⚠️ 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 — 2026-10-10

🏗️ Institutional Change — Bloomberg’s October 6, 2026 Reporting Documents DeepSeek’s Pre-IPO Round of At Least ¥80 Billion with CATL and Tencent as the Largest Committed Investors, Paired with Sina/Phoenix Tech’s Documentation that DeepSeek’s June 2026 First External Round Was Executed as a Limited-Partnership Structure in Which External Investors Hold No Voting Rights and a Five-Year Lockup While Liang Wenfeng Retains Close to 100 Percent of Voting Rights

1. Bloomberg (October 6, 2026, per RFI/Sina/iHeima re-reporting) — DeepSeek approaching at least ¥80 billion (approximately US$11.93 billion) in pre-IPO round, with final size potentially reaching ¥100 billion.

2. DeepSeek — round originally targeted at ¥50 billion with a target valuation of ¥500 billion (approximately US$74 billion); subscriptions currently at ¥80 billion, meaning the raised amount has already significantly exceeded the original target.

3. CATL (宁德时代, China’s largest battery manufacturer) and Tencent Holdings — the two largest committed investors in the round.

4. Reuters (via RFI, September 30 to October 6, 2026) — DeepSeek is expected to close the round in October 2026.

5. Sina/Phoenix Tech (October 8, 2026) — DeepSeek’s June 2026 first external round raised over ¥50 billion (approximately US$7.4 billion), the largest single AI funding round in the Chinese industry to date at the time of completion.

6. DeepSeek June 2026 round — founder Liang Wenfeng (梁文锋) personally contributed approximately ¥20 billion, Tencent contributed approximately ¥10 billion, CATL group contributed approximately ¥5 billion.

7. DeepSeek June 2026 round governance structure — limited-partnership (有限合伙) architecture; external investors hold no voting rights and share lockup of five years; Liang Wenfeng retains close to 100 percent of voting rights.

8. Reuters (September 2026) — DeepSeek engaged CITIC Securities (中信证券) to prepare a potential STAR-board IPO; listing timing, valuation, and raise size not yet finalized; a 2027 filing is now documented in the Sina/Phoenix Tech piece.

9. Bloomberg Intelligence (October 5, 2026) — post-V4.1-Flash launch, the US-China top-model benchmark testing gap has narrowed to approximately 3 percent, down from approximately 9 percent in May 2026 and approximately 15 percent at the start of the year.

10. DeepSeek — V4.1-Flash released September 10, 2026 with native multimodal visual understanding; GPQA Diamond 90.9; LiveBench rank 6th with a score of 81.1, the highest Chinese-model rank since DeepSeek’s 2025 R1 breakout, at significantly lower inference price points than Anthropic and OpenAI comparable systems.

11. Bloomberg Intelligence — Tencent’s stated motivation for the round is to integrate AI capabilities into its platform portfolio; CATL’s stated motivation is to supply components and data-center infrastructure for the AI industry.

12. DeepSeek — Huawei Ascend chip collaboration continues, including software adaptations to Ascend hardware (CANN adaptations); DeepSeek’s V4-series open-sourced under MIT license on April 24, 2026 with V4-Pro and V4-Flash both supporting 1 million-token context windows and “thinking / non-thinking” dual modes.

Bloomberg’s October 6, 2026 reporting on DeepSeek’s approaching pre-IPO round, paired with the Sina/Phoenix Tech October 8, 2026 piece documenting the governance structure of DeepSeek’s June 2026 first external round, is the first-documented instance in this series of a Chinese-frontier-AI-lab pre-IPO round being operationalized at the limited-partnership-plus-founder-retained-99-percent-voting-rights-plus-external-investor-five-year-lockup-plus-CITIC-Securities-STAR-board-preparation institutional-form layer simultaneously with this series’ October 7 briefing’s CNBC-DeepSeek-15-billion-dollar-funding-round-consideration documentation, this series’ October 7 briefing’s Reuters-DeepSeek-12-billion-funding-round documentation, and this series’ October 8 briefing’s Crypto Briefing DeepSeek-plus-Huawei-Ascend-950DT-160000-chip-1-gigawatt documentation.

The sharpest institutional-economics fact of the day: DeepSeek’s June 2026 first external round was executed as a limited-partnership structure in which external investors hold no voting rights and share lockup of five years, while founder Liang Wenfeng personally contributed approximately ¥20 billion out of a total raise of over ¥50 billion, and retained close to 100 percent of voting rights. Per Sina/Phoenix Tech: “该轮交易采用有限合伙架构，外部投资方多不具投票权、股份锁定五年，梁文锋仍掌握近100%表决权” (The round was executed as a limited-partnership structure; external investors hold no voting rights and shares are locked for five years; Liang Wenfeng still holds close to 100 percent of voting rights). The institutional reading is that the DeepSeek-limited-partnership-plus-99-percent-voting-rights-plus-five-year-lockup layer is being operationalized at the Chinese-frontier-AI-lab pre-IPO governance-form layer — a first-documented instance in this series of a Chinese-frontier-AI-lab pre-IPO round being executed as an institutional-form object that separates capital-influx (¥50 billion-plus first round, ¥80 billion-plus pre-IPO round) from voting-control (Liang Wenfeng’s approximately 99 percent voting rights retained through the two rounds and a five-year external-share lockup).

