Open Source Robotics Foundation revenue was $2,596,521 in 2024, and most of the growth was not revenue

The Open Source Robotics Foundation is the non-profit behind ROS, Gazebo and Open-RMF. Its Form 990 for the year ended 31 December 2024 reports total revenue of $2,596,521, against $1,943,778 in 2023 and $6,015,244 in 2022. Expenses were $2,029,767, leaving a surplus of $566,754. Net assets closed the year at $5,797,644.

Every figure below comes from the foundation's own filed Form 990 returns for FY2023 and FY2024, read from the IRS filed XML rather than from a summary. Where a number is derived rather than printed, it says so.

The headline revenue figure is accurate and it is the least interesting number in the filing. Net assets grew by $1,563,676 in 2024. The surplus was $566,754. The gap is not revenue at all, and tracing it leads to a wholly owned taxable company that has almost no public existence.

Four answers, none of which agree

Search for this organisation's revenue and page one returns four different numbers:

None of them is lying. They are answering at different vintages, and one of them is answering about a different organisation than it thinks.

The Growjo entry is the outlier worth naming. It estimates revenue at roughly $5M from a headcount of 54, at about $91,800 of revenue per employee. The foundation's own return reports one employee, in both 2023 and 2024. The 54 figure is a plausible memory of Open Robotics when it had an operating company attached to it. It is not a description of the filer today.

Wikipedia carries no revenue figure for this organisation at all. That is silence, not staleness, and silence is the better failure.

Why 2022 is not a baseline

The obvious story is a collapse: $6,015,244 down to $1,943,778 is a fall of 67.7%, followed by a 33.6% recovery to $2,596,521. On that framing 2024 is still 56.8% below 2022.

That framing is wrong, because 2022 was not a normal year. Reading the prior-year column of the FY2023 return, 2022's revenue splits as $561,911 of contributions, $2,146,775 of program service revenue, $2,119,077 on the investment income line and $1,187,481 of other revenue. So 55.0% of that year sat on two lines that are neither donations nor programme work.

December 2022 is also when, in the foundation's own words on its foundation page, "the business of OSRC and OSRC-SG was acquired by Intrinsic, an Alphabet company". A transaction year is not a run rate. The honest comparison is 2023 against 2024, and on that basis programme revenue rose from $1,609,950 to $1,863,290 and the organisation went from a $26,773 deficit to a $566,754 surplus.

One caveat stated plainly: the FY2022 return itself was not retrieved for this piece, so the split between ordinary investment income and any gain on that transaction is not established here. What the FY2023 return supports is the total and the line it sat on, and that is all that is claimed.

The $994,639 that appears three times

Here is the part no aggregator reports. Net assets rose $1,563,676 while the surplus was $566,754. The FY2024 return reconciles the difference itself, and it foots exactly:

So 63.6% of the year's net asset growth arrived on the unrealised gains line rather than through revenue.

Now follow that $994,639. It appears three times in the same return, and the coincidence is not a coincidence:

• Part XI, as the net unrealised gain on investments: $994,639
• Schedule D, as an other-asset line item labelled INVESTMENT IN OSRM: $994,639
• Schedule R, as the foundation's share of end-of-year assets in Open Source Robotics Media Inc: $994,639

The foundation revalued its stake in a wholly owned subsidiary, and the entire revaluation ran through net assets without ever touching revenue. That is an ordinary accounting treatment. It is also the reason a reader who takes $2,596,521 as the measure of the year's performance will be wrong about the direction and size of what happened.

Two subsidiaries changing places

Schedule R is where this organisation is legible, and the useful move is to diff its columns across the two returns rather than read either alone.

Both are 100% owned C corporations controlled by the foundation. In FY2024 the income and asset elements for the Corporation are not zero, they are absent from the filing entirely, while the Media entity's appear for the first time. The two swapped places in a single year.

This also corrects a widely repeated claim. The Wikipedia article states that "the OSRC was sold to Intrinsic". The foundation's own page says the business of OSRC was acquired. Those are different statements, and the filings side with the foundation: OSRC is still listed as a 100% owned, controlled corporation in the FY2024 Schedule R, two years after the transaction. The operating business left; the legal entity did not.

