⚠️ 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 — 2026-11-03

🏛️ Policy & Regulation — The Word Moves Into the Theme: What the Wuzhen Summit Did With “Open Source” When It Stopped Being a Sub-Forum

1. The event, stated plainly. The 2026 World Internet Conference (WIC) Wuzhen Summit opened on 2026-11-02 and runs to 2026-11-05, hosted by the WIC and organised by the Zhejiang Provincial People’s Government, under the theme — in the WIC’s own English rendering — “Open Source and Openness, Joint Contribution and Shared Benefits — Jointly Building a Community with a Shared Future in Cyberspace.” Reported scope: an opening ceremony, a main forum, and 27 sub-forums; the “Light of the Internet” Expo with 671 companies and institutions from 54 countries and regions showing over 1,000 technological achievements, more than 20 of them debuts, and — for the first time — a dedicated zone for 6G, satellite internet, quantum communication and ultra-wideband. The main forum carries a new “AI Wuzhen Dialogue” replacing last year’s Six Little Dragons session, bringing together “leaders of prominent global technology companies, leading open-source large language model enterprises, and AI scientists” on “building a global open-source AI ecosystem.” The Global Youth Leadership Program, in its fourth edition, drew 2,600-plus candidates from 140-plus countries and selected 20.

2. What is actually new, and what the Chinese coverage says is new. Multiple independent Chinese write-ups state the same claim: the word 开源 (open source) appears in the summit theme for the first time in the conference’s history. Sina’s BigNews piece adds a second, more specific claim — that this is the first time a theme was proposed by Zhejiang and adopted by the organising committee. That second claim is single-sourced and should be treated as reported, not verified. The first is stated consistently enough across Xinhua-syndicated commentary, Guangming Online, Sohu and ifeng that it can be treated as the day’s organising fact.

The trajectory the Chinese commentary itself reconstructs is worth recording, because it is the institutional history rather than the marketing: 2018 — the China RISC-V Alliance founded at the summit (secretary-general Bao Yungang, Institute of Computing Technology, CAS); 2021 — the first Open Source Ecosystem Forum, with CAS vice-president Zhang Yaping calling open source “the crystallization of human wisdom… a major era transformation”; 2022 — an Open Source Technology Ecosystem Innovation Forum, where CAICT vice-president Wei Liang predicted China would produce more application-layer open-source projects because of its scenario volume; 2024 — five domestic OS communities — openEuler, OpenAnolis, OpenCloudOS, Ouge/鸥栖 and openKylin — jointly issued an Open Source Ecosystem Development Cooperation Initiative; 2025 — ROCm as an “AI open-source software platform” and Qwen as an “open and open-source foundation model” took the WIC Leading Technology Award, and the Straight to Wuzhen competition added an open-source track for the first time. Open source went from a sub-forum topic (2021) to a track within a sub-forum’s competition (2025) to the theme of the whole summit (2026). In twelve years, the item moved up the agenda until it became the frame around the agenda.

3. What was built to fill the frame. The concrete objects are public and checkable:

  • The national first open-source-themed experience hall (开源主题生态体验馆) opened at the Wuzhen World Internet Science and Technology Museum on the eve of the summit, showing “major achievements in AI open source and openness” — from base software to large models. Ant Group’s senior vice-president and chair of its Technology Strategy Committee, Ni Xingjun, gave the keynote for the corporate side.
  • Direct-through-Wuzhen global internet competition added an open-source track, split into an open-source model application contest and an open-source challenge contest, “open to global developers for the first time.”
  • “Wuzhen Talk · Open Source Youth Forum” (乌镇Talk·开源青年说) — developers under 40 debating how open models address real industrial needs; Xinhua ran a series of individual talks (earth-science foundation models, mathematics, scientific research teams, astronomy). Opened to public booking — the summit’s own description of 2026 is the first year general visitors could register.
  • The 2027 work item. On the same day, the WIC AI Professional Committee held its 2026 plenary and third chair meeting in Wuzhen, chaired by Wang Jian (CAE academician, Zhejiang Lab director, Alibaba Cloud founder), on-site and online. Members voted on deputy-chair additions, changes to the leads of its three advancement plans — Standards; Security and Governance; Industry — and approved the 2026 summary and the 2027 work plan. In 2026 the committee ran 8 thematic seminars and 6 research discussions, producing 1 research report and 1 initiative. The single forward-looking sentence in the official readout is the one that matters: “In 2027, the committee will actively promote the transformation of the open-source openness initiative into action” — 积极推进开源开放倡议转化为行动 — through seminars, joint research, and published initiatives and outputs. Dame Wendy Hall and Zhang Ya-qin closed with a recommendation that the committee track agent-era risk and deepen research on “the development, application and governance mechanisms of open source and openness.”

