⚠️ 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 — 2026-11-14
🏛️ Institutional Change — The First Export That Is a Distribution: openEuler at 61%, and HECOS 1.0 in Pakistan
1. The event, and the three numbers in it. On 2026-11-13 the Agentic Infra OS Conference & openEuler Summit 2026 opened at the Zhongguancun International Innovation Center in Beijing, organised by openEuler, the community incubated by the OpenAtom Open Source Foundation (开放原子开源基金会). The community released an operating system positioned for Agentic Infra (智能体基础设施) — the substrate for agent runtimes, tool invocation, task-state management, memory systems and identity control — shipping openEuler 24.03 LTS SP5, with sandbox, KV Cache, memory-system and Agent security capabilities, and stated commercial deployment at leading users in internet and finance. Three numbers were published from the stage. Xie Shaofeng (谢少锋), chair of the OpenAtom Foundation, said cumulative domestic installs of openEuler-family operating systems are expected to pass 22 million units. Xiong Wei (熊伟), chair of the openEuler committee, cited Frost & Sullivan (沙利文) projecting a 61% share of China’s new server-operating-system market in 2026. And the third, which is the one this entry is about: Pakistan’s Higher Education Commission formally released HECOS 1.0, described as the first indigenous operating system outside China built on openEuler.
2. What the community disclosed about itself. The community reports 2,200-plus member organisations and over 30,000 open-source contributors, with deployments across finance, telecom and energy, and entry into satellite systems and embodied intelligence. Huawei’s Peng Honghua (彭红华), president of ICT strategy and business development, spoke as “strategic donor and core contributor” to the community and committed to three things: continuing to output AI core capabilities (sandbox, KV Cache, memory systems, agent security governance); increasing investment in community infrastructure, contributing SkillHub and other AI tools to make community development services agentic; and co-building with industry partners. Xiong Wei stated the strategic pivot explicitly: the community’s focus moves from pursuing market scale to going deep and strong (从追求市场规模,全面转向做深做强), on two mainlines — accelerating into AI, and globalisation. On the community’s own operations, the plan is that by end-2026, 80% of CVEs will be handled with AI assistance and 100% of community services will be agent-supported.
3. The upstream history, stated by the sources themselves. openEuler began as Huawei’s in-house server OS EulerOS, was open-sourced in September 2019, was repositioned in September 2021 from a server OS to “an operating system for digital infrastructure,” and in November 2021 was donated by Huawei with community partners to the OpenAtom Foundation, at which point “the development model shifted from enterprise-led to industry co-building and community self-governance (由企业主导转向产业共建、社区自治).” For scale: in November 2024 openEuler reported cumulative installs passing 10 million after five open years, with over 5 million added in 2024 alone. Today’s 22 million figure means the base roughly doubled in two years.
Institutional Analysis & Commentary:
For a fortnight this series has been measuring a single asymmetry — the ecosystem pays, generously and quickly, for the outbound half of the commons and has no budget line for the inbound half — and today it produced the first counter-example that actually holds.
Begin with what HECOS 1.0 is, institutionally, because the distinction is the whole entry. Every export instrument documented in the previous briefings moved a capability across a border and then kept custody of it: the AI export registration queue makes publication a filing; the model-licence regime governs what a recipient may do with the artefact; the faculty programme trains people who go home and teach a stack whose direction they do not set. A national operating system released by a foreign higher-education commission is a different kind of object. When a ministry-level body in another country ships HECOS 1.0 as its own national system, it acquires the support obligation, the localisation burden, the security-patch relationship and the curriculum — and it acquires a downstream, in the strict open-source sense: it takes the upstream as a dependency and is thereafter structurally interested in the upstream’s health, because its own universities run on it. This is the first time in this run that a Chinese-originated project has left the country and created an institution abroad whose interest is aligned with maintaining it. That is closer to how Red Hat, SUSE and Canonical internationalised than to how an export licence works, and it is worth recording plainly, because a framework that cannot distinguish between the two has stopped being useful.
