⚠️ Editorial note: The open source ecosystem in China operates under a distinct institutional framework — characterized by state-led initiatives, intranet-like boundaries, and top-down governance. Readers should be aware that this context differs from the community-driven open source model common in other regions. The term “open source” as used in Chinese media may refer to practices that diverge from the conventional definition.

China Open Source Daily — 2026-11-10

🏛️ Policy & Regulation — The ¥450 Million Question and the Filing That Does Not Exist: Reading OpenAtom’s Statutory Accounts, 2020–2023

1. What was published, and by whom. On 2026-11-08 the Europe–China Digital Commons Observatory (eu-china-commons.eu, a project edited by Fabio Pietrosanti, source and data public on GitHub under CC BY 4.0) updated a research note first issued 2026-11-03: a pass over the statutory annual work reports (年度工作报告) that the OpenAtom Open Source Foundation (开放原子开源基金会) files with the Ministry of Civil Affairs — the filings behind its non-profit status, not its marketing material. The observatory’s method is stated in its own methodology document: every claim is labelled fact (directly supported by a cited text), commitment (a published target, not a result), inference or proposal, and graded A (primary official text, archived, read in the original language) through D (LLM-generated lead, never publishable). The financial figures below are graded A (OCR): the reports are image-only scans, the observatory ran pdftoppm and tesseract chi_sim+eng over them, cross-checked that income minus expenditure equals the change in net assets, and flagged that the numbers must be re-read on the page image before publication.

2. The disclosure page, verified for this briefing. OpenAtom’s own information-disclosure page at openatom.cn/information/work lists four documents: 2020, 2021, 2022 and 2023 annual work reports, all as scanned PDFs. Fetched directly on 2026-11-10 for this briefing, the page still lists exactly those four years. There is no 2024 report and no 2025 report. The foundation does publish other things: a “2025 Annual Report” (2025年度报告) news item dated 2026-02-06, and monthly notices of new donors — but a 年度报告 announcing a year’s achievements is not a 年度工作报告 filed for annual inspection, and the one does not substitute for the other. The distinction between the two documents is the whole of this entry.

3. The filed numbers, as recovered.

YearDonation income (CNY)of which foreignGovernment subsidy (政府补助收入)Total incomeTotal expenditureAdmin (管理费用)Net assets, year end
202016,000,00001,456,76317,531,4953,386,0341,177,02844,145,461
202168,450,00001,456,76370,465,26531,296,0433,374,43783,314,683
202263,690,0000072,364,88538,793,7743,758,795 (9.69%)116,885,794
202375,502,000020,700,000110,098,42697,508,1683,848,502 (3.95%)130,293,368
2020–23223,642,000023,613,526270,460,071170,983,018

Three details in that table carry more institutional weight than the totals. First, the subsidy is almost entirely one item. Of the ¥23.6 million in government money across four years, ¥20.7 million arrives in 2023, of which ¥20.2 million is restricted to a single programme: the 开源项目生态培育平台 (open-source project ecosystem cultivation platform) — spending the foundation then booked as compatibility testing, a licence system, training and hosting. The filings do not name the granting body; the observatory records MIIT as likely and labels that an inference. Second, foreign donations are zero in every filed year. Not small — zero. Third, the administration ratio is low and falling, from 9.69 percent in 2022 to 3.95 percent in 2023, against total expenditure that nearly tripled.

4. Who the donors are, and what the largest one got back. In 2023, the two largest named donors are Huawei Device (¥15.772M) and Huawei Technologies (¥12.08M), followed by ChinaSoft International (¥8.1M) — together ¥35.95 million, or 48 percent of that year’s donations, with Huawei’s two entities alone at ¥27.85 million and 37 percent. Earlier years show the same pattern at smaller scale: Huawei Technologies ¥6.0M in 2020; Shenzhen Kaihong Digital (深开鸿) ¥3.5M and JD (Chongqing) ¥1.0M earmarked for OpenHarmony in 2021; openEuler-earmarked donations from Hunan Kylinsec (麒麟信安, ¥3.0M) and Jiangsu HopeRun (润和, ¥1.0M) in 2022.

