Cursor's $20/Month Pricing Strategy: Sell AI Like a Planet Fitness Membership

Cursor has created a pricing model that seems to defy financial logic.

For just $20 per month, users get access to 500 fast generative requests plus unlimited slow ones. The math behind this pricing reveals an interesting relationship between what customers pay and what the service costs to provide.

Looking at the actual economics: each fast request consumes approximately 1,200 tokens (800 input + 400 output tokens).

With 500 requests, this equals about 600,000 tokens per month.

Using GPT-4.1, these costs break down to much less than initially appears - input costs of $0.80 (400K tokens at $2 per million) and output costs of $1.60 (200K tokens at $8 per million) for a total of just $2.40 per month in raw API costs.

Even with much lower API costs, the business faces other expenses. When considering additional technical factors such as context windows, streaming tokens, retries, agent mode (which requires multiple LLM calls per credit), and extra processing logic, costs can still add up significantly.

But it still begs the questions about the underlying business model.

Venture Capital Makes Aggressive Pricing Sustainable

Venture capital funding plays a crucial role in Cursor's pricing strategy. The company has secured substantial investment—$800 million from firms like Andreessen Horowitz (A16Z), Thrive, and Accel—allowing it to operate with thin margins or occasional losses while building market share.

This approach follows an established pattern in the tech industry, where companies prioritize growth over immediate profitability. The technology sector has numerous examples of companies that operated at a loss for years before turning profitable.

VC Subsidization Creates Artificial Market Conditions

The current pricing environment for AI tools demonstrates how venture capital creates market conditions that wouldn't exist in a purely profit-driven environment. Companies offer services at aggressive price points to capture users, knowing the economics will evolve over time.

Uber exemplifies this pattern—Manhattan Uber pool rides cost a flat $5 in 2015 (or $6 for two people). Today, a single rider pool easily costs $30, with the company having both survived and become profitable through this evolution.

Only a few AI coding tools will survive long-term. In the bigger picture, 2-3 of these tools will survive and charge actual cost plus profit, while the rest become forgotten.

Low Utilization Rates Create Profit From Inactive Users

Cursor's business relies heavily on the same economic principle that makes gym memberships profitable: most subscribers don't fully use what they pay for.

Planet Fitness charges a monthly fee of $10 per month but most people don't use the gym regularly making the business extremely profitable. This model allows the gym to generate significant revenue from inactive members.

Many subscribers sign up with good intentions but use the service minimally or not at all. Similarly, many developers pay for subscriptions for months without using them, effectively giving the company free revenue.

This "gym membership model" works because companies make money from people who appreciate having access to a service but rarely use it enough to make the relationship unprofitable. CTOs might buy licenses for entire teams on hype, but only a few engineers actually use it regularly.

Usage-Based Pricing Creates Revenue From Power Users

The basic subscription serves as an entry point, but substantial revenue comes from users who exceed their limits and move to usage-based pricing.

Entry level pricing for regular fast requests are 4 cents per task while heavy charges for MAX models are up to 130 cents per task. That's 30 times as expensive.

$20/month plan acts as a strategic entry point. This plan targets users who may not reach their limit. If they stay under the 500 request cap, Cursor makes a profit. If they exceed it, Cursor takes a small loss but converts them to a more profitable pricing tier.

The incentive for Cursor is to keep the plan threshold low and capture big spenders as fast as possible in the pricing hierarchy. The company doesn't want to give people already prepared to spend $300/month an extra set of discounted queries. That's the reason they don't have a $60/month plan.

Windsurf, a Cursor competitor, offered a $60/month ultimate plan that they have removed recently for this reason.

Premium features command premium prices: 5¢ per tool call for Gemini 2.5 Pro Max and Claude 3.7 Sonnet. While this seems expensive compared to the base subscription, businesses see it as a minimal cost to make their employees more productive.

Business Model Depends On Asymmetric Usage Patterns: Heavy Users Subsidized By Light Users

Subscription services work because user behavior varies dramatically. Some subscribers extract maximum value, while others barely touch the service.

Roughly one-third of customers use services to their maximum extent, while two-thirds underutilize them. This asymmetry creates a sustainable economic model where light users effectively subsidize power users.

SaaS businesses operate on the assumption that most people don't fully utilize the service. This extends beyond software—almost every mass-marketed product works this way.

Companies optimize pricing for average users while knowing some will barely use the product and others will push it to its limits.

Most mass-market products work this way. Imagine if Apple only sold iPhones to 1,000 people. If all of them heavily used their phones and wore out the batteries quickly, the cost of handling battery warranties would likely raise the price for everyone. But in the mass market, where Apple sells to millions of people, they can set a price that works for the average user. Some people use their phones lightly, some heavily, and most are in between. This way, everyone helps cover the warranty costs for those who use their phones a lot and wear out the batteries faster.

Enterprise Pricing Creates Substantial Provider Discounts

Cursor benefits from economies of scale that individual developers cannot access.

Unlike individual developers who pay retail prices for API access, Cursor likely receives significant volume discounts from AI providers like OpenAI, Anthropic, and Google.

AI IDE tools represent an important investment in the future of how these tools will be used. This strategic importance earns them incredibly discounted rates compared to individual developers or small businesses.

