In February, AI coding startup Windsurf was in talks to lift an enormous new spherical at a $2.85 billion valuation led by Kleiner Perkins, at double the valuation it hit six months earlier, sources advised TechCrunch on the time. That deal didn’t occur, based on a supply accustomed to the matter. As an alternative, information broke in April that the startup deliberate to promote itself to OpenAI for roughly the identical valuation: $3 billion.
Whereas that deal famously fell aside, one larger query stays: If the startup was rising that quick and attracting VC curiosity, why wouldn’t it promote in any respect?
Insiders inform TechCrunch that for all the recognition and hype round AI coding assistants, they will truly be massively money-losing companies. Vibe coders usually, and Windsurf particularly, can have such costly constructions that their gross margins are “very damaging,” one individual near Windsurf advised TechCrunch. Which means it value extra to run the product than the startup may cost for it.
That is as a result of excessive prices of utilizing giant language fashions (LLMs), the individual defined. AI coding assistants are significantly pressured to at all times provide the newest, most superior, and costliest LLMs as a result of mannequin makers are significantly fine-tuning their newest fashions for enhancements in coding and associated duties like debugging.
This can be a problem compounded by fierce competitors within the vibe-coding and code-assist market. Rivals embody corporations that have already got large buyer bases like Anysphere’s Cursor and GitHub Copilot.
Probably the most easy path to bettering margins on this enterprise includes the startups constructing their very own fashions, thereby eliminating prices of paying suppliers like Anthropic and OpenAI.
“It’s a really costly enterprise to run in the event you’re not going to be within the mannequin sport,” stated the individual.
Techcrunch occasion
San Francisco
|
October 27-29, 2025
However that concept comes with its personal dangers. Windsurf’s co-founder and CEO, Varun Mohan, in the end determined towards the corporate constructing its personal mannequin — an costly enterprise, the individual stated.
As well as, mannequin makers are already competing straight. Anthropic affords Claude Code and OpenAI affords Codex, as an illustration.
Promoting the enterprise was a strategic transfer to lock in a excessive return earlier than it may very well be undermined by the very corporations that provided its AI, together with OpenAI and Anthropic, which had been additionally getting into the AI coding market.
A number of folks imagine that the identical strain on margins Windsurf confronted may very well be impacting Anysphere, the maker of Cursor, in addition to vibe coders like Lovable, Replit, and others.
“Margins on all the ‘code gen’ merchandise are both impartial or damaging. They’re completely abysmal,” stated Nicholas Charriere, founding father of Mocha, a vibe-coding startup and back-end internet hosting resolution serving small and medium companies (SMBs). He added that he believes the variable prices for all of the startups within the sector are very shut, seemingly inside 10% to fifteen% of each other.
In contrast to Windsurf, Anysphere has been rising so quick that it intends to stay an unbiased firm, having already turned down acquisition affords, together with, stories say, from OpenAI.
And Anysphere introduced in January that it’s making an attempt to construct its personal mannequin, which may give it extra management over its bills. In July, the startup employed two leaders from Anthropic’s Claude Code workforce, the Info reported, however two weeks later, these workers returned to work at Anthropic.
Along with constructing a mannequin, Anysphere may anticipate the price of LLMs to lower over time.
“That’s what everybody’s banking on,” stated Erik Nordlander, a normal associate at Google Ventures. “The inference value as we speak, that’s the most costly it’s ever going to be.”
It’s not fully clear how true that’s. Somewhat than falling as anticipated, the price of among the newest AI fashions has risen, as they use extra time and computational sources to deal with difficult, multistep duties.
When that may change stays to be seen. On Thursday, as an illustration, OpenAI launched a brand new flagship mannequin, GPT-5, with charges which are considerably lower than its competitor, Anthropic’s Claude Opus 4.1. And Anysphere instantly provided this mannequin as a alternative for Cursor customers.
Anysphere has additionally not too long ago modified its pricing construction to move alongside the elevated prices of working Anthropic’s newest Claude mannequin, significantly to its most lively customers. The transfer caught a few of Cursor clients unexpectedly, since they didn’t anticipate extra prices on prime of its $20-per-month Professional plan. Anysphere CEO Michael Truell later apologized for unclear communication concerning the pricing change in a weblog put up.
That is the rock and the onerous place. Though Cursor is without doubt one of the hottest AI purposes, having reached $500 million in ARR in June, the corporate’s person base might not be so loyal to the product if one other firm develops a device that’s superior to Cursor, traders say.
Anysphere didn’t reply to a request for remark.
Given the aggressive panorama and the prices, Windsurf’s determination to get out could show to be comprehensible. After the OpenAI deal fell by way of, the founders and key workers left to hitch Google in a deal that led to a $2.4 billion payout to key shareholders. The remaining enterprise then offered itself to Cognition.
Whereas many, together with distinguished VCs, criticized Mohan for leaving roughly 200 workers with out roles at Google, a supply accustomed to the deal insisted the acquisition truly maximized the outcomes for all workers.
Past Cursor, different AI coding instruments are additionally among the many quickest rising startups of the LLM technology, like Replit, Lovable, and Bolt, and all of them depend on mannequin makers as effectively.
Moreover, if this extraordinarily well-liked enterprise sector, already producing lots of of tens of millions in income or extra a yr, has problem constructing on prime of mannequin makers, what may it imply for different, extra nascent industries?