Second, the strategic pivot is legible and it is the correct one — which is precisely why the interest in it is not disinterested. Xiong Wei’s sentence about moving from market scale to depth is the right sentence for a project at 61% of new domestic server-OS share. A projector that already holds the majority of new installs cannot grow by winning more of the same market; it grows by moving up the value stack (Agentic Infra — sandboxing, KV Cache, memory, agent security, the layers where the margin is) and outward (national distributions abroad). Both moves are rational, and both moves are, from the commons’ point of view, one move: escape the commodity layer where the price is set by whoever copies fastest. The dependency this exposes is honest and worth naming: the 61% is a domestic number, on Linux and on a server market shaped by procurement practice, and the Agentic Infra release is where the project must now earn its position technically rather than institutionally.
Third, and this is the entry’s real subject, the community’s own operating numbers are the most revealing thing published on that stage. By end-2026, 80% of CVEs handled with AI assistance; 100% of community services agent-supported; SkillHub contributed so that community development services become agentic. Read this alongside the size of the community — 2,200 member organisations, 30,000-plus contributors — and a structural problem this series has documented across the whole ecosystem becomes arithmetic: a security-response process is the single most labour-intensive, least glamorous, most commons-generating function any open-source project performs, and its cost scales with the size of the codebase while its practitioners do not scale with the size of the community. Automating 80% of CVE triage is not a stunt; it is the correct answer to a real capacity crisis, and it is the first time in this run that anyone has put a number on the inbound workload. But note what the number does and does not say. Handling 80% of CVEs with AI assistance is a claim about throughput. It is not a claim about who decides whether a fix is right, who signs the advisory, or who carries the disclosure relationship with the reporter. A triage system that classifies and drafts is a magnificent instrument; an advisory that ships on the authority of a system rather than a person is a different institution, and the difference is exactly the difference between a maintainer and a pipeline.
Fourth, the counterweight that the openEuler record earns. This is the project the ecosystem has been running longest, and its history is the cleanest available specimen of the donation-to-foundation transition this series keeps invoking. EulerOS → open-sourced 2019 → repositioned 2021 → donated to OpenAtom November 2021 → self-governance described as the operating model → 10 million installs 2024 → 22 million installs 2026 → a foreign national distribution 2026. That is a seven-year sequence from an in-house asset to a base other countries build their own systems on, and the mechanism is the one identified on 2026-11-13 in the certification context: the project went from Huawei’s product to the floor — the layer 2,200 other organisations can stand on. Whether it is a commons depends on the governance question, not the install count, and the recovery of that question is the point of the next section.
Fifth, the test, stated so that it can fail. If HECOS 1.0 is maintained by the Higher Education Commission’s own people and upstreams patches to openEuler, the Pakistani distribution is a genuine downstream and the internationalisation is real. If HECOS 1.0 is a relabelled openEuler release with a support contract and no local patch flow, it is a regional SKU — and the vocabulary of “first indigenous operating system” will have been borrowed rather than earned. The evidence to look for is unglamorous and checkable: commits to openEuler from Pakistani institutions, a HECOS-specific issue tracker, and whether a HECOS-originated fix ever appears in the upstream repository under a non-Chinese corporate identity. The second test is the one from yesterday, now with a sharper instrument: when openEuler’s AI-assisted pipeline handles 80% of CVEs, who is named on the advisory? If the answer is a named security team, the automation is a tool. If the answer is a system, the ecosystem has automated the one function that requires a person to be responsible.
Checkable questions.
(i) Does HECOS 1.0 have a local repository, and does anything from it land upstream? A national distribution that cannot push a patch upstream is a fork with a support contract. The answer is in git log, not in a press release.
(ii) What is the composition of the 30,000-plus contributors — and specifically, how many hold merge rights? Contributor counts measure participation; committers and maintainers measure governance. openEuler has published the first and not, in these materials, the second.
(iii) Who signs the advisory when the AI pipeline handles 80% of CVEs? A named team means automation under accountability. A system means the accountability has been outsourced to a process at exactly the moment the commons most needs a person’s name.