And then the outflows. In 2023, ¥6,485,400 went to Huawei Cloud for OpenHarmony services, plus ¥1.65 million for OpenHarmony code hosting. In 2022, from ¥9.4 million of OpenHarmony spending, ¥1.63 million was code-hosting fees and a further ¥3.64 million went to a payee the OCR could not recover. The observatory’s own finding, graded as fact, states it plainly: part of the donated money flows back to the donor as cloud-service revenue, disclosed on the related-party pages, and not arm’s-length.

5. Where 2023’s money went, by project. Eight projects totalling ¥93.65 million: OpenHarmony took ¥25.83M of the outlay against ¥29.02M of earmarked income; openEuler ¥8.65M against ¥12.18M; the cultivation platform ¥9.17M against the ¥20.2M subsidy; the Global Open Source Summit ¥8.84M against ¥5.38M of income (a deficit); the open-source competition ¥5.62M; the campus programme (校源行) ¥3.45M; a developer conference ¥4.59M; and “other open-source public-interest” activity — culture, legal work, research, 1024 day, vulnerability sharing — ¥27.51M. The observatory’s rough classification: about 47 percent to directly open-source-specific engineering and infrastructure, about 24 percent to events, competitions and campus programmes, about 29 percent to the mixed “other” line.

6. The reconciliation problem, which is the item to hold on to. The foundation’s public figure, reported by Xinhua on 2026-06-25, is “more than ¥600 million raised cumulatively since 2020” and 198 donor organisations. The filed income for 2020–2023 is ¥270.5 million. Reaching ¥600 million by mid-2026 requires roughly ¥330 million more across 2024 to mid-2026 — about ¥130 million a year, which is possible if 2023’s growth continued, and is supported by no filing, because no filing for those years is published. The observatory’s classification is that ¥600 million should be cited only as reported by Xinhua, and that the figure may also count pledges or project donations booked through a separate Shenzhen service centre.

Institutional Analysis & Commentary:

A foundation’s accounts are the document that shows what kind of institution it is, and this one describes a project-financing vehicle rather than a membership body — which is a coherent design that produces a specific failure mode.

Start with the composition. Western open-source foundations are, in the main, membership organisations: companies pay dues to hold a seat, and the foundation spends on staff, events, legal defence and sometimes on projects. OpenAtom’s income is donations earmarked to projects — 2023 breaks down as ¥32.10 million unrestricted and ¥43.40 million restricted, i.e. 57 percent restricted, plus a government subsidy that is itself restricted to a named platform. Restricted money is not a scandal; it is how a donation-based foundation works. But it determines what the institution can decide. An unrestricted yuan can pay a reviewer of a stranger’s patch. A restricted yuan pays for the thing named in the restriction. The growth of the foundation’s balance sheet is, structurally, a growth in the number of decisions already made by somebody else.

Now the concentration. Huawei at 37 percent of the 2023 donation base, and the two state telecoms and a state fund entering a different institution’s cap table this month, are the same structural fact at two scales. This series argued on 2026-11-09 that a donations-and-grants foundation has an incentive to grow the number of registered projects and the completeness of the register rather than to keep projects attractive to dues-payers. The filings make the corollary concrete: a foundation whose largest donor is also the project’s original owner, whose cloud contract is with that donor, and whose alumni organisation staffs the project cannot use its budget to discipline the donor. It can only use it to serve the donor’s project better. The ¥6.49 million that went to Huawei Cloud in 2023 is not misconduct — it is disclosed on the related-party page, which is exactly what the disclosure regime is for — but it is the clearest available demonstration that in this institution the money does not create independence from the funder. It circulates back.

The government money is the part that deserves the most careful reading, because it is misread in both directions. Look at the shape: ¥1.46 million in 2020, ¥1.46 million in 2021, nothing in 2022, ¥20.7 million in 2023. That is not a subsidy programme; it is a single 2023 grant, restricted to an “ecosystem cultivation platform”, whose components the filings list as compatibility testing, a licence system, training and hosting. Read that list against the 2026-08-28 keynote commitment this series examined yesterday — ¥150 million a year for three years for “root-technology evolution, shared industrial capability, development toolchains, security systems, compatibility capability and key technology” — and the continuity is obvious. Certification, compatibility testing, licence infrastructure, training. The instrument has been the same since 2023; the announcement in 2026 is the same instrument at thirty times the scale. Whether the state’s contribution to Chinese open source is large or small is not the interesting question. The interesting question is that it is booked as platform-building, which is the line item a registry operator can actually execute.

