Posts

“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.”

“CAICT (China Academy of Information and Communications Technology), jointly with COPU (Cloud Computing Open Source Industry Alliance), formally releases the ‘Token Factory · Intelligence-Computing Capability Grading Evaluation System’ — a five-tier (T1–T5), five-dimension administrative-grading apparatus that converts China’s proliferating ‘Token Factory’ compute-facility market from an unregulated ‘price-per-token’ competition surface into a state-anchored, certificate-based governance surface with operational/design dual-track certification; COPU honorary chairman Lu Shouqun (陆首群) is formally invited to serve on the chairmanship group of the GOAI World AI Open-Source Competition (杭州云谷中心, 5-million-RMB prize pool, four tracks) — a structurally new institutional-bridge event converting Lu Shouqun from the COPU-internal governance axis to a GOAI-foundation external-anchor axis, validating the ‘COPU-chairman-as-ecosystem-bridge’ hypothesis with a concrete appointment; Huawei opens the Northeast-region registration window of the Ascend CANN HCCL Communication-Library Innovation Contest (hosted by Harbin Institute of Technology, prizes up to 20,000 RMB plus Huawei fast-track hiring) — an empirical observation of Huawei’s talent-pipeline institution that routes university developers into the Ascend+CANN+GitCode ecosystem through certificates + prizes + employment-incentive stacking, converting ‘community contribution’ into a recruitment-funnel input.”

“Huawei’s CANN (Compute Architecture for Neural Networks) heterogeneous-compute compiler stack is formally open-sourced at the 2026 CCF China Open Source Conference — a structurally new institutional move that shifts the Chinese AI-chip-compiler ecosystem from a proprietary-vendor-stack architecture to an open-compiler-institution architecture, paralleling the BAAI FlagOS/FlagTree move (August 20 briefing) and completing a two-source confirmation of the ‘domestic AI-chip compiler open-source’ institutional claim; Tiangong Kaiwu Open Source Foundation and MoLe (Modelers) Community announce a formal strategic partnership — the first institutional bridge between a state-anchored ‘Chinese-characteristic’ open-source foundation and a CSDN-anchored developer-platform community, with the explicit framing ‘构建中国特色AI开源协作新范式’ (construct a Chinese-characteristic AI open-source collaboration paradigm); ASPI (Australian Strategic Policy Institute) analyst Alex Colville publishes ‘Some Chinese AI will go closed-source, but not all’ (August 2026) — a structurally new institutional observation that deconstructs Xi Jinping’s WAIC ‘开源开放’ (open-source-and-openness) slogan as a conceptually-substituted umbrella term spanning models, tools, datasets, and chip architectures, allowing frontier-model weights to be withheld under the same slogan.”