Sources: 新浪财经 — openEuler系操作系统国内累计装机量预计突破2200万套,预计2026年新增市场份额将攀升至61% (2026-11-13; the Agentic Infra OS release; Xie Shaofeng on 22 million installs; the Pakistan Higher Education Commission and HECOS 1.0; Huawei’s Peng Honghua as strategic donor and core contributor; the three-part contribution commitment including SkillHub) · IT之家 — 开源欧拉 openEuler 发布面向 Agentic Infra 的操作系统,国内累计装机量预计超 2200 万套 (2026-11-13; 24.03 LTS SP5 released by Xiong Wei; the Agentic Infra definition; 2200-plus member organisations and 30,000-plus contributors; the 26.09 innovation release of 2026-10-01 with dual 6.6/6.18 kernels; the Ubuntu SuperPoD collaboration; the new platinum donor; 20 leading commercial-practice cases; the EulerOS → 2019 open-sourcing → 2021 donation history; 2024’s 10 million installs) · 腾讯新闻 / IT时代网 — openEuler升级为智能体基础设施操作系统 (2026-11-13; sandbox, KV Cache, memory system and agent security; the 80% CVE and 100% agent-service targets; HECOS 1.0 as the first overseas indigenous OS built on openEuler) · 搜狐 / 黄海峰看科技 — 2200万装机、61%新增份额!openEuler亮出Agentic时代的开源底牌 (2026-11-13; Xiong Wei’s speech; the pivot from market scale to depth; the four Agentic Infra technical solutions including the open圆融 KVCache project; SuperPoD OS carrying JD.com’s 11.11 traffic peak; ten-plus vendors shipping SuperPoD OS products; the Agentic Infra co-building initiative with China Telecom Cloud, China Mobile and Huawei) · It’s FOSS — SuperPoD OS and Agentic Infra Take Center Stage at openEuler Summit 2026 (2026-11-09; the 13–14 November Beijing programme; 24.03 LTS SP4’s 2,006 developers across 39 member organisations; the Ubuntu SuperPoD session) · This series’ 2026-11-05 briefing — openEuler’s upstream history and the cost of upstream loss · This series’ 2026-11-10 briefing — the openEuler-earmarked donations in OpenAtom’s accounts
🎓 Education & Institutions — Eight Hundred Courses and No Governance Textbook: How HarmonyOS Entered the University
1. The event, and where it was held. On 2026-11-12 China Youth Daily (中国青年报), via Sina Finance, reported on the HarmonyOS University Talent Development Forum (鸿蒙高校人才发展论坛) held on 6 November at CCF ChinaSoft 2026 in Tianjin — the Chinese Software Conference — with fifty-plus faculty from computer-science and software-engineering departments in the Beijing–Tianjin region plus industry representatives. The forum’s stated themes were “the transformation of HarmonyOS teaching paradigms in the AI era, the reconstruction of course forms, and deep industry–academia integration.”
2. The numbers, which are the institutional content. The article reports that HarmonyOS coursework now covers more than 650 universities, that HarmonyOS knowledge has been integrated into more than 800 courses, and that more than 460,000 student developers are registered. Globally, HarmonyOS developers exceed 11 million, and more than 600,000 people hold HarmonyOS application-developer certification. HarmonyOS 6 & 7 device count is expected to cross the 100 million mark in late November. And a line that matters more than any of the above: this year the HarmonyOS University Innovation Competition entered the list of competitions recognised in the National Ranking of Disciplines and Majors in Ordinary Higher Education Institutions (全国普通高校学科竞赛排行榜内竞赛项目) — it became a national-level competition (国赛).
3. The delivery architecture, which is the part nobody advertises. The programme does not run through a foundation or a community; it runs through ministry instruments. The article names them: the Ministry of Education’s industry–academia cooperative education programme (教育部产学合作协同育人项目), the Ministry of Education–Huawei “Smart Base” industry-education integration base (教育部—华为"智能基座"产教融合协同育人基地), and the MoE–Huawei “Smart Base” HarmonyOS-direction course virtual teaching-research office (虚拟教研室). Huawei’s talent pipeline is described as a complete chain — “learn, practise, examine, certify” (学—练—考—证) — with new course directions in AI application development, HarmonyOS PC development and smart-wearable development, and with parallel teacher training so that “course update speed keeps pace with system iteration speed.”
4. What the teachers actually said, because two of them said something worth quoting. Gong Xiaoli (宫晓利), professor at Nankai University’s College of Computer Science and School of Cryptography and Cyber Security, described why he moved HarmonyOS into his long-running operating-systems course: the OS knowledge framework “has barely changed in decades — classic means stable, and stable means slow to update.” Students found the material “irrelevant to real life” and treated it as “the art of dragon-slaying (屠龙之术)” — no enthusiasm. HarmonyOS filled the gap because “it is the system running in their own phones, on the devices around them.” His formulation of the industry–university alignment is the sharpest: “Enterprises have returns from spreading an ecosystem, and the process of spreading the ecosystem simultaneously completes the cultivation of people (企业铺开生态是有收益的,而生态铺开的过程,同时就完成了对人的培养).” And on AI: “Students need to form independent judgement — to judge whether what AI generates is good, whether it is a valuable direction.” He likened the human to “the harness (马具)” on the wild horse that is AI.