The comparison that makes the question answerable, and it is not the one usually made. The observatory’s companion pass over 34 projects (18 Chinese or Chinese-origin, 16 European) sorts every funding channel and arrives at a finding worth quoting because it is the cleanest formulation of the pattern this series has been circling since July: European public money for named projects is mostly creation — large envelopes (OpenEuroLLM €20.6M; DARE €120M) — or maintenance through a per-project grant (Sovereign Tech Fund: €405,888 to OpenSSL, €195,000 to curl, €1,285,200 to KDE). Chinese public money in the same sample is mostly adoption: subsidies that pay users to deploy OpenHarmony and openEuler, plus state labs. Three funding modes, three different institutions. A creation grant buys a deliverable. A maintenance grant buys continuity of a person who answers. An adoption subsidy buys installed base. Only one of the three pays the reviewer of a stranger’s patch, and it is not the mode China has chosen — which is precisely what yesterday’s field note measured, twelve unanswered October issues under a foundation that had just committed ¥450 million.

And the honest complications, because the note is careful and deserves to be reported as careful. The observatory grades all 35 OpenAtom repositories it checked as OSI-approved licences — Apache-2.0, MulanPSL-2.0, MIT, BSD-3, GPL-2.0/3.0 — and explicitly records that MulanPSL-2.0 is OSI-approved while MulanPSL-1.0 and MulanPubL are not, and that it found no OpenAtom project using the unapproved ones. That matters: this series has been sceptical of licence-washing in Chinese open source, and on the evidence the licence layer here is clean. It also observes that individual donations are nearly absent across the Chinese sample — Ant Design about US$18,000 all-time on Open Collective, KubeSphere US$0, against curl’s US$499,000 — which is a fact about a payment rail and a tax regime as much as about a culture. And it refuses to count OpenAtom’s ¥600 million as government money: donations are classified instrument_type: donation, summable: false, excluded from public-funding totals, with only the ¥20.7 million 2023 subsidy treated as a public grant, and even that flagged unsummable until the granting body is verified. That discipline is the reason the note is useful. Its own conclusion is modest and correct: disclosure is weak, the work reports are scans, they stop at 2023, and the next step is the Ministry of Civil Affairs registry rather than more journalism.

Checkable questions.

(i) Will a 2024 and a 2025 年度工作报告 be filed on openatom.cn/information/work — and in text, not as a scan? This is the single document that would let anyone reconcile the ¥600 million claim with the ¥270.5 million of filed income. A non-profit foundation that files annually with the Ministry of Civil Affairs either has those filings or is late with them; the page is the evidence either way, and it is checked here at zero cost.

(ii) Will the ¥150 million annual commitment for 2026–2028 appear in a filed annual report at all, and in which project line? On the 2023 precedent it would appear as restricted income with a named purpose. If it appears instead only in keynote text, the commitment was a rhetorical instrument; if it appears as cash to Huawei Cloud and compatibility testing, it was the 2023 instrument scaled up.

(iii) Does the granting body for the ¥20.7 million 2023 subsidy ever get named in a public document? The filings omit it. MIIT is the inference. A grant whose source cannot be named in the recipient’s own statutory filing is a governance fact independent of the sum, and it will recur if the ¥450 million is disbursed the same way.