The capital-influx-plus-voting-control separation layer. Per Sina/Phoenix Tech: the June 2026 round raised over ¥50 billion (approximately US$7.4 billion) — Liang Wenfeng ¥20 billion, Tencent ¥10 billion, CATL ¥5 billion — and the round’s governance structure is a limited-partnership with external investors holding no voting rights and a five-year share lockup. Per the same piece: “在最新一轮中继续加码，腾讯的意图并不难理解，彭博社称，该公司希望将AI能力融入旗下各个平台；宁德时代看中的，则是为AI行业供应零部件、为数据中心供货这一日渐有利可图的生意” (Tencent’s motivation is not hard to understand — Bloomberg reports that Tencent wants to integrate AI capabilities into its platforms; CATL’s motivation is to supply components and data-center infrastructure for the AI industry). The institutional reading is that the capital-influx-plus-voting-control separation layer is being operationalized at the Chinese-frontier-AI-lab pre-IPO governance-form layer simultaneously with the DeepSeek-limited-partnership-plus-99-percent-voting-rights-plus-five-year-lockup layer — a two-layer-simultaneous DeepSeek pre-IPO governance-form observation.

The CATL-plus-Tencent dual-lead-plus-DeepSeek-voting-rights-plus-Huawei-Ascend-inference-transition layer. Per Bloomberg: CATL and Tencent are the two largest committed investors in the current pre-IPO round; CATL’s motivation is supplying components and data-center infrastructure for the AI industry; Tencent’s motivation is integrating AI capabilities into its platforms. Per Crypto Briefing (this series’ October 8 briefing): DeepSeek is planning to deploy at least 160,000 Huawei Ascend 950DT chips at Ulanqab, Inner Mongolia data center starting Q4 2026 or Q1 2027, and is pulling forward the Ascend 960DT training chip launch to Q1 2027. The institutional reading is that the CATL-plus-Tencent dual-lead-plus-DeepSeek-voting-rights-plus-Huawei-Ascend-inference-transition layer is being operationalized at the Chinese-frontier-AI-lab pre-IPO capital-influx-plus-sovereign-hardware-stack-institutional-form layer simultaneously with the DeepSeek-limited-partnership-plus-99-percent-voting-rights-plus-five-year-lockup layer.

The CITIC-Securities-STAR-board-preparation-plus-reorganization-plus-headcount-expansion layer. Per Reuters (September 2026): DeepSeek has engaged CITIC Securities to prepare a potential STAR-board IPO. Per Sina/Phoenix Tech (October 8, 2026): “本轮融资落定后，DeepSeek下一步将是推进公司重组，为IPO做准备。另一方面，凤凰网科技从多个不同信源处了解到，融资完成后DeepSeek的主要任务之一也是扩充公司人员规模” (After this round closes, DeepSeek’s next step is to advance the company reorganization to prepare for IPO. Separately, Phoenix Tech has learned from multiple sources that after the funding closes, one of DeepSeek’s main tasks is to expand the company’s headcount). The institutional reading is that the CITIC-Securities-STAR-board-preparation-plus-reorganization-plus-headcount-expansion layer is being operationalized at the Chinese-frontier-AI-lab pre-IPO-institutional-form-transition-plus-organizational-scale-transition layer simultaneously with the DeepSeek-limited-partnership-plus-99-percent-voting-rights-plus-five-year-lockup layer — a two-layer-simultaneous DeepSeek pre-IPO institutional-form observation.

The Bloomberg-Intelligence-3-percent-US-China-benchmark-gap-plus-V4.1-Flash-LiveBench-rank-6 layer. Per Bloomberg Intelligence (October 5, 2026, per Sina/Phoenix Tech): the US-China top-model benchmark testing gap has narrowed to approximately 3 percent since V4.1-Flash launch, down from approximately 9 percent in May 2026 and approximately 15 percent at the start of the year. Per Sina/Phoenix Tech: DeepSeek V4.1-Flash ranks 6th on LiveBench with a score of 81.1, the highest Chinese-model rank since DeepSeek’s 2025 R1 breakout, at significantly lower inference price points than Anthropic and OpenAI comparable systems. The institutional reading is that the Bloomberg-Intelligence-3-percent-US-China-benchmark-gap-plus-V4.1-Flash-LiveBench-rank-6 layer is being operationalized at the Chinese-frontier-AI-lab pre-IPO-numerical-evidence-plus-capability-evidence layer simultaneously with the DeepSeek-limited-partnership-plus-99-percent-voting-rights-plus-five-year-lockup layer — the capability-evidence layer providing the basis on which the capital-influx-plus-voting-control separation layer is being executed.