Cursor's constant, high-volume usage makes them valuable customers for AI companies. While an individual developer has inconsistent usage and low profitability, platforms like Cursor use APIs constantly and in large volume, making them more profitable for AI providers to serve.

Context Limitation And Caching Create Cost Savings

Cursor employs various techniques to minimize token usage and maximize efficiency.

The platform uses heavy caching for similar requests, summarizes and limits context, and optimizes system prompts. While these techniques sometimes cause challenges for developers, they significantly reduce costs.

Predictable usage patterns allow Cursor to right-size resources, and the fast/slow request mechanism helps with load balancing. Context caching substantially cuts down costs, especially for Anthropic models as Claude allows caching.

Caching means if you call the same API endpoint with the same parameters (let's say create an image of a rocket), the response will be cached (stored in one location for more than 1 hour) and returned immediately (whenever you call the same prompt, i.e, create an image of a rocket), reducing the need to pay for additional API calls.

The mechanism of fast/slow requests plays a role in resource allocation, using already paid-for resources but shifting queue priorities to help with load balancing requests to lower slot/slower servers.

User Data Creates Alternative Value Streams: Code Example Collection Builds Proprietary Training Data

Data is the new oil. Access to developer workflows and code examples provides additional value beyond subscription revenue.

Cursor likely collects data that helps build better coding models. Having programmers accept/decline coding suggestions provides exceptionally valuable information for model training. This is similar to RLHF (Reinforcement Learning from Human Feedback).

TikTok moderators job was something similar. The mods were forced to left-swipe videos of poor, ugly, and disabled people to make the feed more attractive and addictive.

Similarly, OpenAI gives free credits to developers who share code as mentioned in GPT 4.1 Livestream, suggesting data sharing has tangible value in the AI ecosystem.

This creates a virtuous cycle where user interactions improve the product, making it more valuable to both current and future users.

Blitzscaling Model Ends In Price Hikes

The current pricing environment cannot last forever. History suggests prices will rise once companies build sufficient market dominance.

Cursor follows a "blitzscaling" approach, designed to operate at a loss until a sizable number of clients rely on them. Once users become dependent on the service, prices will be hiked.

The pattern repeats across tech services: Hulu once offered better value, Netflix used to be cheaper, and Google once displayed just two simple ads.

The long-term goal might eventually be charging $2,000/month for an AI that functions as a mid-level engineer driven by an employee earning $5,000/month rather than hiring a $12,000/month senior engineer.

AI Model Pricing Decline May Save Current Business Model

Despite the challenges, falling AI costs might make Cursor's current pricing sustainable in the long run.

AI model pricing continues to drop, with newer options like DeepSeek, Gemini 2.5 Pro, and ChatGPT 4.1 offering better economics. Cursor likely expects LLM costs to decline, improving margins over time without raising prices.

This strategy works if user engagement scales and cost curves drop as expected. If AI costs fall faster than user consumption rises, the business model becomes sustainable without price increases.

Competition With Other AI Coding Tools Intensifies

Cursor faces growing competition from established AI Providers and newer startups like Windsurf, rumoured to be purchased by OpenAI for $3 billion.

Anthropic's Claude Code takes a different approach, with zero price subsidization—users pay for API usage on regular terms. This contrasts with Cursor's subsidized model.

OpenAI also released Codex today, a competitor to Claude Code.

Microsoft GitHub Copilot offers 300 requests (including Claude 3.7 Sonnet Thinking) for $10/month. As a deep-pocketed competitor, Microsoft can offer services for free until competition dies out.

The release of VSCode Agent demonstrated that Cursor lacks a significant moat protecting its business, suggesting the Cursor's competitive advantage might be temporary.

For now, developers benefit from this unusual economic environment. Whether through VC subsidies, the gym membership effect, or future cost reductions, today's pricing creates an opportunity for developers to access powerful AI coding tools at a fraction of their true cost.

Top Tweets of the day

1/

Scrolling TikTok while only looking for info is extremely hard since its addictive with all the entertainment.

2/

ChatGPT memory gives ChatGPT a MOAT and everyone will buy their own accounts instead of sharing one between family now.

Love this prompt: "remove any self-care bullshit"

3/

o4-mini-high is extremely slow but its real good at tool use.

This is what it means to have experts in their own field. The LLMs output is going to get a hell of a lot better.

Some like Tyler Cowen are calling it AGI. Now it doesn't matter how intelligent you are cause the AI went above human intelligence.

Rabbit Holes

1. How We Build Effective Agents: Barry Zhang, Anthropic by AI Engineer

2. Where Everyday Human Reasoning Still Surpasses Frontier Models by SimpleBench

3. Reaching a $100K MRR within 12-18 months of a product launch in a B2B SaaS seems to be the ideal growth rate recommended by most investors. How can founders structure their startup, product, and sales before launching to hit those numbers? by Jason Lemkin

What’d ya think of today’s newsletter? Hit ‘reply’ and let me know.

Do me a favor and share it in your company's Slack #marketing channel.

First time? Subscribe.

Follow me on X.

More Startup Spells 🪄

1. X Growth Hack: Piggyback on High-Follower Accounts using Quote-Tweets (LINK)

2. How Honey Stole Money from Influencers (Scam) (LINK)

3. From $0 to $30M exit in 9 weeks (LINK)

4. The Ultimate Guide To Social Listening (LINK)