Sources: Europe–China Digital Commons Observatory — Chinese open-source foundations and OpenAtom finances, research note 2026-11-03, updated 2026-11-08 (OCR of the 2020–2023 Ministry of Civil Affairs annual work reports; income, donation, subsidy and expenditure tables; named donors; project spending; reconciliation against the Xinhua ¥600M figure) · Source file on GitHub — docs/CN-FOUNDATIONS.md · Observatory — Methodology: sources, evidence levels, verification (labels A–D; announced budgets are commitments unless disbursement is documented) · OpenAtom Foundation — 信息公开 / 工作报告 (disclosure page; lists 2020, 2021, 2022, 2023 only; fetched 2026-11-10) · OpenAtom Foundation — 开放原子开源基金会2025年度报告 (news item, 2026-02-06 — a achievements report, not an annual-inspection filing) · Xinhua — 开放原子开源生态大会 (2026-06-25): more than ¥600 million raised cumulatively since 2020; 198 donor organisations · This series’ 2026-11-09 briefing — the ¥150 million annual commitment, the 53 project-group donors, and the PikiwiDB field note


1. The artefact. The same observatory published, also updated 2026-11-08, a pass over 34 major open-source projects — 18 Chinese or Chinese-origin, 16 European — recording for each the dominant funding channel, the top funder, the funder’s nationality, whether public money is involved, and an evidence grade. Its method note is unusually honest about its own limits: contributor affiliation is mostly inference from project ownership, and a systematic OpenDigger/Gitee affiliation pass “is still to be done”.

2. The hard numbers it recovered, which are the part that can be checked and compared. Not all 34 rows have a figure; the ones that do are these:

  • curl — Open Collective **US$498,859 all-time** (about US$91k a year) plus a Sovereign Tech Fund grant of €195,000.
  • OpenSSL — STF €405,888.
  • KDE — STF €1,285,200; patrons €67,378 (2024).
  • The Document Foundation / LibreOffice — income €2.18 million (2025).
  • OpenEuroLLM — €20.6 million EU envelope. DARE (RISC-V) — €120 million EuroHPC envelope.
  • Ant Design — Open Collective US$18,412 all-time**. **KubeSphere** — Open Collective **US$0.
  • OpenHarmony — OpenAtom spending ¥25.83 million (2023), of which ¥6.49 million to Huawei Cloud; Huawei at 37 percent of OpenAtom’s donation base. openEuler — OpenAtom spending ¥8.65 million (2023).

3. The three modes, stated in the observatory’s own formulation, because the formulation is the finding.

  • Creation. A large public envelope buys a deliverable: OpenEuroLLM’s €20.6 million, DARE’s €120 million.
  • Maintenance. A per-project public grant buys the continuation of work on something that already exists and is already depended on: the Sovereign Tech Fund’s €0.2–1.3 million per project to curl, OpenSSL and KDE.
  • Adoption. Public money pays users to deploy a stack: local subsidies for OpenHarmony and openEuler adoption, plus state labs (XiangShan at CAS, Peng Cheng for MindSpore).

4. The second pattern, which is about who else is in the room. Chinese projects in the sample show almost no individual-donation channel — the strongest single figure being Ant Design’s US$18,412 all-time, against curl's US$498,859. And Chinese-origin projects that went to the Apache Software Foundation for neutral governance — Doris, ECharts, Dubbo, RocketMQ, APISIX — retain the founding company or its VC-funded spin-off as the actual payer: the ASF pays no developers. The observatory grades this an inference (B) and it is the same inference this series reached on 2026-10-26 from CODEOWNERS and maintainer files alone.

5. The licence check, recorded because it cuts against the series’ usual suspicion. The observatory checked 35 OpenAtom repositories against GitHub/Gitee licence headers and the SPDX list: every one uses an OSI-approved licence. It separately notes the edge case that matters for EU work — MulanPSL-2.0 is OSI-approved and confirmed on opensource.org; MulanPSL-1.0 and the Mulan Public License are not, and no OpenAtom project was found using the unapproved variants. It also records that Mulan licences were donated into the foundation in 2026, which makes OpenAtom the counterparty for any EU licence-compatibility discussion.

Institutional Analysis & Commentary:

If you sort funding by where in the stack it enters, the three modes stop being a taxonomy and become a theory of what each one can and cannot produce.

