JetBrains slips to its first net loss on record revenue of $710 million

JetBrains lost about 2% of revenue in 2025 ^[8], in line with the negative margin reported in the dev.to post ^[5]. EBITDA fell much further. Working back from the 58.3% annual drop ^[3], EBITDA was about CZK 2.2 billion in 2024 ^[12]. Over two years it fell about 78% ^[13].

Revenue kept growing the whole time ^[2]. According to the post, annual growth slowed from 20% to 6.3% over four years while costs climbed ^[6]. In the same year sales rose about 6% and EBITDA fell by more than half, so costs grew faster than sales [c3, c7].

EBITDA is still positive. The CZK 315 million loss converts to $14 million in the post, a rate of about 22.5 koruna per dollar ^[9]. At that rate, 2025 EBITDA comes to roughly $41 million ^[10], close to 6% of revenue ^[11]. Using one rate for a full year of koruna earnings is an approximation.

All of these figures come from one post. It does not name the filing behind them or give the base for the 240% growth in AI paid users ^[16]. Nor does it break down where the extra cost went or report any change to what JetBrains charges. On this record, new pricing or bundling for IDE customers is a forecast. By its own account, the margin went into agents and models ^[15].

Competition is how the post explains the squeeze. It argues that coding agents belong in the terminal, and it names Claude Code, OpenAI's Codex CLI and Gemini CLI. Its reasons are that agents need shell access and filesystem access, and that parallel sessions can be isolated in separate worktrees ^[17]. It then compares JetBrains with Cursor. According to the post, Cursor went from zero to $4 billion in annual recurring revenue in under three years, the fastest SaaS revenue growth in history ^[18].

The better case for JetBrains comes from the same post. Its contrarian section says JetBrains is betting that code is getting cheaper to generate and more expensive to verify. On that view, an IDE's diffing, type checking and test visualisation matter more when agents write the code ^[19]. I think JetBrains has picked the right bet. In the post's telling, the IDE sold deep language intelligence, integrated tooling and less for the developer to keep in their head ^[21]. Checking agent output needs all three. The post leaves open whether review work justifies a subscription ^[20].