Tor Project revenue was $12,279,993 in 2025, and the US government share fell to 22.6%

Quick answer. The Tor Project reported total revenue of $12,279,993 for the fiscal year ended 30 June 2025, against expenses of $8,967,599, a surplus of $3,312,394, and net assets of $10,042,968. That is a record, and it is 68.5% above the $7,287,566 of FY2024. The more interesting number is underneath it: government grants as reported on the return fell to $1,932,985, or 15.74% of revenue, from $3,684,992 in FY2023. On the Tor Project's own published definition the US government share is 22.62%, down from the 53.5% it published for FY2022. One aggregator has the FY2025 totals right. None of them reports the composition, and the Tor Project has not published this year itself: the return was filed with the IRS on 15 May 2026 and has not yet appeared on its own reports page.

The Tor Project, Inc. is a Massachusetts-domiciled 501(c)(3) formed in 2006, EIN 20-8096820, filing from a post box in Winchester, New Hampshire. Its fiscal year runs 1 July to 30 June. Everything below comes from three consecutive filed returns, pulled as XML from the IRS bulk archive of Form 990 filings: the FY2023 return (object 202421369349300337), FY2024 (202501359349316200) and FY2025 (202611359349300241).

The headline, and the two other headlines

Search for this and you will be handed four different numbers. All of the following are published, and three of them are correct.

The first two are both right, and the gap between them is exactly $700,288. Schedule D Part XI itemises it: donated services and use of facilities $643,794, net unrealised investment gains $45,148, and a foreign currency exchange gain $11,346. Those three sum to $700,288 to the dollar, and $12,980,281 minus $700,288 is $12,279,993. The return excludes in-kind contributions from revenue and books them as an expense instead; the audit includes them on both sides.

This matters because the Tor Project itself publishes both, and said so plainly in its FY2023-2024 transparency post: the difference "is not a mistake", it reflects "differences in rules about how you must report tax revenue to the IRS and how you must report revenue in audited financial statements."

There is a third figure for FY2024, and it is worth pinning down because it trips people up. The blog post gives audited "Revenue and Support" of $8,005,661 against the 990's $7,287,566, and attributes the difference to in-kind contributions of $768,413. Add those and you get $8,055,979, which is $50,318 too much. The residual is exactly FY2024's investment income of $50,318, which the audited statements present below the "Revenue and Support" subtotal rather than inside it. Schedule D then reports the audited total as $7,993,079, and $8,005,661 plus investment income $50,318 plus the FX gain $9,092 minus the unrealised loss $71,992 lands on $7,993,079 exactly. Three figures, one year, all internally consistent once the subtotal boundary is visible.

Where the $12,226,787 of contributions came from

The Tor Project does something almost no filer does: it names its funders, donor by donor, in Schedule O. Here is FY2025 against the two prior years, taken from those narratives.

Four movements carry the year. Mullvad, which had been the single largest corporate funder, fell 78.2% from $1,402,954 to $306,225, taking it from 19.39% of FY2024 contributions to 2.50% of FY2025's. The startsmall Foundation tripled to $3,000,000. Individual donations tripled to $3,462,303, up $2,359,684 in one year. And the Open Technology Fund rose 147.9% to $844,580 while the State Department's main grant fell another 22.6%.

Two lines now carry more than half the organisation: startsmall at 24.54% of contributions and individual donations at 28.32%, or 52.85% between them.

The smaller entries are worth a glance for what they say about who funds privacy infrastructure. Proton AG gave $40,000, DuckDuckGo $25,000 for the third year running, Brave Software $10,000 for the third year running, the Brewster Kahle Charitable Foundation $25,000, the Ethereum Foundation $5,000, the Internet Society Foundation $1,200, and a website called Word Unscrambler $11,400.

"What share is government money" has two right answers, and both are in the filing

This is the question the search results ask and none of them answers. The FY2025 return reports $1,932,985 on Part VIII line 1e, "government grants", which is 15.74% of revenue. The same return's Schedule O narrative puts $2,777,564 under the heading "funding from US government sources", which is 22.62%.

The difference is entirely the Open Technology Fund. OTF is a US-government-funded non-profit rather than a federal agency, so the Tor Project lists it under the government heading in prose and excludes it from the government line item. The arithmetic confirms the rule holds across all three years, to within a dollar:

A dollar in each direction across three years is rounding, not error. What it establishes is that line 1e is the direct federal money only, and that anyone quoting it as "Tor's government funding" is understating by whatever OTF gave that year, which in FY2025 was 6.88% of revenue on its own.