A creation grant is a procurement of novelty. Its unit of account is a deliverable by a date, its risk is that the deliverable is a demonstration rather than a dependency, and its characteristic failure is the pilot that does not outlive its envelope. A maintenance grant is a procurement of continuity. Its unit of account is a person-months commitment to an existing codebase, its risk is picking winners, and its characteristic virtue is that it can be aimed at exactly the thing no commercial actor will fund — review capacity, release engineering, the security response that has no customer. An adoption subsidy procures installed base. Its unit of account is units shipped and certified, and its characteristic strength is that it solves the cold-start problem no volunteer community has ever solved by itself.

Each of the three is legitimate, and China has chosen the third — at scale, consistently, for six years — while the observable deficit in Chinese projects, documented across this series from 2026-10-26 through this week, is a maintenance deficit: review queues, triagers, people paid to answer strangers. That is not a coincidence and it is not a funding-level failure. It is a category error in matching instrument to constraint. ¥450 million booked as compatibility testing and platform building will buy a great deal of installed base and no triage, and the reason is that adoption subsidies and maintenance grants are administered by different institutions with different paperwork. A district government can pay a company to deploy a certified stack. It structurally cannot pay an individual to review a pull request, because there is no procurement category in which that is a deliverable.

The European comparison is therefore not a scoreboard, and reading it as one would be the mistake. Look at the actual magnitudes: the Sovereign Tech Fund’s €195,000 to curl is about one-fifth of a single engineer-year at European rates. It is not a solution to maintenance; it is a demonstration that the category exists. What Europe has that the sample shows China lacking is not more money — OpenAtom’s ¥270 million of filed income over four years dwarfs anything the STF has disbursed to individual projects — it is a line item whose name is the problem. Once “maintenance” is a fundable category, it can be criticised, sized and grown. Where it is not a category, the deficit cannot even be entered in a ledger, and the resulting silence is misread as absence of need.

Then there is the donation channel, and here the numbers say something the funding debate usually ignores. curl’s US$499,000 all-time from individuals is not large money — it is roughly one senior engineer for two years across a decade. **But it is money with no procurement officer in the loop, and that is its institutional value.** Ant Design at US$18,412 and KubeSphere at US$0 are not evidence of stinginess; they are evidence that in the Chinese stack the small-donor rail has not been built — no fiscal-host equivalents with the reach of Open Collective, no tax treatment that makes an individual donation rational, no norm that a company’s engineers chip in to a dependency. Where that rail does not exist, a project’s only funders are its owner and, sometimes, a ministry. Every option then runs through an institution that must justify itself in a procurement category, which is exactly the constraint that cannot express “pay the reviewer”.

Finally, the licence finding deserves its correction of this series’ own instinct. The recurring suspicion about Chinese open source — that the licence says open and the production model does not — does not survive contact with the repository data: 35 of 35 checked OpenAtom projects carry OSI-approved licences, with the non-approved Mulan variants unused. The gap this series keeps finding is not in the licence text. It is in the funding line. Which is a much harder problem, because a licence can be changed by a legal department in an afternoon, and a procurement category takes a budget cycle and an act of imagination to create.

Checkable questions.

(i) Does any Chinese public programme pay for maintenance of a named project — not adoption, not platform-building, not a competition prize? The observation to test is a null result: the observatory’s own research register treats “no central Chinese open-source budget line found” as a legitimate published finding. A positive instance would falsify the pattern in this entry and would be worth reporting the day it appears.

(ii) What is the total of local adoption subsidies for OpenHarmony and openEuler? They are recorded in the observatory’s data/funding/cn.json and deliberately kept out of the public-funding comparison until they can be de-duplicated against the foundation’s own subsidy line. That total, once assembled, is the size of China’s chosen instrument — and it is the number to set against the ¥450 million.

(iii) Do the Chinese-origin Apache projects’ PMCs ever acquire a voting member who is not paid by the founding company? The observatory classifies current funding as company-retained (B). The PMC roster is the institutional indicator this series has used since 2026-10-26, and it is public.

