China

“凤凰 AI研究院 dialogue with Wang Chenglu (王成录, deepin AI Labs / 深开鸿 CEO, ex-Huawei HarmonyOS co-creator, 24-year Huawei veteran, Harbin Institute of Technology PhD) publishes the first-documented institutional reading of the OpenHarmony base layer by the ex-Huawei architect who now runs the ecosystem’s second-largest contributor: OpenHarmony has reached 14 billion device installs, of which only ~3 billion are Huawei phones and 11 billion are non-Huawei devices — the article’s own claim; Huawei’s contribution share inside the OpenHarmony community has declined from ~90% to ~70% (Wang Chenglu’s own framing of a “natural process”); 哈勃投资 (Hubble Investment, Huawei’s venture arm) held a 20% stake in 深开鸿 at founding — an outlier for Hubble’s usual <10% rule; 深开鸿 has been led for four years by Wang Chenglu after he left Huawei in 2022 with what Ren Zhengfei framed as “the widest moat for Huawei itself” — an institutional fact about the vendor-foundation separation that the previous briefings’ Western-neutral-foundation question did not document at the vendor-official agenda layer; the article documents the M-Robots OS donation to OpenAtom Foundation and OpenAtom’s own official timeline of M-Robots OS 1.0 (April 2025) → full open-sourcing (July 2025) → donation with independent PMC (November 2025) → 2.0 release (May 2026), which is the vendor-foundation separation pattern being applied to a new domain (robot operating systems) with a 76% dependency-on-foreign-ROS base as the diagnostic; 工信部 (MIIT) is described as pushing a unified IoT interconnection standard for OpenHarmony — an administrative-industrial-policy placement that pairs with the September 6 briefing’s MIIT AI SME Startup Support Plan; the article’s own claim that 深开鸿’s 2028 target is to become “a pure ecosystem-platform enterprise, like today’s Google or Microsoft” and that “the base layer for the Internet of Everything is most likely to be realized in China” — a structurally new vendor-official claim that pairs with the “万物互联的底座=未来谷歌微软” framing in the WeChat input’s own metadata.”

“Two institutionally new placements on the AAIF / Western-neutral-foundation question in the same 48-hour window — (1) Huawei’s Chief Open-Source Contact Officer Zhang Shanzhou (张山舟, the article’s own name for Ren Xudong’s AAIF-side role at AGNTCon + MCPCon China 2026, Shanghai, September 6–7) delivers the first Huawei keynote on the Agentic AI Foundation (AAIF) that formally names Huawei as one of the first Chinese enterprises to join AAIF (首批加入 AAIF 的中国企业), offers an L1–L5 Agent-AI evolution ladder with three agent-native paradigm shifts (project / organization / infrastructure), positions openJiuwen at 40K+ stars / 3M downloads / 650 contributors / 50+ industry deployments, introduces A2A-T (a telco-scoped extension of Anthropic’s A2A protocol) and OpenAN (open-sourced with China Mobile and Orange), and closes with a lineage framing of Huawei’s 2015 CNCF founding membership as the precedent for today’s AAIF entry — a structurally new institutional move that absorbs the vendor-official analytical layer’s AAIF-seat recommendation (from the September 5 Ren Xudong essay) into a vendor-official agenda; (2) COPU’s WeChat account, citing COPU Deputy Secretary-General Tan Zhongyi’s on-the-ground confirmation, documents that Lu Shouqun’s keynote at KubeCon China 2026 is scheduled as the opening session (September 8, 09:00, Shanghai International Convention Center) and that CNCF + OpenInfra + PyTorch will jointly present Lu Shouqun with a "Lifetime Achievement Award for Open Source Leader" (开源领袖终身成就奖) — the first-documented Western-foundation-layer formal award to a COPU-named Chinese institutional voice at the KubeCon-China stage, and a structurally new move on the same Western-neutral-foundation question the September 5 Ren Xudong essay first raised.”

“MIIT’s office issued the AI SME Startup Support Plan 2026-2028 on September 4, 2026 — a 15-item three-year programme that makes five institutionally new moves on the same unsolved object as the prior four briefings (Ren Xudong’s governance-layer-deficit essay on September 5; CAC’s Qinglang Phase 2 on September 3; Chongqing’s 15th-Five-Year-Plan on September 2; COPU’s KubeCon address on September 2): (1) Item 9 formally names AtomGit as ’national-level AI open-source community’ (国家级人工智能开源社区), the first administrative-document designation of AtomGit at the national-community level and a structural confirmation of AtomGit’s role as the institutional anchor of China’s open-weight release apparatus; (2) Item 4 explicitly names ‘open-source community core contributors’ as a new entrepreneurial subject with an encouraged commercialization path — the first administrative recognition that open-source contribution itself is an entrepreneurial subject class in China; (3) Item 13 introduces compute-equity and data-equity arrangements (算力入股、数据入股), an institutional innovation that treats compute and data as equity inputs into SME ventures rather than market-priced factor inputs; (4) Item 4 also explicitly names ‘one-person companies / super individuals’ (一人公司 / 超级个体) as an encouraged entrepreneurial subject — a structurally new administrative recognition of the individual contributor as an SME-form object; (5) the evaluation-metrics clause (last paragraph) explicitly names open-source project star counts as one of four official performance indicators for identifying high-growth-potential AI SMEs — the first administrative-document use of the GitHub-style star metric as an institutional evaluation metric in China.”