4. The documented policy lineage, so the theme is not mistaken for a bolt from the blue. The theme has a paper trail. At WAIC 2025 (Shanghai, 2025-07-26/29), the Ministry of Industry and Information Technology had the China–BRICS AI Development and Cooperation Centre join the OpenAtom Foundation and others to publish an International AI Open Source Cooperation Initiative (《国际人工智能开源合作倡议》) with five directions: lead innovation; cultivate the ecosystem; enable conversion; protect rights — including support for cross-border mutual recognition and compliant use of open-source licences; and open sharing. Direction two is explicit that the state’s role is to support “open-source foundations and other open-source organisations in strengthening communication” and to “guide joint construction of AI standards, promoting coordinated development of the open-source ecosystem and the standardisation process.” In 2026, at the BRICS summit, open-source AI inclusion was elevated again as a state-level proposal. The Wuzhen theme is therefore the naming step in a sequence — initiative (2025) → state proposal (2026) → theme (2026-11) → committee work plan to convert initiative into action (2027).

Institutional Analysis & Commentary:

Start with the honest part of the claim, because it is real and it is not trivial.

A conference theme is not a press release. It is a priority-setting instrument — it determines which submissions are accepted, which sub-forums exist, which executives fly in, which competition tracks carry prize money, and which exhibits the museum installs. An organisation that puts a word into its theme for the first time has spent something real: the annual, non-repeatable slot that could have said AI, or digital economy, or inclusion. Zhejiang and the WIC spent it on open source. For anyone who tracks this ecosystem, that is a genuine signal and it should be recorded as one — the kind of signal that is costly to fake and therefore informative. And the substance behind it is not nothing: from the RISC-V alliance (2018) through five OS communities cooperating (2024) to ROCm and Qwen winning the technology award (2025), there is a ten-year record of real artefacts.

The institutional problem is what kind of instrument a conference theme is.

Williamson’s L1→L4 sequence is useful here precisely because it predicts which layer a slogan can move. A theme is L1 — social embeddedness, the taken-for-granted vocabulary by which actors describe what they are doing. The two layers that determine whether anything is actually open are L2 (the formal rules: licences, contribution agreements, IP regimes, procurement codes) and L3 (the governance arrangements: who holds the copyright, who holds the trademark, who decides what merges, who bears the residual cost of maintenance). A theme cannot reach L2 or L3. No conference theme has ever merged a pull request. So the correct reading of 2026-11-02 is not “China has become more open” but “the vocabulary layer has moved ahead of the rule layer” — which is either a leading indicator or a substitution, and the 2027 committee plan is the test.

That is why the AI committee’s single operative sentence deserves more weight than the entire theme.

“Actively promote the transformation of the open-source openness initiative into action.” Set this beside the committee’s 2026 output accounting: 8 seminars, 6 discussions, 1 report, 1 initiative — every one of those is a meeting or a document. Not one is a licence, a deposit, an access rule, or a maintained repository. The unit of account is the event. In 2027 the unit of account is scheduled to change: initiative → action. If the conversion happens, this is a genuine upgrade and the theme will have been a leading indicator. If the 2027 output accounting again reads in seminars and reports, then the theme was substitution — a norm stated at L1 in place of a rule written at L2 — and the vocabulary became a substitute for the institution rather than a precursor to it.