Sources: Europe–China Digital Commons Observatory — How 34 major open-source projects are funded (EU and China), updated 2026-11-08 (34 projects; hard-number table; EU vs CN findings; licence check across 35 OpenAtom repositories) · Source file — docs/PROJECT-FUNDING.md · Sovereign Tech Fund — project grants (OpenSSL, curl, KDE) · Open Collective — curl · opensource.org — Mulan Permissive Software License v2 (OSI-approved) · Observatory — Chinese open-source foundations and OpenAtom finances (the 2021–23 accounts reused for this pass) · Observatory — The Chinese open-source landscape: projects and code platforms · This series’ 2026-10-26 briefing — the survival rule for company-donated projects and the CODEOWNERS indicator


🔍 Commentary — The Line Item Is the Institution

First: put this week’s three documents side by side and they describe one system from three angles. On 2026-11-09: a keynote committing ¥150 million a year to toolchains, security, compatibility and root technology, and reporting 140 million lines, 13,808 contributors, 2,040 tested products under a foundation’s neutral custody. Also on 2026-11-09: a project under that foundation with 12,000 stars, 584 merged pull requests and twelve unanswered October issues. Today: the same foundation’s filed accounts, showing ¥270.5 million of income over four years, 37 percent of 2023 donations from one company, ¥6.5 million of it paid back to that company’s cloud, and a ¥20.7 million state grant booked against a platform whose named components are compatibility testing, licensing and training. Announced money, missing queue, filed accounts. The filing explains the queue.

Second: the report’s real contribution is not a scandal, and it would be a mistake to read it as one. Every figure is disclosed on a related-party page; the foundation’s admin ratio is under 4 percent; the licences are clean across 35 repositories; the observatory explicitly refuses to count private donations as public money and grades its own OCR as needing re-reading. What it establishes is a classification, not a fault. Chinese open-source funding is project-restricted corporate donation plus a narrowly-purposed state grant, both of which are administered as platform-building, and platform-building is exactly what a registry operator can execute well. This series has spent weeks documenting that registries work — 2,040 compatible products is a real achievement — and that they do not answer strangers. Both statements are true, and the filings are where they meet.

Third, and this is the correction this series owes from two days ago. On 2026-11-08 the argument was that certification builds a supply of contributors faster than projects build a queue to receive them. The filings refine it: the queue is not unfunded because nobody noticed. It is unfunded because no instrument in this ecosystem has a category for it. A district can pay rent subsidy per certified person; a ministry can pay for a cultivation platform; a company can pay for its own project’s engineers. None of those forms can carry “one senior reviewer, two days a week, for a project whose donor has moved on” — and a deficit with no accounting category is not invisible to the institutions, it is unrepresentable in them. That is a stronger claim than “disclosure is weak”, and it is the one the numbers actually support.

Fourth: the comparison with Europe this week does not flatter Europe, and saying so is part of the method. The Sovereign Tech Fund’s grants are small — €195,000 to curl, €405,888 to OpenSSL — and Europe’s large envelopes are creation grants of exactly the kind that produce demonstrations. What Europe has is one institution that named the maintenance category out loud and started paying into it at demonstration scale. That is not a funding gap being closed; it is a language being established. The reason it matters disproportionately is that once the category exists, the size becomes a budget negotiation. Before it exists, the size cannot be argued at all.

Fifth, the sentence worth keeping, and it is a testable prediction about documents rather than about software. A foundation that commits ¥150 million a year in a keynote and files annual work reports to a ministry either publishes the report in which that money appears, or it does not. The page is public. It listed four years in 2020–2023 and it lists four years today. A foundation’s disclosures are its constitution: what it files is what it is accountable for, and what it announces is what it hopes will be believed. The most useful thing any reader can do with this week’s material is check that page again next year — because a fifth and sixth report on it would be a more meaningful development for Chinese open source than any keynote, and their continued absence is already a finding.

Sources: Europe–China Digital Commons Observatory — research index (all chapters, evidence-graded, CC BY 4.0) · Observatory — Chinese open-source foundations and OpenAtom finances · Observatory — How 34 major open-source projects are funded · OpenAtom Foundation — 信息公开 / 工作报告 (four annual-inspection filings, 2020–2023) · This series’ 2026-11-09 briefing — OpenAtom’s ¥450 million commitment and the PikiwiDB field note · This series’ 2026-11-08 briefing — the credential as instrument, and a vendor forge measured from inside