Li Gen (李根) of Tianjin University has taught “School–Enterprise Cooperative Practice” since October 2022, with HarmonyOS application development as the main content, requiring a working app delivered in groups at the end. He noted that any HarmonyOS textbook “may have missed the iteration window by the time it is written and published,” so the course uses official documentation and updates dynamically. His course structure after four complete cycles — now in its fifth — is worth naming because it is a curriculum design and not a slogan: “manual baseline — AI collaboration — red-team review.” The first two weeks ban AI tools; students hand-write a minimal runnable version. Then the requirement is rewritten as “executable acceptance criteria” handed to an agent — “beautiful interface, smooth interaction” is not accepted; it must become “cold start under two seconds, no dropped frames on list scroll.” The final stage has AIs attack each other’s work while students judge whether a finding is a real defect, a false positive, or a design trade-off. His summary: “The threshold for tools is falling; the threshold for judgement is rising — that is exactly where a university course belongs.” And: “Code can be generated; responsibility cannot (代码可以生成,责任不能生成).”
5. The labour-market line, offered as the justification. Recruitment-platform data: the HarmonyOS development job-posting index rose from 14.04 in January–April 2025 to 26.11 in the same period of 2026, a 86% increase, and the HarmonyOS talent supply-demand ratio fell from 1.47 to 0.92 — from relative surplus to shortage.
Institutional Analysis & Commentary:
This is the most consequential education story this series has covered, and it is not consequential because of 800 courses. It is consequential because of the word “国赛.”
First, why that one line is the whole story. A curriculum is an offer; a national-competition ranking is a selection instrument. When the HarmonyOS University Innovation Competition entered the Ministry of Education’s ranking list of recognised discipline competitions, it moved from something students may do to something students are measured by — because rankings feed scholarships, graduate admissions, faculty evaluation and institutional assessment. That is the moment a technology becomes a currency in the academic incentive system, and the conversion is not performed by Huawei, by a foundation, or by a community. It is performed by the ministry’s recognition list. The article reports the fact but does not notice the mechanism, which is why it is worth spelling out here. Compare this with the 2026-11-13 certification story: OpenAtom charges a candidate a fee and issues a credential nobody is yet obliged to accept, and this series set the falsifiable test that the credential becomes institutional the day a procurement spec or a maintainer appointment cites it. HarmonyOS has completed that step — not through a hired credential but through the university ranking. The path was different and the destination is the same, and it arrived faster because the instrument already existed.
Second, look at the shape of the curriculum, because it is unusually good, and saying so is not flattery. Gong’s diagnosis is correct and it is an indictment: the operating-systems syllabus taught in Chinese universities has barely changed in decades because classic material is stable and stable material is slow. That is a textbook case of a curriculum with no feedback channel to production. Li Gen’s course structure is the opposite of what an industrial-training programme normally looks like. Banning AI for two weeks to force a manual baseline; requiring requirements to be rewritten as machine-checkable acceptance criteria; then making students judge adversarial output rather than produce it — that is a serious, honest response to the actual condition of software work in 2026. And his sentence — “code can be generated, responsibility cannot” — is an institutional claim, not a motivational one: responsibility is the one input that does not fall in price when generative capacity rises, and it is precisely the input a certification chain cannot manufacture. Compare it with the certification ladders this series examined on 2026-11-08 and 2026-11-13, which measure what a person can do; Li Gen is teaching what a person will be accountable for. The gap between those two things is the gap between a workforce and a profession.