Three specific tests, all cheap, all checkable, all restating no more than the committee’s own words.

(i) Does the initiative become a text with articles? An action has a document number, a scope, and a party that bears obligations. The WAIC 2025 initiative is five directions and a call to “join hands” — five aspirations addressed to “global developers, enterprises, academic institutions and public departments.” Aspirations do not bind. Test: does the 2027 output name who does what, by when, and who verifies? Note that the WAIC 2025 initiative already contains the most concrete and most interesting commitment in the whole corpus — support for cross-border mutual recognition and compliant use of open-source licences (支持开源许可协议跨境互认和合规使用). Mutual recognition of licences is a hard, legal, checkable L2 commitment with real consequences for Chinese firms shipping code abroad and foreign projects accepting Chinese contributions. If 2027 produces anything, it should produce progress on that sentence. If it produces another “initiative,” the pattern is confirmed.

(ii) Does the experience hall’s inventory get a licence? The hall is a display of “major achievements in AI open source and openness.” A hall is an exhibition; a repository is an offer. They differ by exactly the instruments open source runs on: a licence, a copy of the code or weights anyone can pull, an access rule, and a named maintainer with obligations. Every exhibit in an experience hall is simultaneously a publication and not a commons — to reuse the distinction this file has drawn before about sectoral datasets. The test is embarrassingly simple: is there anything in the hall a remote developer could download that is not already downloadable from GitHub or ModelScope? If the answer is yes, the hall added something. If it is no, the hall added visibility, which is not the same thing, and which open source does not need because its default distribution channel is already public.

(iii) Do 40,000 pieces of code get read? The Open Source Youth Forum selected young developers and moved a competition onto an open-source track. A competition selects a winner, not a committer. Participation in a contest is a tournament — and tournaments are excellent incentive devices; they produce effort. But they do not produce maintainers, because maintainership requires the one input a tournament cannot supply: a second party who depends on your code and will file a bug against it next year. So the right question about the youth forum is not how good the talks were. It is: did any of the young developers who presented end up with commit rights in a repository outside their own organisation? That is the single metric that separates a developer ecosystem from a talent showcase, and it is not reported — which is, once again, the ordinary incompleteness of a first-year event and not a scandal.

Now the sharpest line in the whole day, and it comes from a company, not a ministry.

At the opening of the experience hall, Ant Group’s Ni Xingjun said: “We look forward to open source becoming the basic public good of the digital world” (期待开源能够成为数字世界的基础公共品), and added the correct justification — that for developers and small firms in developing countries, “high software licensing fees and difficulty participating in core technology R&D remain real barriers,” and that “wherever you are and however senior you are, every participant can contribute equally and benefit equally.” He also named the three conditions for sustainability: maintain community trust through open collaboration with global firms and developers; take seriously open-source governance and sustainability — long-term maintenance, community governance, contributor incentives; and build a “contribution — use — re-contribution” positive cycle.

That is, almost verbatim, the specification of a commons. It is also the most institutionally sophisticated paragraph any corporate speaker has contributed to this subject all year, and it is being said by a company that also open-sourced OceanBase (5 million-plus lines, 200,000-plus deployed clusters per its own account), Ant Design (99,000-plus GitHub stars) and the LingBot embodied-AI model series (2.2 million downloads in nine months). None of those numbers is independently verified for this briefing — they are the firm’s own accounting. But the form of the claim is exactly right, and the form is where the institutional content lives.

The problem is that “public good” is being used where the Chinese open-source system has been built to produce a “club good.”