“Huawei’s Chief Open-Source Strategy Officer Ren Xudong (任旭东) publishes in Silicon Time (硅基时间) the first systematic institutional-framework essay from a Chinese vendor official on AI open-source governance — a four-layer openness taxonomy (open interface / open weights / open training / open governance), a three-layer competition model (model / runtime / governance), and an explicit claim that China’s most dangerous open-source weakness is at the governance layer, where Chinese participation in AAIF, MCP, and Open VSX is nearly absent; a four-layer Chinese policy recommendation (strategy / mechanism / ecosystem / governance) is offered as a systematic proposal, and the essay is the first documented instance of a Huawei official positioning Chinese open source not as model-cost-advantage but as an institutional-governance deficit that requires a Chinese-seat strategy at neutral Western foundations.”

“COPU’s September 1 weekly meeting minutes, co-signed by Lu Shouqun (陆首群, acting in his first COPU-named role as honorary chairman) and Jiang Tao (蒋涛), document two institutional moves on the open-weight layer in a single sitting: (1) Lu Shouqun’s first address to a Western foundation — KubeCon China 2026 (September 7–9, Shanghai International Convention Center, co-located with OpenInfra Summit + PyTorch Conference, per Linux Foundation’s June 19 press release) — is drafted with a three-axis framework (performance / cost-reduction / invocation) and a ‘50-vs-2’ open-weight framing, explicitly casting the 50 founding members of WAICO as a majority vote against Anthropic and AWS on the open-weights question; (2) Jiang Tao’s essay ‘Chinese large models are world-class’ is attached in full to the COPU minutes, publishing its four-account-book framework and eight-layer architecture in a COPU-branded institutional surface — a structurally new placement that the prior Silicon Time briefing (August 31) did not document. In parallel, Chongqing’s 15th-Five-Year-Plan AI framework, analyzed by the Tiangong Kaiwu Open Source Foundation (天工开物开源基金会), is the first city-level 15th-Five-Year-Plan AI plan to elevate ‘open source’ from a technical option to an explicitly named ‘industrial infrastructure’ and ‘organizational form of industrial ecology’ — alongside explicit clauses on ‘open-source-safety testing, supply-chain maturity assessment, and supply-chain-break risk prevention’, making Chongqing the first municipal plan in which open-source infrastructure and compliance are codified in the same document.”

“Moonshot AI closes a $3.5B round at a $35B valuation — tripled in six months — with China’s National AI Industry Investment Fund as anchor, before reaching out for a new round at a $50B pre-money valuation ahead of a Hong Kong IPO; Reuters reports DeepSeek is planning a fresh raise at ~$74B valuation ahead of a mainland IPO, with the same National AI Industry Investment Fund negotiating to become lead investor — meaning China’s two largest frontier labs are being consolidated with state-directed capital at a combined valuation approaching $85B, an industrial-policy consolidation rather than competitive-market dynamics; a separate institutional layer — the tiered-governance apparatus for open-weight models (Beijing MoC + NDRC meetings with Alibaba/ByteDance/Z.ai documented by Reuters July 7; formalized by a May Supreme People’s Court roundtable summary; analyzed by Jamestown August 13) — moves from academic-scholarship to policy-formation, defining open-weight not as a community norm but as a state-graded release category; Chatham House’s August 13 article documents how China’s open-weight strategy is structurally the counter-offer to a U.S. closed-weight model, and how both Washington and Beijing are now converging on restricting access to their frontier AI.”

“Huawei’s openJiuwen has four papers accepted at EMNLP 2026 — 2 Findings, 1 Industry Track, 1 Demo — publishing the first time a Huawei 2012-Lab open-source agent platform has received top-tier ACL-family academic endorsement at the industry-track level, using peer-reviewed academic legitimacy to promote commercial products; Jiang Tao (硅基时间/Silicon Time) publishes the second half of ‘Chinese large models are world-class, so why is the developer ecosystem still in others’ hands?’ with the first quantified ‘derivative-mass gap’ — 34,000 Qwen derivatives on ModelScope vs. 151,000 on Hugging Face — an empirical anchor for the default-path institutional-economics thesis; Xie Lan (清华AIRI院长, State Council Advisory Group) writes in 《中国社会科学报》 — China’s most authoritative social-science official media — explicitly using the ‘Great Divergence’ term in AI governance discourse and positioning ‘open source + high-quality foundation models’ as an instrument of cultural sovereignty and diplomatic statecraft.”

Brief: key policy, legal, and institutional developments in Chinese open source

“CAICT’s ATH (Agent Trusted Handshake) protocol, initiated May 2026 and released on AtomGit+GitHub, has crossed a structural threshold: 《中国改革》(China Reform, Caixin’s governance-policy flagship journal) formally cites ATH and its OSCAR-Trustworthy-Open-Source stewardship in its July 2026 ‘Intelligent-Agent Governance’ cover-feature special section (Issue 448, dated 2026-07-01) — a first documented institutional-legitimacy event that carries CAICT’s agent-governance technical standard from the industry-standards layer into the national policy-discourse layer, with 季卫东 (Ji Weidong, SJTU China-Institute-of-Law-and-Society director) as the senior-academic co-anchor; 凤凰WEEKLY财经 (Phoenix Weekly / Fortune) publishes ‘海信出海之路并不好走’ (Hisense’s overseas-expansion road is not smooth sailing) — an empirical data-driven piece documenting Hisense’s 28.22% domestic vs 11.21% overseas gross-margin gap, providing a fresh data point on the ‘出海 as institutional rhetoric’ thesis first developed in August 2026’s institutional-analysis program: when gross margins collapse by more than half overseas, ‘going global’ becomes not market expansion but the institutionalization of domestic competitive-failure through the ‘going-global’ rhetorical frame.”