Add Sida, a Swedish government agency, and all public-sector-derived money is $3,370,380, or 27.45% of revenue. So the honest answer is a range, and the filing supports every point in it:

The middle row is the one to use for comparisons, because it is the Tor Project's own method. Its FY2023-2024 post reports 35.08% for FY2024, and dividing that year's Schedule O US-government total of $2,556,473 by revenue of $7,287,566 reproduces 35.08% exactly. Applied consistently, the series is 53.5% for FY2022 (the Tor Project's own published figure), then 49.26%, 35.08%, and 22.62%.

That is the thing the filing settles. The organisation wrote in December 2025 that "our long-term mission to reduce our dependence on U.S. government funding has been coming to fruition." The return filed five months later shows the share roughly halving again.

What the diversification cost

Escaping one concentration bought another, and Schedule A prices it. Public support fell from 98.73% of the five-year base in FY2022 to 96.49%, 89.69% and 83.35%. The mechanism is the substantial-contributor exclusion, which strips the portion of any single donor's gifts above 2% of the base:

The excluded amount grew 6.3x in two years. The Tor Project still clears the 33 1/3% public-support test by 50.02 percentage points, so nothing is at risk, and saying otherwise would be alarmism. But the direction is real: replacing a small number of large government grants with a small number of large private ones does not reduce donor concentration, it relocates it.

The reserve is genuinely healthy, and the restriction split barely matters

Net assets rose 50.5% to $10,042,968, against annual expenses of $8,967,599. That is 408.8 days of cover. The number that usually deflates a figure like that is the donor-restricted share, and here it does almost nothing:

Restricted funds fell in absolute terms while net assets grew by half, so the unrestricted reserve is 381.8 days and the gap between the headline and the usable figure is a factor of 1.07. Liabilities are $466,705, or 4.44% of assets. Within the year the Tor Project also moved cash into interest-bearing accounts, with savings and temporary cash investments up 151.0% to $6,650,205 while pledges and grants receivable fell 27.3% to $798,557.

Where the money goes: 93.95% of it is people

Salaries, benefits and payroll taxes came to $5,043,606 in FY2025. Fees for services, other, came to $3,381,233. Together that is $8,424,839, or 93.95% of all expenses. Almost nothing else on the expense schedule is material: information technology $88,888, conferences and meetings $161,100, occupancy $4,390.

The split between those two lines is the organisational fact. The return reports 22 employees, up from 20 and 18. But five independent contractors were paid more than $100,000 each in FY2025, totalling $694,997, up 73.9% from three contractors and $399,600 in FY2024, and their roles are core engineering rather than overflow: applications team lead, community team lead, project manager, and two applications engineers. Schedule F separately reports 49 employees, agents and independent contractors outside the United States.

Foreign spending reached $2,470,150, or 27.55% of expenses, and its shape shifted hard: Europe up 89.5% to $1,081,696, South America up 65.8% to $1,041,116, Asia up 334.8% to $325,628, and non-US North America down 89.7% to $21,710. A Middle East and North Africa region present in FY2024 at $80,936 does not appear in FY2025.

Donated services, meanwhile, are shrinking while cash grows: $831,179, then $768,413, then $643,794, down 22.54% across two years. The FY2024 post valued that year's contribution at "7,614 hours of software development, 1,894,720 words translated, cloud hosting services for 23 servers, and 26 certificates", and noted that the estimate "does not include relay operator time and cost, but it should."

The crypto line, and a useful contrast

Schedule M reports one non-cash category in all three years, described only as "crypto currency" with no coin named.

The valuation method is fair market value, the organisation reports no property it must hold, and the average gift in FY2024 and FY2025 is essentially identical at $263 while the count fell 15.8%. It is also the opposite composition to the other crypto-heavy filing on this desk. The GNOME Foundation's record revenue for its own 2025 year, which ended 30 September, was 77.7% a single cryptocurrency donation of $2,248,286 from one contributor, sold at a $39,828 loss. Same asset class, two overlapping 2025 fiscal years, and one of them is a crowd of 2,391 while the other is one person.

Grading the answers already on the page

ProPublica's Nonprofit Explorer page is correct and current. It carries $12,279,993, $8,967,599, net income $3,312,394, net assets $10,042,968, contributions $12,226,787 at 99% of revenue, and executive compensation $1,021,489 at 11% of expenses. Every one of those reconciles to the return. Its API is a different matter: the structured filings_with_data array stops at FY2023, so an automated consumer gets figures two years older than the web page for the same organisation. The fair verdict is accurate and radically incomplete, because nothing on it names a funder, separates OTF from line 1e, or mentions the second audited revenue figure.