The institutional-object pairing with this series’ October 7 briefing’s CNBC-DeepSeek-15-billion-dollar-funding-round-consideration documentation and this series’ October 7 briefing’s Reuters-DeepSeek-12-billion-funding-round documentation. This series’ October 7 briefing documented the CNBC and Reuters rounds of coverage on DeepSeek’s funding round consideration. This cycle’s briefing documents the same DeepSeek funding-round surface at the Bloomberg-October-6-approaching-80-billion-yuan-plus-final-size-potentially-reaching-100-billion-yuan-plus-CATL-and-Tencent-dual-lead layer simultaneously with this series’ October 7 briefing’s CNBC-15-billion-dollar-funding-round-consideration layer and this series’ October 7 briefing’s Reuters-12-billion-funding-round layer — a two-source-convergent documentation of the same DeepSeek funding round at the approaching-80-billion-yuan-plus-CATL-Tencent-dual-lead layer.

The institutional-object pairing with this series’ October 8 briefing’s Crypto Briefing DeepSeek-plus-Huawei-Ascend-950DT-160000-chip-1-gigawatt documentation. This series’ October 8 briefing documented Crypto Briefing’s September 22, 2026 exclusive on DeepSeek’s plan to deploy 160,000 Huawei Ascend 950DT chips at a new Ulanqab data center. This cycle’s briefing documents the same DeepSeek institutional object at the pre-IPO-round-plus-CITIC-Securities-STAR-board-preparation-plus-June-2026-limited-partnership-99-percent-voting-rights layer simultaneously with the same DeepSeek-plus-Huawei-Ascend-sovereign-hardware-stack surface this series’ October 8 briefing documented at the 160000-chip-Ulanqab-plus-1-gigawatt layer. The institutional reading is that the same DeepSeek institutional object is being documented at two institutionally distinct layers — the pre-IPO-institutional-form layer and the sovereign-hardware-stack layer — in the same 48-hour window.

The institutional-economics question this cycle’s briefing documents but does not answer. Whether the DeepSeek-limited-partnership-plus-99-percent-voting-rights-plus-five-year-lockup-plus-CITIC-Securities-STAR-board-preparation apparatus converts the Chinese-frontier-AI-lab pre-IPO governance form from a capital-influx-plus-voting-control object into a specific institutional-form object at the same layer at which the DeepSeek-plus-Huawei-Ascend-sovereign-hardware-stack surface this series’ October 8 briefing documented at the 160000-chip-Ulanqab-plus-1-gigawatt layer, and whether the same separation of capital-influx from voting-control apparatus converts the same DeepSeek institutional object into a founder-control-plus-state-capital-influx-plus-strategic-platform-corporation-synergy-object rather than a normal-pre-IPO-object at the same institutional layer.

Institutional significance: Bloomberg’s October 6, 2026 reporting plus Sina/Phoenix Tech’s October 8, 2026 piece is the first-documented instance in this series of the DeepSeek-80-billion-yuan-pre-IPO-round-plus-CATL-Tencent-dual-lead-plus-100-billion-yuan-target-plus-June-2026-limited-partnership-plus-99-percent-voting-rights-plus-five-year-lockup-plus-CITIC-Securities-STAR-board-preparation institutional-form layer being operationalized at the international-mass-media-numerical-evidence-plus-governance-structure-documentation-plus-organizational-reorganization-plus-headcount-expansion-plus-3-percent-US-China-benchmark-gap-plus-V4.1-Flash-LiveBench-rank-6 layer simultaneously with this series’ October 7 briefing’s CNBC-15-billion-dollar-funding-round-consideration layer, this series’ October 7 briefing’s Reuters-12-billion-funding-round layer, and this series’ October 8 briefing’s Crypto Briefing DeepSeek-plus-Huawei-Ascend-950DT-160000-chip-1-gigawatt layer — a four-surface-simultaneous DeepSeek pre-IPO institutional-form observation pairing capital-influx-plus-voting-control separation with sovereign-hardware-stack-plus-STAR-board-preparation-plus-organizational-reorganization layers.

Sources:

🏗️ Institutional Change — Hongxing/Red Star News’s October 6, 2026 Reporting Documents Moonshot AI (月之暗面) Completing its Final Pre-IPO Round at a Valuation of Approximately US$50 Billion with a Q1 2027 Hong Kong Exchange Listing Target, Paired with Sina/Hongxing's Documentation that Moonshot AI has Completed the Business-Registration Change from Limited Liability Company to Joint Stock Company as a Pre-IPO Procedural Signal, Paired with the Piece's Documentation of Moonshot AI's Six Funding Rounds in 2026 with Valuation Rising from US$10 Billion to US$35 Billion Across January-to-July

1. Hongxing/Red Star News (October 6, 2026, per Sina Technology re-reporting) — Moonshot AI completed its final pre-IPO private funding round at a valuation of approximately US$50 billion and is advancing toward a Q1 2027 Hong Kong Exchange IPO.