Third, and now the caveats, because there are three and they are real. (a) The article names no foundation. Unlike openEuler, which went through OpenAtom and describes itself as community-governed, HarmonyOS’s university presence runs through MoE programme instruments and one company. The delivery architecture is ministry-plus-vendor, and there is no neutral trustee in the account at all. (b) Gong’s sentence is the most honest thing in the piece and it is also a governance description: “the process of spreading the ecosystem simultaneously completes the cultivation of people.” He means it approvingly, and he is right that it is efficient — but read coldly it says that ecosystem expansion and talent formation are the same operation, which means the curriculum’s purpose is set by the party doing the expanding. That is not a scandal; it is what every vendor education programme does. It is worth stating because the ecosystem’s own vocabulary — co-build, co-share, co-govern — invites the reader to assume a governance that the delivery architecture does not contain. (c) Nothing in the 800 courses is about governance. Course, practice, exam, certification, competition — the chain covers using an operating system comprehensively and building on it well. It does not cover reviewing a contribution, triaging an issue, reading a licence, holding a maintainer’s judgement, or deciding a technical direction in a body where your vote is one of many. Stripped of branding, this is the same asymmetry this series recorded in the certification ladder on 2026-11-13: the ecosystem has industrialised the outbound competence — using, developing, certifying, hiring — and has no curriculum, so far as any of these sources disclose, for the inbound competence on which a commons depends.
Fourth, the deeper point, and it is about what a curriculum does to a community’s reproductive rate. The openEuler entry above ends with a governance question: who holds merge rights, who signs the advisory. This entry supplies the other half of that question: who is coming up. 650 universities and 460,000 registered student developers is a recruitment base of a size no open-source community in the world reproduces organically — the Linux kernel’s contributor graph has never had a pipeline of that shape. If even a small fraction of those students are taught to submit a patch and receive a review, the ecosystem’s longstanding inbound bottleneck dissolves within a cohort. If they are taught to build applications against an API and pass a certification, the ecosystem gets a very large, very competent user base and the bottleneck remains exactly where it is. Both outcomes are consistent with today’s numbers, and the numbers reported are the ones that count users. The question to ask is not how many courses; it is how many of those 800 courses require a merged contribution to pass.
Fifth, the prediction, in the form that can be falsified. If, within two academic years, any of those 650 universities publishes a course whose assessment includes a merged upstream patch — to HarmonyOS, to openEuler, to anything — the pipeline has crossed from consumption to contribution and the distinction this series has drawn for a fortnight will have been answered on the ground rather than in a press release. The observable marker is a syllabus with a link to a repository in it. And the second marker, sharper: whether any HarmonyOS or openEuler community governance document ever cites university coursework as a qualification for anything. On 2026-11-13 this series said no Chinese project had yet cited a certification in its CONTRIBUTING or governance document. That remains true, and the university channel is now the shortest path by which it could stop being true.
Checkable questions.
(i) Of the 800-plus courses, how many require a merged contribution to pass, and to which repository? A course that ends in an app is a training course; a course that ends in a reviewed patch is a commons institution. The syllabus is public at 650 universities.
(ii) What exactly did the “国赛” recognition change about student incentives — credit, scholarship, graduate admission? The answer is in university assessment rules, not in the article. This is the instrument that converted a competition into a currency, and its exchange rate is written somewhere.
(iii) Is there any governance content in the HarmonyOS university curriculum — licence reading, review practice, steering-committee process? If 460,000 student developers are being trained on a platform and none of them is being taught how a platform is governed, the ecosystem is producing users at scale and maintainers by accident.
Sources: 新浪财经 / 中国青年报 — 开课数量超800门,高校加速布局鸿蒙人才培养 (2026-11-12; the HarmonyOS University Talent Development Forum at CCF ChinaSoft 2026, Tianjin, 6 November, fifty-plus faculty; 650-plus universities, 800-plus courses, 460,000-plus registered student developers; 11 million global developers and 600,000-plus certified; HarmonyOS 6 & 7 expecting 100 million devices in late November; the MoE cooperative-education programme, the MoE–Huawei “Smart Base” base and virtual teaching-research office; the learn–practise–examine–certify chain; Gong Xiaoli of Nankai and Li Gen of Tianjin University; the 14.04→26.11 posting index and the 1.47→0.92 supply-demand ratio; the innovation competition entering the national ranking list) · This series’ 2026-11-08 briefing — from a certificate of contribution to a condition of employment · This series’ 2026-11-13 briefing — OpenAtom’s certification fee, the five tracks, and the missing maintainer rung
🔍 Commentary — Two Curricula, One Question
First, today’s two stories are the same instrument pointed in opposite directions, and the contrast is the finding. openEuler sent an operating system out — to a foreign higher-education commission that will run it, localise it, and depend on its upstream. HarmonyOS brought a curriculum in — into 650 universities, 800 courses, 460,000 students, through ministry programme instruments, ending in a national competition ranking. One creates a maintainer abroad; the other creates a user base at home. Both are, by any fair measure, successes, and neither is the thing a commons needs.