A public good is non-excludable and non-rival: anyone may use it, and one person’s use does not diminish another’s. A club good is non-rival but excludable: shared among members, closed to non-members. Everything in the architecture described above is club-shaped. The WIC is a membership organisation. The AI Professional Committee has members and deputy chairs and three advancement plans with named leads — it is a governance body with a roster. The experience hall admits visitors. The competition admits entrants. OpenAtom is funded by member dues and runs projects it has accepted into incubation — its own 2025 accounting puts it at 60-plus projects receiving full-lifecycle support. Each of these is a legitimate and effective way to organise collective action. Each of them has a gate. The thing about a gate is that it is the definition of excludability, and the thing about open source is that its entire institutional innovation was the discovery that you can produce a commons without a gate — by attaching a licence to a copy and letting the copy do the admitting.

So the ambition in Ni’s sentence is real and the structure is not yet the one that would satisfy it. “Basic public good” is not achieved by a hall that people visit. It is achieved by a licence that people accept. And the gap between those two is not rhetorical — it is exactly the gap between L1 and L2.

One more thing about the framing, and this is where 「开源之道」 has to be precise.

A recurring line in this year’s Chinese commentary positions open source against “small yard, high fence” (小院高墙) and technological blocs, with the WIC secretary-general Ren Xianliang framing the theme as “transcending the practice of building walls and drawing distinctions between near and far”, and Chinese analysts describing the US as trending toward closed source and “silos.” Read charitably, this is a real argument: open licensing genuinely does lower entry costs, and the fact that a foundation model can be downloaded by a university in Lagos is genuinely different from a model available only through an API contract.

But the institutional claim has to be made carefully, because it can be made in a way that destroys itself. Open source is not a position in a geopolitical contest; it is a set of licence terms. The whole force of the argument against walls comes from the fact that an Apache-2.0 or MIT file cannot be revoked, cannot be geo-fenced, and cannot be held hostage — that is what makes it credible. The moment open source is described as a national path, a civilisational choice, or our side’s model, that specific force is spent: a norm that is asserted as a national instrument is a norm that can be withdrawn as a national instrument, and everyone can see that. The strongest possible defence of open source in this argument is a document number and a licence header, not a theme. If the objective really is to make open source the default for the next decade of AI, the 2027 deliverable that would settle it is embarrassingly small and embarrassingly concrete: a published list of Chinese-origin AI projects whose weights, training code and evaluation harnesses are under OSI-approved licences, with a named maintainer and a public issue tracker for each. That is a list a foreign developer can act on. A theme is a list a foreign developer can quote.

What to watch, concretely, between now and the 2027 summit.

Whether the AI committee’s 2027 plan produces a document with obligations instead of another initiative. Whether the experience hall publishes an item-level inventory with licences attached — the way the culture-and-tourism pilot-test base published its dataset counts, and then did not publish an access rule. Whether the cross-border licence mutual recognition sentence from WAIC 2025 gets any legal follow-through, because that single clause is worth more to the global commons than the entire theme. And whether the 27 sub-forums on agent interoperability, embodied intelligence and standards produce one shared specification with an implementation anyone outside the room may adopt — or fourteen more declarations of intent. The summit ends 2026-11-05. Whatever it produces will have been produced by then, and the committee’s next plenary is when we learn whether “initiative into action” was a plan or a way of postponing one.


Sources: China Daily — WIC Wuzhen Summit to unveil new vision for open source, openness (2026-10-23) · Global Times — Open source and practical AI take center stage at Wuzhen Summit · Guangming Online via Sina — 乌镇,何以"开源" (2026-10-30) · Sohu — 开源不只是代码 今年乌镇为何把开源放在C位? (2026-11-02) · ifeng/Cover News — 乌镇峰会开源主题生态体验馆开馆 (2026-11-02) · Guangming Online via QQ News — 世界互联网大会人工智能专业委员会 2026 全体会议及第三次主任委员会议 (2026-11-03) · Xinhua — 20项成果获2026年世界互联网大会领先科技奖 (2026-11-02) · CAC — 2026 精品案例在乌镇发布 (2026-11-02) · OpenAtom Foundation — 联合发布《国际人工智能开源合作倡议》(WAIC 2025) · Guangming Online via Sina — 乌篷载得星河去,开源引得世界来 (2026-11-02)