2. Moonshot AI — investor-intent preparation has already begun; preliminary investor meetings possibly starting this month; IPO timing may still shift.

3. Moonshot AI valuation progression (2026): January-to-February three consecutive rounds raising valuation from US$10 billion to US$18 billion; May D-round to US$20 billion; June E-round pre-money to US$31.5 billion; July F-round to US$35 billion; final pre-IPO round to approximately US$50 billion.

4. Six funding rounds in the first half of 2026; valuation roughly 5x within six months.

5. Moonshot AI — flagship Kimi K3 launch in July 2026: launched 2 days after release, ranked 1st on Hugging Face’s international open-source AI model trends within 30 minutes of release.

6. Moonshot AI CEO (president) Zhang Yutong (张予彤, July 18, 2026) — post-K3 enterprise ARR grew several-fold; June 2026 ARR broke through US$300 million.

7. Moonshot AI founder Yang Zhilin (杨植麟, born 1992) — National Informatics Olympiad Guangdong First Prize, Tsinghua University Computer Science department (graduated top of class, 2015), Carnegie Mellon University PhD; pre-entrepreneurship experience at Google Brain and Meta AI; returned to China in 2023 to found Moonshot AI; founded the company at age 31.

8. Moonshot AI — angel round of approximately ¥2 billion (approximately US$290 million) completed within less than two months of founding.

9. Moonshot AI — completed business-registration change from “limited liability company” to “joint stock company” (有限责任公司变更为股份有限公司), which is a common procedural precursor to IPO preparation.

10. Moonshot AI — founder Yang Zhilin’s personal equity, employee equity, and board governance structure have not been publicly documented by Moonshot AI.

Hongxing/Red Star News’s October 6, 2026 reporting plus the Sina/Hongxing documentation of Moonshot AI’s business-registration change is the first-documented instance in this series of a Chinese-frontier-AI-lab pre-IPO round being operationalized at the limited-liability-company-to-joint-stock-company-conversion-plus-dual-track-Hong-Kong-and-A-share-pre-IPO-signal-plus-six-rounds-in-first-half-of-2026-plus-US$10-billion-to-US$50-billion-valuation-increase-in-six-months-plus-Kimi-K3-Hugging-Face-rank-1-plus-ARR-300-million-dollars layer simultaneously with this series’ October 7 briefing’s Euronews-Moonshot-50-billion-valuation-Hong-Kong-IPO-eyes documentation and this series’ September 24 briefing’s Moonshot-50-billion-dollar-valuation-plus-pre-IPO-preparation documentation.

The sharpest institutional-economics fact of the day: Moonshot AI has completed the business-registration change from “limited liability company” to “joint stock company” as a procedural precursor to IPO, paired with a final pre-IPO round at approximately US$50 billion, paired with a six-round valuation progression from US$10 billion to US$50 billion across January-to-October 2026 (approximately 5x within ten months), paired with a June 2026 annual recurring revenue (ARR) that broke through US$300 million on the back of Kimi K3’s July 2026 launch. Per Sina/Hongxing: “月之暗面已经完成工商变更登记，市场主体类型由’有限责任公司’变更为’股份有限公司’，这一股改动作通常被视为企业筹备上市的前置信号” (Moonshot AI has completed the business-registration change; the market entity type has been changed from “limited liability company” to “joint stock company”; this stock-change action is commonly regarded as a pre-IPO precursor signal). The institutional reading is that the Moonshot-limited-liability-to-joint-stock-conversion-plus-pre-IPO-round-plus-6-rounds-valuation-progression-plus-Kimi-K3-plus-ARR-300-million layer is being operationalized at the Chinese-frontier-AI-lab pre-IPO procedural-plus-valuation-plus-business-registration-change layer — a first-documented instance in this series of a Chinese-frontier-AI-lab business-registration change being executed at the pre-IPO procedural layer simultaneously with the Moonshot-6-rounds-valuation-progression layer.

