JetBrains did not lose money selling software in FY2025. The Czech statutory accounts for JetBrains s.r.o. show CZK 315 million of red ink — about $14 million at media conversion rates, an approximation rather than a filed figure — on record revenue of CZK 16.01 billion, up 6.3% year over year, with operating profit (EBIT) still positive at CZK 751 million. [1][2] It is the first annual net loss in tracked financial history, and the statutory filing behind it is real and public: a 45-page bundle of annual accounts and annual report entered the Czech commercial register's document collection on August 20, 2026. [3][4] The question running through the developer discussion — first loss, so does this end an era? — splits in two. The loss itself is a financing-and-currency event. The era question lives somewhere else in the filing: costs that grew about five times faster than revenue, and a gross margin that fell from 53.7% in 2021 to 10.7% in 2025. [1][2] If you read one set of numbers this week, read those, not the headline.

Why the ink went red

The direct cause is arithmetic in the financing lines. Net financing costs reached CZK 1,078 million in FY2025; the year before, the same line produced CZK 810 million of financing income — a swing of roughly CZK 1.9 billion that by itself exceeds the net loss. [2] The operating side never stopped earning: EBIT of CZK 751 million, return on equity collapsing from 96.0% to −11.1% on the financing items below the operating line, and CZK 7,584 million of net cash (about $336 million, again an approximation) with CZK 1,444 million in cash and equivalents at year-end. [1][2] A company with those balances is not in a liquidity crisis.

What sits inside the financing line matters for how the loss should be read. Interest expense was only CZK 21.088 million for the year, while CZK 1,531.1 million was booked as other financial expenses — per ic.work's reading of the filing's Note 26, mostly realized and unrealized foreign-exchange losses from cross-border settlement. [1] ByteNote, cross-checking the same filing, cautions that the filing does not break the financing costs down and treats the currency-and-revaluation explanation as an inference rather than a disclosed fact. [2] Both readings agree on what is filed and verifiable: operations earned a profit, and the financing lines swallowed it. These are the standalone statutory accounts of the Prague entity, not a group consolidated report; consolidated reporting and transfer pricing sit with the Amsterdam parent, according to ic.work's account. [1]

The real alarm is the margin line

The part worth worrying about is not the loss. Gross margin ran 53.7% in 2021, 35.2% in 2023, 18.8% in 2024, and 10.7% in 2025. [1] EBITDA margin — earnings before interest, taxes, depreciation and amortization, a rough cash-profit proxy — went from 48.9% in 2021 to 5.73% in 2025. [1] The absolute figure fell from CZK 4,155 million in 2023 to CZK 918 million in 2025, about a 78% drop across those two years on those figures; [2] the widely quoted "58.3%" is the year-over-year figure as reported. [6] In the same period, staff expenses rose 34.2% and operating expenses 26.5% against that 6.3% revenue growth. [2][5]

Why the margin slid is where the record stops: AI inference costs are the common explanation in the discussion, but the filing attributes nothing, so it stays an analysis, not a finding. [2] The competitive backdrop is genuine pressure, reported rather than filed. Cursor announced direct connections to JetBrains IDEs through the Agent Client Protocol on March 4, 2026. [1] Reported figures — about $4 billion of annual recurring revenue (ARR) for Cursor by June 2026, Claude Code near 54% of enterprise AI-coding usage by mid-2026, and JetBrains' own ecosystem survey reading 29% Copilot, 18% Cursor, 18% Claude Code — all belong in the "as reported" category. [2] JetBrains itself has pushed back on the spending narrative, according to ByteNote's account of the exchange: a company representative stated on Hacker News that the Junie team is 30–40 people, the IntelliJ teams number in the thousands, and Junie does not subsidize usage. [2] The 343,000-users figure sometimes cited for Junie is a vendor self-claim relayed by the same report. [2]

The dev.to piece "The Terminal Won" pairs the $14M loss with Cursor's ARR and declares the IDE era finished [6] — but the filed numbers contradict that headline: operating profit stayed positive all year. [1][2] The Hacker News thread on the results had its title changed, first from one that was too negative — "since the company is not in any sort of financial trouble" — to one that is too positive. [5] An AI-generated digest of that thread captures the same split on the margin question: some readers saw the numbers as large-scale reinvestment and accounting effects rather than distress, others saw early trouble if the margins keep shrinking. [7]

Metered billing is the new seat math

For subscription SaaS vendors, this filing previews a specific profit-and-loss transition, and the mechanics are already first-party. JetBrains sells AI as metered quotas: AI Pro at $10 USD per 30 days with 10 AI credits, AI Ultimate at $30 USD per 30 days with 35 credits; credits are consumed when features rely on cloud models; additional credits can be purchased on both paid tiers; and using ACP agents does not require a JetBrains AI subscription. [8] Where the service itself is restricted — it has been unavailable in China since July 13, 2026 — the vendor points users to AI Assistant with their own API keys or third-party and local models. [8] Those terms come from JetBrains' own help documentation, current as of the AI Assistant 2026.3 version dated October 14, 2026. [8] The reported overage rate is about $1 per credit, according to ic.work. [1] Monthly AI cost therefore has two components — the subscription plus whatever the credits burn. As a hypothetical example: at that reported rate, a developer who exhausts the 10-credit AI Pro quota and needs ten more credits in the same 30-day window would pay $10 on top of the $10 subscription, doubling the month's AI bill.

Base prices moved first: on October 1, 2025, the All Products Pack individual price went to $299 in year one, $239 in year two, and $179 from year three, according to ic.work. [1] The market's reference point is moving the same way — GitHub Copilot shifted to quota-limited usage billing on June 1, 2026 per ic.work's account, [1] and GitHub's current documentation now files the older request-based model under "Request-based billing (legacy)". [9] Seat-priced software is converging on the meter, and the JetBrains accounts show what a gross-margin line looks like while that happens.

The meter has a user-side cost too: "I've been a JetBrains subscriber for over a decade. Nobody is going to use their tools in five years. I certainly don't need my subscription anymore," one subscriber wrote in the discussion. [5] That is an anecdote, not churn data — but it is the sentiment any metered scheme has to overcome.

So — end of an era?

On this filing's evidence, no. Record revenue, positive operating profit, and CZK 7.58 billion of net cash do not describe a company at the end of anything. [1][2] What the era question is really pointing at shows in three lines, and all three moved the same way in FY2025: the gross-margin trajectory, 53.7% to 10.7% in four years; the cost-growth gap, expenses up 26–34% against 6.3% revenue; and the industry's turn from flat seats to metered AI billing. [1][2][8][9] Those are the numbers that will say whether an era is ending — and in the latest year on record, they moved, while the "AI did it" framing did not survive contact with the filed record.