AI

High costs and thin margins threatening AI coding startups

In February AI Coding Startup Windsurf was in conversation to set up a large new round on a rating of $ 2.85 billion under the leadership of Kleiner Perkins, on double the appreciation that it struck six months earlier, sources WAN told at the time. That deal did not happen, according to a source that is familiar with the issue. Instead, the news broke in April that the startup was planning to sell itself to OpenAI for roughly the same rating: $ 3 billion.

Although that deal fell famous, there is a bigger question: if the startup grows so fast and attract VC interest, why would it sell completely?

Insiders say techcrunch that they can actually be massive loss of money for all popularity and hype around AI coding assistants. Atmosphere coder in general, and in particular windsurf, may have such expensive structures that their gross margins are ‘very negative’, a person told WAN close to Windsurf. Which means that it costs more to run the product than the startup for it could charge.

This is due to the high costs of using large language models (LLMS), the person explained. AI coding assistants are mainly put under pressure to always offer the most recent, most advanced and most expensive LLMs, because model makers in particular coordinate their latest models for improvements in coding and related tasks such as error detection.

This is a challenge that has been exacerbated by fierce competition in the market for atmospheric coding and code-assist. Rivals include companies that already have huge customer bases, such as the cursor and Github Copilot from Anysphere.

See also  Rocket MortGage pays up to $ 5,000 in closing costs for tenants

The simplest path to improve the margins in this company includes the startups that build their own models, which eliminates the costs of paying suppliers such as anthropic and openi.

“It is a very expensive company to run if you don’t sit in the model game,” said the person.

WAN event

San Francisco
|
27-29 October 2025

But that idea comes with his own risks. Windsurf’s co-founder and CEO, Varun Mohan, finally decided against the company that built his own model, the person said.

Moreover, model makers already compete immediately. For example, Anthropic offers Claude Code and OpenAi, for example, offers Codex.

The sale of the company was a strategic step to lock a high efficiency before it could be undermined by the companies that have delivered his AI, including OpenAi and Anthropic, which also entered the AI coding market.

Several people believe that the same pressure on margins Windsurf can be confronted on Anysphere, the maker of Cursor, as well as atmospheric codes such as Lovable, Replit and others.

“Margins on all ‘code gene’ products are neutral or negative. They are absolutely terrible,” said Nicholas Carriere, founder of Mocha, an atmospheric startup and back-end hosting solution for small and medium-sized companies (SMBs). He added that he believes that the variable costs for all startups in the sector are very close, probably within 10% to 15% apart.

In contrast to windsurf, Anysphere has grown so quickly that it is planning to remain an independent company that have already rejected acquisition bids, including reports, from OpenAi.

See also  Beyond math and coding: New RL framework helps train LLM agents for complex, real-world tasks

And AnSphere announced in January That it tries to build his own model, which could give more control over his expenses. In July, the startup Rent two leaders from the Claude code of Anthropic Team, the information reported, but two weeks later, these employees returned To work with anthropic.

In addition to building a model, Anysphere could expect that the costs of LLMS will fall over time.

“That is what everyone is banking,” said Erik Nordlander, a general partner at Google Ventures. “The conclusion costs today, that is the most expensive it will be.”

It is not entirely clear how that is. Instead of descending as expected, the costs of some of the newest AI models have risen because they use more time and computational sources to handle complicated, multistap tasks.

When that will change, it can still be seen. On Thursday, for example, OpenAI introduced a new flagship model, GPT-5, with reimbursements that are considerably less than his competitor, Anthropic’s Claude Opus 4.1. And AnSphere Immediately offered This model as a choice for cursor users.

Anysphere has also recently changed its price structure to pass on the increased costs of carrying out the newest Claude model from Anthropic, in particular for the most active users. The move caught some of the cursor customers, because they did not expect any extra costs on top of the pro-plan of $ 20 per month. Anysphere CEO Michael Truell later apologized for unclear communication about the price change in one Blog post.

This is the rock and the hard place. Although cursor is one of the most popular AI applications, the company’s user base of the company may not be that loyal to the product if another company is a tool that is superior to cursor, investors say.

See also  Startups launch products to catch people using AI cheating app Cluely

Anysphere did not respond to a request for comment.

Given the competitive landscape and the costs, the decision of Windsurf to get out can prove to be understandable. After the OpenAI deal had endured, the founders and important employees left to become a member of a deal that led to a payment of $ 2.4 billion in important shareholders. The remaining company then sold itself to cognition.

While many, including prominent VCs, criticized Mohan for leaving around 200 employees without a role at Google, a source that was familiar with the deal, it insisted that the acquisition actually maximized the results for all employees.

In addition to Cursor, other AI coding tools also belong to one of the fastest growing startups in the LLM generation, such as replit, sweet and bolt, and they also rely on model makers.

Moreover, if this extremely popular business sector, which already generates hundreds of millions of income or more per year, has difficulty building on top of model makers, what could it mean for other, more emerging industries?

Source link

Back to top button