The six-round-valuation-progression-plus-dual-track-Hong-Kong-and-A-share-pre-IPO-signal layer. Per Sina/Hongxing: Moonshot AI’s 2026 valuation progression is characterized as “月之暗面2026年的估值增速极快，1至2月连续完成3轮融资，估值从100亿美元攀升至180亿美元；5月D轮融资后估值达200亿美元；6月E轮融资后投前估值升至315亿美元；7月F轮融资后达350亿美元” (Moonshot AI’s 2026 valuation growth is extremely fast; January-to-February three consecutive rounds raised the valuation from US$10 billion to US$18 billion; May D-round valuation reached US$20 billion; June E-round pre-money valuation rose to US$31.5 billion; July F-round valuation reached US$35 billion). Per Sina/Hongxing: the piece also references 17173 (September 10, 2026) and Phoenix Tech (September 10, 2026) reporting that Moonshot AI is considering dual listing in Shanghai and Hong Kong to raise more capital. The institutional reading is that the six-round-valuation-progression-plus-dual-track-Hong-Kong-and-A-share-pre-IPO-signal layer is being operationalized at the Chinese-frontier-AI-lab pre-IPO valuation-progress-and-listing-strategy layer simultaneously with the Moonshot-limited-liability-to-joint-stock-conversion-plus-pre-IPO-round-plus-6-rounds-valuation-progression layer.

The Kimi-K3-plus-ARR-300-million-dollars layer. Per Sina/Hongxing: “今年7月，Kimi K3上线两天就被挤爆。Kimi K3发布后在硅谷引发轰动效应，上线仅30分钟便登顶国际开源AI模型社区Hugging Face趋势榜。月之暗面总裁张予彤7月18日在社交媒体披露，K3模型发布后企业ARR有了数倍的快速增长。6月中旬，月之暗面ARR突破3亿美元” (In July this year, Kimi K3’s server was crowded within 2 days of launch. Kimi K3’s release caused a sensation in Silicon Valley; it topped Hugging Face’s international open-source AI model trends within 30 minutes of release. Moonshot AI CEO Zhang Yutong disclosed on July 18 on social media that after the K3 model launch, enterprise ARR grew several-fold; mid-June, Moonshot AI’s ARR broke through US$300 million). The institutional reading is that the Kimi-K3-plus-ARR-300-million-dollars-plus-Hugging-Face-rank-1 layer is being operationalized at the Chinese-frontier-AI-lab pre-IPO-numerical-evidence-plus-business-evidence layer simultaneously with the Moonshot-6-rounds-valuation-progression-plus-dual-track-pre-IPO-signal layer — the business-evidence layer providing the basis on which the valuation-progression layer is being executed.

The Yang-Zhilin-founder-profile-plus-Google-Brain-plus-Meta-AI-plus-Tsinghua-CMU-layer. Per Sina/Hongxing: Moonshot AI founder Yang Zhilin (杨植麟, born 1992) was a National Informatics Olympiad Guangdong First Prize winner, received early admission to Tsinghua University Computer Science department, graduated top of his class in 2015, then pursued a PhD at Carnegie Mellon University. Prior to founding Moonshot AI, he worked at both Google Brain and Meta AI, the two leading AI research labs. In 2023, Yang Zhilin returned to China to found Moonshot AI; the company completed a near-¥2-billion angel round within less than two months of founding. The institutional reading is that the Yang-Zhilin-founder-profile-plus-Google-Brain-plus-Meta-AI-plus-Tsinghua-CMU layer is being operationalized at the Chinese-frontier-AI-lab founder-institutional-background layer simultaneously with the Moonshot-6-rounds-valuation-progression-plus-dual-track-pre-IPO-signal layer — the founder-institutional-background layer providing the institutional-form context for the valuation-progression-plus-pre-IPO-signal layer.

The institutional-object pairing with this series’ October 7 briefing’s Euronews-Moonshot-50-billion-valuation-Hong-Kong-IPO-eyes documentation. This series’ October 7 briefing documented the Euronews Moonshot-50-billion-dollar-valuation-plus-Hong-Kong-IPO-eyes documentation. This cycle’s briefing documents the same Moonshot institutional object at the Hongxing-October-6-final-pre-IPO-round-plus-ARR-300-million-dollars-plus-Kimi-K3-plus-business-registration-change-plus-6-rounds-valuation-progression layer simultaneously with this series’ October 7 briefing’s Euronews-50-billion-dollar-valuation-plus-Hong-Kong-IPO-eyes layer — a two-source-convergent documentation of the same Moonshot institutional object at the approaching-50-billion-dollar-plus-business-registration-change-plus-6-rounds-valuation-progression layer.

The institutional-object pairing with this series’ October 1 briefing’s National-AI-Industry-Investment-Fund’s-8.8B-capital-consolidation-of-Moonshot-at-35B-and-DeepSeek-at-74B layer. This series’ October 1 briefing documented the National AI Industry Investment Fund’s $8.8B-in-capital consolidation of Moonshot at $35B and DeepSeek at $74B valuation. This cycle’s briefing documents the same Moonshot institutional object at the Hongxing-October-6-final-pre-IPO-round-plus-50-billion-dollar-valuation-plus-business-registration-change-plus-6-rounds-valuation-progression layer and the same DeepSeek institutional object at the Bloomberg-October-6-approaching-80-billion-yuan-plus-final-100-billion-yuan-target layer. The institutional reading is that the same two Chinese-frontier-AI-labs this series’ October 1 briefing documented at the National-AI-Industry-Investment-Fund capital-consolidation layer are now being documented at the pre-IPO-round-plus-business-registration-change-plus-dual-track-listing-strategy layer simultaneously — the state-capital-consolidation layer giving way to the pre-IPO-institutional-form-transition layer.