Second, the asymmetry this series has tracked for a fortnight has now been located precisely, and it is not where it looked. It is not that China fails to contribute upstream; openEuler’s donation and HECOS 1.0 both refute that. It is not that China does not train people; 460,000 registered student developers refute that. The asymmetry is between two kinds of teaching and two kinds of dependency. A downstream that runs your code will notice when it breaks, and that is a real if indirect channel of accountability. A student who builds apps on your platform will notice when the API changes, and that is a real if indirect channel of demand. Neither is the channel through which a stranger’s patch gets read on a Thursday afternoon. Every instrument in this run — the register, the certificate, the voucher, the residency, the course, the competition — produces people who use or people who carry. Not one of them produces the person who is entitled to say no to a change.
Third, and credit where it is due, because this week earned more than most. HECOS 1.0 is the first export in this run that creates an institution abroad with a stake in the upstream’s health — closer to how open source actually internationalises than any registration queue. Li Gen’s course is the most institutionally serious curriculum design this series has seen from the Chinese ecosystem: manual baseline, executable acceptance criteria, adversarial review, students as judges, and a closing sentence — code can be generated, responsibility cannot — that names the exact input that generative capacity does not cheapen. And Gong Xiaoli’s complaint is correct and should be repeated in every faculty meeting in every country: an operating-systems syllabus that has not changed in decades is not classical, it is disconnected. The fact that a vendor ecosystem was the thing that forced the update is an indictment of the curriculum, not of the vendor.
Fourth, the question that makes the week legible, and it is the same question as yesterday, with a clearer answer available. Who writes the syllabus? On 2026-11-13 the answer was a foundation convening twenty-odd experts across fourteen organisations to build a question bank — a real, if narrow, form of upstream work, and the first budget line in this entire run that pays the people who know the code well enough to set the standard. Today the answer is different and it should be said plainly: at 800 courses across 650 universities, the syllabus is written by the vendor whose platform is being taught, delivered through ministry programme instruments, and now feeds a national ranking. That is an extraordinarily effective distribution machine. It is not a governance institution, and the difference is not a technicality — a governance institution is one in which the people being taught have standing to change what is taught.
Fifth, the sentence to keep. Li Gen teaches in Tianjin that “the threshold for tools is falling; the threshold for judgement is rising.” That is the thesis of every briefing in this run, stated by a teacher rather than an analyst. The ecosystem has built, at remarkable speed and scale, every instrument needed to lower the tools threshold — certifications, courses, competitions, national distributions, agentic community services handling 80% of CVEs. What it has not built, anywhere in these sources, is an instrument that raises the judgement threshold by giving a newcomer the authority to exercise it. A merged patch reviewed by a stranger is that instrument. It costs nothing to install and nobody has installed it. That remains the whole of the finding, and this week it is finally visible in a form that can be checked: 460,000 students, 30,000 contributors, 2,200 member organisations, 22 million installs, one foreign national distribution — and the question of who gets merge rights still has no published answer.
Sources: 新浪财经 / 中国青年报 — 开课数量超800门,高校加速布局鸿蒙人才培养 (2026-11-12; Li Gen’s manual-baseline/AI-collaboration/red-team structure and “code can be generated, responsibility cannot”; Gong Xiaoli on the decades-static OS syllabus and on ecosystem expansion completing talent cultivation) · 新浪财经 — openEuler 2200万装机与61%新增份额 (2026-11-13; HECOS 1.0; the OpenAtom foundation context) · 搜狐 — openEuler 亮出 Agentic 时代的开源底牌 (2026-11-13; the 80% CVE and 100% agent-service targets; 2200-plus member organisations and 30,000-plus contributors) · IT之家 — openEuler 发布面向 Agentic Infra 的操作系统 (2026-11-13; the EulerOS → OpenAtom donation history) · This series’ 2026-11-13 briefing — who gets to be the counterpart · This series’ 2026-11-09 briefing — a foundation that pays its own projects’ rent