Wikipedia is stale in a specific and forgivable way. Its infobox gives revenue $5,999,891 and expenses $4,853,334, both labelled "(2021)". Those are the figures for the year ended 30 June 2022, so the label is a year out as well as the data being three fiscal years behind. It understates current revenue by 2.05x. Two smaller things: the infobox gives headquarters as Winchester, Massachusetts, whereas the return's mailing address is Winchester, New Hampshire, with Massachusetts as the state of legal domicile, and both towns exist. And the article's "a working staff of 22 people" is a sentence about a 2020 pandemic layoff, not a headcount; it coincidentally equals FY2025's 22 employees, which is the kind of agreement that should not be treated as corroboration.

LeadIQ is the outlier. Google's displayed snippet for its page reads "As of July 2026, The Tor Project's annual revenue is estimated to be $100M - $250M." The figure is not in the page's static markup, so it is quoted here as the snippet. Against $12,279,993 that is an overstatement of 8.1x to 20.4x.

The 2018 press coverage is the one still doing damage. A widely cited 2018 article reports $4.2 million of income for 2017 with 51% from government funds. Both halves are eight years old, and the government half is the number people still repeat.

The checks that passed

A filing this clean deserves to have that said first. Fifty-two structural identities were recomputed from the XML across the three returns, and 52 of 52 passed exactly: the Part I revenue and expense columns chain correctly between consecutive years, the revenue composition identity holds all three years, the balance sheet identity holds all three years, the Part IX functional-expense schedule foots to the printed total all three years, the Part XI net-asset roll-forward foots all three years, the Part X restriction split sums to net assets all three years, Schedule D's revenue and expense reconciliations foot all three years, Schedule A's five-year table and public-support subtraction foot all three years, and Schedule F's per-region employee and expenditure columns sum to their printed subtotals.

Four artefacts, none of them errors, are worth recording so nobody reports them as findings:

• The Part IX "other expenses" line appears twice in every one of the three returns. Summing only the first occurrence produces a phantom shortfall.
• Schedule A's printed public-support percentage differs from the recomputed value in the fifth decimal place in all three years (0.96489 against 0.96490, 0.89694 against 0.89690, 0.83351 against 0.83350). Rounding.
• The Schedule O donor narrative does not tie exactly to total contributions: it is $1,914 over in FY2023, $56,974 under in FY2024 and $10,002 under in FY2025. For FY2024 the Tor Project's own post resolves it, publishing an "Other" category of $107,293 which is investment income $50,318 plus $56,975 of contributions it did not itemise. FY2023's residual has the opposite sign and so cannot be an omission; on a $8.46m base it is 0.02% and the filing does not settle what it is.
• The percentages in that published FY2024 category table sum to exactly 100.00, and the amounts sum to $7,287,565 against line 12's $7,287,566. One dollar.

Two typographical items in the FY2025 return: Part VII lists a "staff acountant", and Schedule L names the same related-party employee as CARA ABT in FY2024 and CALA ABT in FY2025, the latter matching the Part VII spelling. The transaction itself is properly disclosed both years, an employee with a familial relationship to the CFO, compensated $106,205 then $130,312.

What this filing does not settle

The audited statements are checked as audited in all three years, but there was no audit committee in any of them, which is disclosed rather than hidden. A federal single audit was required and performed in all three years. The board was eight voting members in FY2023, seven in FY2024 and eight in FY2025, every one of them independent in all three years.

The Tails merger appears nowhere in the return. The Tor Project announced on 26 September 2024 that the two had "joined forces and merged operations", incorporating Tails into its own structure, which places the event inside FY2025. Yet "Tails" occurs zero times across the FY2024 and FY2025 filings, and neither carries a Schedule R or a Schedule N. That absence is not by itself a defect, because the schedules that would record a transfer are generally filed by the terminating organisation rather than the receiving one. It does mean the FY2025 revenue jump cannot be attributed to the merger from this document, and it should not be.

Nor does the return explain why individual donations tripled. A $2,359,684 increase in one year on a base of $1,102,619 is the single largest movement in the filing, and Schedule O aggregates it into one line.

The number to actually use

For "Tor Project revenue", the answer is $12,279,993 for the year ended 30 June 2025, on the Form 990 basis, or $12,980,281 on the audited basis, and you should say which. For "how much of Tor is government funded", the answer is 22.62% on the organisation's own method, down from 53.5% three years earlier, and 15.74% if you only count the government grants line. Anything between 15.74% and 27.45% is defensible if you name the definition. Anything above 50% is describing FY2022 or earlier.