The institutional-economics question this cycle’s briefing documents but does not answer. Whether the Moonshot-limited-liability-to-joint-stock-conversion-plus-pre-IPO-round-plus-6-rounds-valuation-progression-plus-dual-track-Hong-Kong-and-A-share-pre-IPO-signal apparatus converts the Chinese-frontier-AI-lab pre-IPO institutional form from a state-capital-consolidation object into a specific pre-IPO-institutional-form object at the same layer at which this series’ October 1 briefing documented the state-capital-consolidation layer, and whether the same apparatus converts the same Moonshot institutional object into a specific pre-IPO-plus-founder-control-plus-platform-corporation-synergy-object rather than a state-capital-consolidation-object at the same institutional layer.

Institutional significance: Hongxing/Red Star News’s October 6, 2026 reporting plus the Sina/Hongxing documentation is the first-documented instance in this series of the Moonshot-50-billion-dollar-pre-IPO-round-plus-business-registration-change-plus-6-rounds-valuation-progression-plus-Kimi-K3-Hugging-Face-rank-1-plus-ARR-300-million-dollars-plus-dual-track-Hong-Kong-and-A-share-pre-IPO-signal-plus-Yang-Zhilin-founder-profile institutional-form layer being operationalized at the international-mass-media-numerical-evidence-plus-business-evidence-plus-business-registration-change-plus-listing-strategy layer simultaneously with this series’ October 7 briefing’s Euronews-Moonshot-50-billion-valuation-Hong-Kong-IPO-eyes layer and this series’ October 1 briefing’s National-AI-Industry-Investment-Fund-consolidation-of-Moonshot-at-35-billion-dollar layer — a three-surface-simultaneous Moonshot pre-IPO institutional-form observation pairing business-registration-change with valuation-progression and dual-track-listing-strategy layers.

Sources:

🏗️ Institutional Change — Sina/Phoenix Tech’s October 8, 2026 Documentation of DeepSeek Harness’s National-Day-Holiday 2026 Release of an Experimental Claude Code Mods Plugin Compatibility Layer, Paired with DeepSeek Harness Head Cui Tianyi’s October 4, 2026 Social-Media Statement That “The Main Purpose of Building This Compatibility Layer Is to Verify Whether the Extension Capabilities Claude Code Mods Offers Plugin Authors Are Roughly a Subset of the Capabilities Provided by DSH’s Everything-Is-a-Plugin Architecture”, Paired with Cui Tianyi’s Own Statement That “A Fairly Large Portion of My Daily Working Time Is Spent Recruiting and Interviewing for DeepSeek; DeepSeek’s Goal Is to Achieve AGI, and DeepSeek Harness’s Goal Is to Advance AGI from the Harness Angle”

1. DeepSeek Harness (DSH) — during the 2026 National Day holiday (October 1-8, 2026) released a new version enhancing plugin and cross-platform capabilities, adding an experimental Claude Code Mods compatibility layer.

2. DeepSeek Harness head Cui Tianyi (崔添翼, October 4, 2026 social media) — “做这个兼容层的主要目的，是验证Claude Code Mods向插件作者提供的扩展能力，是否大致是DSH’一切皆插件’架构所提供能力的一个子集” (The main purpose of building this compatibility layer is to verify whether the extension capabilities Claude Code Mods offers plugin authors are roughly a subset of the capabilities provided by DSH’s everything-is-a-plugin architecture).

3. DeepSeek Harness head Cui Tianyi — “我每天工作的相当一部分时间都用于为DeepSeek招人和面试，DeepSeek的目标是实现AGI，DeepSeek Harness组的目标就是从Harness的角度推进AGI” (A fairly large portion of my daily working time is spent recruiting and interviewing for DeepSeek; DeepSeek’s goal is to achieve AGI, and DeepSeek Harness’s goal is to advance AGI from the Harness angle).

4. DeepSeek Harness — “everything-is-a-plugin” (一切皆插件) architecture is characterized by Cui Tianyi as the reference architectural comparison against which Claude Code Mods’s extension capabilities are being tested.

5. Claude Code Mods — an Anthropic (Claude Code) plugin ecosystem; DeepSeek Harness’s compatibility layer is being built to determine whether Claude Code Mods’s plugin-author extension capabilities are a subset of DSH’s capabilities.

6. DeepSeek Harness — DeepSeek’s agent infrastructure product line; the compatibility-layer release is positioned as an agentic-infra-plus-agent-harness-plus-plugin-ecosystem comparison experiment rather than a commercial interop release.