As for Open Source Robotics Media Inc, the return describes its activity as ROBOTICS MEDIA SERVICES and gives a San Jose address. Beyond the filings it is close to invisible: a web search surfaces two pages from a single third party, both being the rules of Intrinsic's AI for Industry Challenge, which name "Employees, interns, contractors, and official office-holders of Intrinsic, Open Source Robotics Media, Inc., and their parent companies, subsidiaries" as challenge entities. A company carrying $994,639 of a public charity's balance sheet and $273,706 of income, with that little public footprint, is worth knowing about. What it does beyond the return's own three-word description is not established here and is not guessed at.

Where the donations came back from

Contributions went $561,911 in 2022, then $28,300 in 2023, then $482,037 in 2024. The 2023 figure is a 95.0% fall and the 2024 figure is a 17.0x recovery.

The recovery has a single identifiable cause, and it is visible in the Part VIII detail. Of the $482,037 of FY2024 contributions, $438,369 is membership dues, which is 90.94% of the total. The FY2023 return has no membership dues line at all; its contributions are $6,780 of government grants and $21,520 of everything else.

A membership dues line appearing for the first time in FY2024 lines up with the launch that year of the Open Source Robotics Alliance, the foundation's membership programme. The return itself does not name the alliance, so that connection is stated as what it is: the line appears in the year the programme launched, and the return does not confirm the attribution.

The alliance's current published fee schedule, labelled 2026, runs from $175,000 a year for a Platinum member with 5,000 or more staff down to $3,500 for a Silver member at a non-profit or a company of ten or fewer, with individuals at $50. Its Platinum members today are Canonical, Intrinsic, NVIDIA and Qualcomm.

That list contains a small circularity worth noticing. Intrinsic, the Alphabet company that acquired the business of the foundation's operating subsidiary in 2022, is now one of four top-tier dues payers to the foundation. The fee schedule is the 2026 one, so it does not price the FY2024 dues line; it describes what the programme charges today.

The payroll moved without anyone being hired

The foundation's salary and benefits line rose from $630,926 to $912,060, up 44.6%, while total expenses rose 3.0%. Employee count was one in both years.

Schedule J explains it. The chief executive's compensation did not grow; it changed employer.

Her combined compensation actually fell 3.4%. What changed is that the share paid by the foundation itself went from 24.3% to 86.2%. With the operating company's business gone, the payroll it used to carry landed on the foundation's own return. The chief technical officer, Geoffrey Biggs, was paid $288,000 by the foundation in both years, so he is not part of the movement.

Four directors received $5,000 each in both years. Brian Gerkey, chairman of the board and the foundation's co-founder, received nothing in either year. Part VII's totals foot exactly: $89,345 plus $288,000 plus four times $5,000 is the declared $397,345 for FY2023, and $306,325 plus $288,000 plus $20,000 is the declared $614,325 for FY2024.

The year the books got audited

FY2024 is the first of the two years in which FSAuditedInd is set. In FY2023 it is 0; in FY2024 it is 1, and an audit committee indicator appears alongside it. On a first audited year the line to check is prior period adjustments, and there is one: $2,283. It is trivially small, which is the reassuring outcome rather than the interesting one.

Diffing Schedule O across the two returns shows the governance language moving in the same year. The description of who reviews the Form 990 goes from "FORM 990 IS REVIEWED BY THE CEO AND CFO" to "FORM 990 IS REVIEWED BY THE CEO AND CFO AND BOARD CHAIR". The compensation process drops a reference to an online tool and gains "AND ASSESSED BY A BOARD SUBCOMMITTEE". The FY2023 return also carried a Part IX line 11g breakdown of technical and payroll services that the FY2024 return does not repeat, so the Schedule O entry count falls from five to four.

The reserve, split properly

Net assets of $5,797,644 against $2,029,767 of annual expenses is about 1,043 days of cover. That number needs two qualifications before it means anything.

