The hottest AI startup in the world has a $29 billion valuation, over $1 billion in annualized revenue, a beloved product, and one of the fastest growth curves in tech history. It’s even admired by other AI winners like Nvidia and OpenAI.

Yet it’s losing money, and no one is sure it will exist in a few years.

I’m talking about Cursor, the explosive AI coding startup that, according to the Wall Street Journal, is burning cash and facing real questions about long-term viability.

It’s not a cheap product either. Cursor is more expensive than established developer tools like GitHub or Jira.

Cursor reportedly has about 250 employees. To be conservative, let’s assume an all-in people cost of $350,000 per employee for an SF high-growth software company. That puts them at around $87 million in payroll. If 80% of a startup’s costs are people, that suggests roughly $108 million in annual operating costs.

That leaves an enormous gap between revenue and expenses. If you take the numbers at face value, Cursor could be spending upwards of $850 million annual run rate on hosting and model inference costs to OpenAI, Google, and others. Even if they build their own models, the hard costs of data centers and GPU buildout don’t disappear.

This is the core challenge: compute-heavy AI businesses simply don’t enjoy SaaS-level gross margins. The familiar 80% gross margin benchmark at scale can no longer be taken for granted.

Cursor’s pricing isn’t totally crazy, they actually use some components of smart AI pricing. They have a hybrid model: seat-based tiers with usage caps, differentiated by model type and request limits. They offer a one-week free trial, so there’s low freemium burden. And the price isn’t low. This isn’t Disney+ launching at $6.99 just to grab attention.

Yet the economics still don’t work.

So what can we learn from Cursor’s challenges?

Why does Cursor’s entry tier cost $20/month? Probably because OpenAI anchored ChatGPT Plus at $20/month. And OpenAI admitted that number was essentially a guess. Cursor likely adopted it for psychological consistency in a crowded market.

Given their age and breakneck growth, it’s doubtful they set aside time for deep willingness-to-pay research. The $60 and $200 Pro and Ultra tiers also seem more like educated guesses than data-backed pricing.

Enterprise buyers make a business case before they buy. Developer time is the most expensive resource inside a software company. Developers are hard to hire, hard to retain, and slowdowns are costly.

Cursor’s viral adoption and rave reviews signal enormous productivity value. A more rigorous look at quantified savings — dollars, hours, throughput — would produce better reference price points. They should have worked pricing backwards from ROI, rather than copying competitors.

There’s one piece of Cursor’s pricing I don’t love:

“On-demand usage is billed monthly at the same rates as your included usage. Requests are never downgraded in quality or speed.”

Flat-rate overages discourage upgrades. Heavy users — who are also the most expensive for Cursor to support — face no pressure to move up to a higher tier. These customers get the highest value, use the most compute, and cost the most money. They should be incentivized to commit to larger fixed-cost plans. Customers should be asking you to upgrade to secure better terms. The feature differentiation between plans is not that large, common for a startup. So they need to rely more on usage to encourage migrations.

Pricing is a lever to nudge customers to premium plans.

Limits set too low frustrate customers. But limits set too high — especially early — can destroy unit economics.

AI usage is still an unknown for most companies. Teams don’t know who will use it, how often, or for what workloads. Cursor was generous with its limits to accelerate adoption. That choice now locks them into costly dynamics.

They’re stuck between two rough options:

A more cautious, experimental approach to usage limits early on would have given them room to maneuver while they searched for a cost breakthrough.

AI vendors want approachable entry pricing but also need to protect margins. For compute-heavy products, per-user pricing is mismatched to costs.

Costs scale with usage. Revenue scales with seats.

That gap creates an arbitrage opportunity customers will take advantage of. 

Many AI companies will need to shift further toward hybrid or usage-centric models:

Aligning revenue to track compute more closely will protect margins.

Whether Cursor can find a rabbit in the hat to fix its model is unclear. Beyond lowering model costs, they could monetize user data, add higher-value features, or expand into new personas beyond developers.

The problem is that In AI, scale doesn’t automatically fix your economics.

Compute doesn’t compress the way traditional software does. Gross margins don’t necessarily improve with time. Pricing discipline matters from day one.

Cursor is a warning shot. 

Before you go to market with an AI-heavy product, put your pricing model through a tougher stress test. Don’t assume it will all work itself out later with a pricing tweak. You need to go to market with a smarter, future-proof model from day one.

“Day one is when the gap forms” - Unknown

Have questions or feedback? Just reply to this. I read every email.