Sina/Phoenix Tech’s October 8, 2026 documentation of DeepSeek Harness’s experimental Claude Code Mods compatibility layer, paired with Cui Tianyi’s October 4, 2026 social-media statement characterizing the compatibility-layer purpose as a subset-testing experiment, is the first-documented instance in this series of a Chinese-frontier-AI-lab agent-infrastructure product line being operationalized at the experimental-Claude-Code-Mods-compatibility-layer-plus-everything-is-a-plugin-architecture-plus-subset-testing-experiment institutional-form layer simultaneously with this series’ October 7 briefing’s DeepSeek-Huawei-Ascend-sovereign-hardware-stack-plus-CANN-adaptations layer and this series’ September 20 briefing’s Tencent-WeChat-OpenClaw-framework-1-billion-users-and-Alibaba-Agent-Native-Cloud layer.

The sharpest institutional-economics fact of the day: DeepSeek Harness head Cui Tianyi characterizes the compatibility layer’s purpose as a “subset-testing experiment” — verifying whether Claude Code Mods’s plugin-author extension capabilities are roughly a subset of DeepSeek Harness’s everything-is-a-plugin architecture’s capabilities, and separately characterizing his own daily work as dominated by recruiting and interviewing for DeepSeek. Per Cui Tianyi: “做这个兼容层的主要目的，是验证Claude Code Mods向插件作者提供的扩展能力，是否大致是DSH’一切皆插件’架构所提供能力的一个子集” (The main purpose of building this compatibility layer is to verify whether the extension capabilities Claude Code Mods offers plugin authors are roughly a subset of the capabilities provided by DSH’s everything-is-a-plugin architecture). Per Cui Tianyi: “我每天工作的相当一部分时间都用于为DeepSeek招人和面试，DeepSeek的目标是实现AGI，DeepSeek Harness组的目标就是从Harness的角度推进AGI” (A fairly large portion of my daily working time is spent recruiting and interviewing for DeepSeek; DeepSeek’s goal is to achieve AGI, and DeepSeek Harness’s goal is to advance AGI from the Harness angle). The institutional reading is that the Cui-Tianyi-subset-testing-experiment-plus-everything-is-a-plugin-architecture-plus-recruiting-and-interviewing layer is being operationalized at the Chinese-frontier-AI-lab agent-infrastructure-plus-agentic-infra-plus-organizational-scale-transition layer — a first-documented instance in this series of a Chinese-frontier-AI-lab agent-infrastructure product line being operationalized at the experimental-compatibility-layer-plus-subset-testing-experiment-plus-agentic-infra-plus-organizational-scale-transition institutional-form layer.

The Claude-Code-Mods-plus-DeepSeek-Harness-plus-AI-agent-harness-architecture layer. Per Cui Tianyi’s October 4, 2026 statement, DeepSeek Harness’s compatibility layer is not a commercial interop product but a subset-testing experiment — testing whether Claude Code Mods’s plugin-author extension capabilities are a subset of DeepSeek Harness’s everything-is-a-plugin capabilities. The institutional reading is that the Claude-Code-Mods-plus-DeepSeek-Harness-plus-AI-agent-harness-architecture layer is being operationalized at the cross-lab agent-infrastructure-architecture-comparison institutional-form layer — a structurally distinct institutional-form object from the cross-lab open-source-ecosystem-partnership layer this series’ September 20 briefing documented at the Tencent-WeChat-OpenClaw-framework-plus-Alibaba-Agent-Native-Cloud layer.

The “everything-is-a-plugin” (一切皆插件) architecture-as-reference-plus-agentic-infra-plus-agent-harness institutional-form layer. Cui Tianyi’s own characterization of DeepSeek Harness’s everything-is-a-plugin architecture as the reference against which Claude Code Mods is being tested is the sharpest institutional-economics fact of the day in this section. The institutional reading is that the everything-is-a-plugin-architecture-as-reference-plus-agentic-infra-plus-agent-harness layer is being operationalized at the Chinese-frontier-AI-lab agent-infrastructure-architecture-reference institutional-form layer — a first-documented instance in this series of a Chinese-frontier-AI-lab agent-infrastructure architecture being characterized as the reference against which a Western-frontier-AI-lab’s plugin-author extension architecture is tested at the subset-testing-experiment layer.

The “DeepSeek Harness组的目标就是从Harness的角度推进AGI” (DeepSeek Harness group’s goal is to advance AGI from the Harness angle) layer. Cui Tianyi’s own characterization of DeepSeek Harness’s role inside DeepSeek is that DeepSeek Harness’s goal is to advance AGI from the Harness angle — DeepSeek’s overarching goal being to achieve AGI. The institutional reading is that the DeepSeek-Harness-group-goal-of-advancing-AGI-from-the-Harness-angle-plus-Cui-Tianyi-recruiting-and-interviewing layer is being operationalized at the Chinese-frontier-AI-lab agent-infrastructure-organizational-form-plus-organizational-scale-transition institutional-form layer simultaneously with the DeepSeek-Harness-Claude-Code-Mods-compatibility-layer layer — the same DeepSeek institutional object this cycle’s briefing documents at the DeepSeek-80-billion-yuan-pre-IPO-round-plus-99-percent-voting-rights-plus-CITIC-Securities-STAR-board-preparation layer.