First, restriction. $5,796,638 of it is unrestricted and $1,006 is donor restricted, a figure unchanged across both year ends. So 99.98% carries no donor strings. That is the healthy reading and it deserves saying as loudly as a worrying one would be.

Second, liquidity. Of those net assets, $994,639 is the carrying value of a subsidiary, not cash. Excluding it gives about 864 days. Cash alone, at $4,543,668, gives about 817 days. Cash itself rose 36.4% during the year.

Liabilities look alarming in isolation, rising from $136,553 to $692,348, a 5.07x increase. The composition says otherwise: $596,186 of the FY2024 total is deferred revenue, 86.1% of all liabilities, up from $52,112. Deferred revenue is money received and not yet earned, which on an organisation whose largest programme line is CONFERENCES at $1,025,217 reads as a forward book rather than distress.

What foots and what does not

Sixty-eight arithmetic checks were run across the two returns, recomputing each total from the components the filings themselves print. Sixty-seven reconcile exactly, with no tolerance.

That includes all four columns of Part IX in both years, Part VIII revenue, the Part XI roll-forward in both years, the Part X balance sheet in both directions, Schedule D's other-asset and other-liability detail against Part X, Schedule F's regions and individual recipients against its own total, Schedule A's five-year table, and both public support percentages recomputed from their own inputs. Schedule A's five-year gift table is also clean across the restatement audit: every overlapping cell between the two filings matches exactly, so nothing was quietly restated.

Two checks initially failed and both were the measuring instrument, not the filing. The public support percentages are printed to two decimal places and were being compared at five, so 67.34% and 69.89% both recompute exactly once tested at the precision the return actually states.

One genuine discrepancy survives, and it is two dollars. Schedule F reports total foreign expenditure of $27,283, footing exactly against both its own region table and its own recipient table. Part IX line 3 reports foreign grants of $27,285. The FY2023 return has no such gap, its Schedule F and Part IX agreeing at $17,155, so this is not a filing convention. It is a two dollar inconsistency in a $2.6M return, reported because the check was run, not because it matters.

Also worth a line: the Part IX compensation block foots exactly to Part I line 15 in both years, which is why the single-employee figure alongside $912,060 of salary needs no exotic explanation.

How to check this yourself

None of this requires paid data.

1. Find the EIN. For this organisation it is 45-5230545.
2. Fetch the IRS index for the year the return was received, at apps.irs.gov/pub/epostcard/990/xml/2025/index_2025.csv, and grep the EIN. The row gives an object ID and the archive that holds it. Note the index year is the year the IRS received the filing, not the tax year, so grep more than one.
3. Pull the filing's XML out of that archive. The archives are hundreds of megabytes, so use HTTP range requests to read the ZIP central directory and then just the one member you want. That turns a 333MB download into about 30KB.
4. Read the schedules, not only Part I. Schedule R is where subsidiaries live, Schedule J is where compensation is split by employer, Schedule D reconciles the balance sheet detail, and Schedule O carries the governance language.

The same approach produced the Software in the Public Interest accounts, where the headline deficit turned out to belong to someone else entirely, and the Open Technology Fund's FY2025 figures, where the biggest revenue year on record sat next to the worst balance sheet.

Grading the sources

ProPublica's Nonprofit Explorer page is correct and current. It publishes the FY2024 revenue, expenses, net income and net assets exactly as filed, along with the revenue composition. Its weakness is scope rather than accuracy: it reports no Schedule R, so the subsidiaries, the revaluation and the compensation split are all invisible there. Accurate and radically incomplete is the fair verdict. Worth noting for anyone building on it, ProPublica's JSON API is a year behind its own web page, with FY2024 absent from the API's extracted filings while the page renders it.

Cause IQ is one year behind but carries a genuinely useful year-over-year comparison. The FOSS Foundations Directory states openly that its figures are approximated from ProPublica's JSON, which is honest sourcing, and it inherits that API's lag. Growjo is the only one that is wrong rather than stale.

The foundation's own about page publishes no financial statements, but it is the most precise source on the corporate structure, and it is the reason the Wikipedia phrasing can be corrected rather than merely doubted. The alliance's membership page publishes the current fee schedule and the full member roster.