Faced with high expenses, software development has fast become the cornerstone for generative AI startups, two years after OpenAI’s ChatGPT went viral, spurring massive developments in the sector.
Known as “code generation” or “code-gen,” these AI tools are attracting enormous valuations as corporate executives explore ways to streamline or even replace costly human programmers.
AI startups must fend off stiff competition from established firms
San Francisco’s Cursor, which can autocomplete single lines and draft entire code segments autonomously, exemplifies the frenzy. In May, it secured a $900 million investment round that valued the company at $10 billion, with backing from Thrive Capital, Andreessen Horowitz and Accel.
Meanwhile, Mountain View startup Windsurf, best known for its Codeium tool that turns natural-language commands into working code, has reportedly held acquisition talks with OpenAI at a $3 billion price tag.
These deals underscore the sense among founders and investors that the window to capture developer mindshare is closing fast, establish your AI assistant now, or risk being sidelined.
Despite these sky-high private valuations, code-gen outfits face ever-rising costs for every API call, since most rely on foundation models from OpenAI, Anthropic or DeepSeek. None are profitable, and all must fend off competition from the likes of Google, Microsoft, and OpenAI itself.
In May alone, each of these giants unveiled, or confirmed they were developing, new AI coding solutions. Microsoft’s GitHub Copilot, the early leader since its 2021 launch, reportedly generated over $500 million in revenue last year and now serves more than 15 million users.
Google says over 30% of its internal code is AI-generated, while Amazon claims to have saved the equivalent of 4,500 developer-years through AI tools.
Satya Nadella added that Microsoft itself uses AI to write about 20–30% of its code, even as the company recently laid off 6,000 employees, over 40% of whom were software engineers in Washington state.
AI startups deal with huge expenses amid rapid growth
Investor sources paint a picture of fast-growing revenues paired with negative gross margins. Cursor, with just 60 staff, has ramped from zero to roughly $100 million in annual recurring revenue by January 2025, its second birthday, while Windsurf, four years old, has hit $50 million in run-rate sales since launching its coding product last November.
Yet each dollar earned is overshadowed by the fees these startups pay per AI query. “Coding assistant costs are only going to rise,” warns Quinn Slack, CEO of Sourcegraph.
Founders in their twenties, many fresh from MIT, are putting in “the hardest work since the first Internet boom,” according to Andreessen Horowitz partner Martin Casado.
But as Redpoint Ventures’ Scott Raney notes, success won’t hinge solely on algorithmic prowess.
“It’s who can market and sell the technology most effectively.”
Raney
And to reduce reliance on external foundation models and the resulting per-query expenses, several code-gen players are striving to train their own. Windsurf recently rolled out its first self-hosted models fine-tuned for software engineering.
Cursor has quietly assembled a research team to develop its own large “frontier-level” models, aiming to cut hefty licensing fees.
However, building a proprietary AI core is a monumental investment in compute and data. Poolside, with over $600 million raised, has partnered with AWS to test its forthcoming model but has yet to launch. Magic Dev, backed with nearly $500 million since 2023, promised a frontier coder last summer but remains silent on delivery. Others, like Replit, have abandoned in-house ambitions after assessing the scale of the challenge.
As code-gen startups jockey for leadership, the broader picture remains uncertain. The question remains if these specialized players will retain enough of their early adopters once the tech giants roll out more integrated, enterprise-ready offerings. Or they will become acquisition targets in a consolidation wave, an outcome that could both validate and vault them into the next stage of AI-driven software creation.
The market only waits to see whether this gold rush yields lasting fortunes.
The smartest crypto minds already read our newsletter. Want in? Join them.
