaaS subscription business. It's a reseller of frontier AI model inference wrapped in a code editor, and understanding how it prices that inference — and who's actually paying for it — explains both its extraordinary growth and why SpaceX just agreed to buy the whole company for $60 billion.

How does Cursor make money?

Cursor makes money by selling tiered monthly subscriptions — from a free Hobby plan up to a $200/month Ultra plan — bundled with a metered pool of AI model usage, then charging enterprise customers custom, per-seat contracts for pooled usage, SSO, and admin controls. On every premium model call beyond the included credit pool, Cursor charges the underlying provider's API price plus roughly a 20% margin, making usage-based markup — not the subscription fee itself — the core profit engine.

That structure is why Cursor's revenue scales so much faster than a typical dev-tools subscription business: every additional prompt a developer runs is itself a billable, margin-bearing transaction, not just a seat that renews once a year.

Cursor's pricing tiers: how the subscriptions actually work

Cursor runs six pricing tiers in 2026, each bundling a different amount of AI model usage rather than just gating features. The table below breaks out what each tier costs and what it includes.

Figures are July 2026 pricing blended from Cursor's public pricing page, DEV Community, and Eesel AI pricing breakdowns. Annual billing applies a 20% discount across all paid plans.

Why enterprise deals are becoming Cursor's real business model

Of Cursor's roughly $4 billion in ARR by June 2026, about $2.6 billion — nearly two-thirds — comes from enterprise B2B contracts rather than individual $20-$200 subscriptions. That mix shift matters because enterprise deals bundle pooled usage across hundreds or thousands of seats, which both raises average revenue per account and gives Cursor more pricing power to protect its margin on every model call, since large customers negotiate custom terms rather than defaulting to list price.

It also explains why Cursor is increasingly compared to enterprise data and AI infrastructure platforms rather than to a consumer productivity app — the company's growth engine now looks a lot more like an enterprise software vendor upselling seats and usage than a $20/month indie-developer tool.

The token-metering trick that protects Cursor's margin

Every paid Cursor plan splits usage into two separate pools: an Auto + Composer pool that runs on Cursor's own in-house model, and a more expensive API pool for third-party frontier models like Claude, GPT, Gemini, and Grok, billed at each provider's own API price plus roughly a 20% margin under Cursor's Max Mode. The economics of those two pools are wildly different — Cursor's own Composer 2.5 model costs about $0.50 per million input tokens and $2.50 per million output tokens, versus $5.00 input and $25.00 output for Claude Opus 4.7/4.8, a 10x cost gap.

By giving Composer its own separate, cheaper allowance and automatically falling back to it once a user exhausts third-party credits, Cursor structurally nudges usage toward the inference it controls and prices itself. That's the real margin lever in the business: subscription revenue funds growth, but the token-metering design is what determines whether each incremental prompt is profitable or a cost center.

What the $60B SpaceX deal says about Cursor's business model

SpaceX agreed on June 16, 2026 to acquire Anysphere in an all-stock deal valued at $60 billion — essentially double the $29.3 billion valuation Anysphere had priced at just seven months earlier in its November 2025 Series D. The valuation jump roughly tracks the revenue jump: ARR nearly doubled from about $2 billion in February 2026 to $4 billion by June, and acquirers were willing to pay a premium multiple on top of that growth given how much of the new revenue is high-margin enterprise usage rather than discounted individual subscriptions.

For anyone tracking how AI companies get priced, the Cursor deal is a useful data point: a company with a clear, usage-metered revenue model and a fast-growing enterprise mix commanded a premium even in stock rather than cash, because the acquirer is underwriting continued ARR growth, not just buying a static asset.

How Cursor's business model compares to other AI coding tools

GitHub Copilot largely monetizes through flat per-seat pricing bundled into Microsoft's broader enterprise agreements, while Windsurf and other competitors have leaned on simpler flat-rate subscriptions without Cursor's dual-pool token metering. That difference matters for margin: Copilot's flat pricing caps Microsoft's exposure to heavy users but also caps upside, whereas Cursor's usage-based markup captures more revenue from power users and enterprise teams running agents constantly, at the cost of more pricing complexity and periodic user backlash over "tokenomics" changes.

We compared the full competitive field, pricing models included, in our AI coding tools ranked 2026 breakdown — Cursor's willingness to price directly to compute is the single biggest structural difference from its closest competitors. That willingness came with real customer friction: Cursor quietly reworked its pricing in mid-2025 after complaints that unlimited-sounding plans were actually gated by opaque usage caps, and the 2026 tiers, with their explicit credit pools and published per-token rates for Max Mode, are a direct response to that backlash. Being transparent about token costs is unusual in this category, and it's arguably part of why enterprise buyers — who need predictable line items for procurement — have been willing to sign the custom contracts that now make up 65% of revenue.

What Cursor's model means for AI infrastructure investors

Cursor is a useful case study for anyone underwriting AI-native software companies, because it shows what happens when a vendor's cost of goods sold scales directly with usage rather than staying fixed like traditional hosting. That's a fundamentally different risk profile from a classic SaaS company: gross margin depends on the spread between what a company charges per token and what the underlying model provider charges, and that spread can compress overnight if a frontier lab cuts API prices or a competitor undercuts on markup. Cursor's answer — building and steering usage toward its own cheaper Composer model — is the same playbook OpenAI, Anthropic, and other model labs use to protect margin as they move down the stack toward end-user applications.

It's also why diligence on AI-native companies increasingly has to ask "what percentage of revenue is pass-through inference cost versus actual gross profit" rather than just looking at top-line ARR growth, since two companies with identical $4 billion ARR figures can have very different underlying economics depending on how much of that revenue is being handed straight to a model provider. Cursor's 20% markup on third-party model usage, layered on top of a cheaper proprietary fallback model, is a relatively investor-friendly structure compared to competitors reselling frontier model access at cost or at a thinner margin just to stay competitive on price.

For LPs and allocators tracking this category through AI valuation multiples, the practical takeaway is that ARR growth alone is an incomplete signal — the mix between enterprise and individual revenue, and the gross margin embedded in the token-pricing structure, matter just as much to whether a $4 billion ARR business is actually worth the $60 billion an acquirer is willing to pay for it.