The institutional-object pairing with this series’ October 7 briefing’s DeepSeek-Huawei-Ascend-sovereign-hardware-stack-plus-CANN-adaptations layer. This series’ October 7 briefing documented the CNBC DeepSeek-15-billion-dollar-funding-round-consideration layer plus the Reuters DeepSeek-12-billion-funding-round layer plus the DeepSeek-plus-Huawei-Ascend-sovereign-hardware-stack layer. This cycle’s briefing documents the same DeepSeek institutional object at the DeepSeek-80-billion-yuan-pre-IPO-round-plus-CATL-Tencent-dual-lead-plus-99-percent-voting-rights-plus-CITIC-Securities-STAR-board-preparation layer simultaneously with the DeepSeek-Harness-Claude-Code-Mods-compatibility-layer-plus-everything-is-a-plugin-architecture-plus-Cui-Tianyi-recruiting-and-interviewing layer. The institutional reading is that the same DeepSeek institutional object is being documented at three institutionally distinct layers — the pre-IPO-institutional-form layer, the sovereign-hardware-stack layer, and the agent-infrastructure-architecture layer — in a 72-hour window.

The institutional-object pairing with this series’ September 20 briefing’s Tencent-WeChat-OpenClaw-framework-1-billion-users-and-Alibaba-Agent-Native-Cloud layer. This series’ September 20 briefing documented the Tencent-WeChat-OpenClaw-framework-1-billion-users and Alibaba-Agent-Native-Cloud layer. This cycle’s briefing documents the same Chinese-AI-agent-infra surface at the DeepSeek-Harness-Claude-Code-Mods-compatibility-layer-plus-everything-is-a-plugin-architecture layer. The institutional reading is that the same Chinese-AI-agent-infra surface this series’ September 20 briefing documented at the WeChat-OpenClaw-framework-plus-Agent-Native-Cloud layer is now being operationalized at the DeepSeek-Harness-Claude-Code-Mods-compatibility-layer-plus-everything-is-a-plugin-architecture layer — a two-layer-simultaneous Chinese-AI-agent-infra institutional-form observation.

The institutional-economics question this cycle’s briefing documents but does not answer. Whether the DeepSeek-Harness-Claude-Code-Mods-compatibility-layer-plus-everything-is-a-plugin-architecture-plus-Cui-Tianyi-recruiting-and-interviewing apparatus converts the Chinese-AI-agent-infra surface into a cross-lab-agent-infrastructure-architecture-reference-object rather than a cross-lab-open-source-ecosystem-partnership-object, and whether the same apparatus converts the same DeepSeek institutional object this cycle’s briefing documents at the DeepSeek-80-billion-yuan-pre-IPO-round-plus-99-percent-voting-rights-plus-CITIC-Securities-STAR-board-preparation layer into a three-layer-simultaneous DeepSeek institutional-form object at the pre-IPO-institutional-form layer, the sovereign-hardware-stack layer, and the agent-infrastructure-architecture layer.

Institutional significance: Sina/Phoenix Tech’s October 8, 2026 documentation is the first-documented instance in this series of the DeepSeek-Harness-Claude-Code-Mods-experimental-compatibility-layer-plus-everything-is-a-plugin-architecture-as-reference-plus-subset-testing-experiment-plus-Cui-Tianyi-recruiting-and-interviewing-plus-DeepSeek-Harness-group-goal-of-advancing-AGI-from-the-Harness-angle institutional-form layer being operationalized at the Chinese-frontier-AI-lab agent-infrastructure-architecture-reference-plus-cross-lab-agent-infrastructure-architecture-comparison layer simultaneously with this series’ October 8 briefing’s Crypto Briefing DeepSeek-plus-Huawei-Ascend-950DT-160000-chip-1-gigawatt layer, this series’ October 7 briefing’s CNBC-DeepSeek-15-billion-dollar-funding-round-consideration layer, this series’ October 7 briefing’s Reuters-DeepSeek-12-billion-funding-round layer, and this series’ September 20 briefing’s Tencent-WeChat-OpenClaw-framework-1-billion-users-and-Alibaba-Agent-Native-Cloud layer — a five-surface-simultaneous DeepSeek institutional-form observation pairing pre-IPO-institutional-form, sovereign-hardware-stack, and agent-infrastructure-architecture layers.

Sources:

🔍 Commentary

Three institutionally dense placements on the same unsolved institutional object this series’ prior fifty briefings have been documenting — the pre-IPO governance form and agent-infrastructure architecture of a Chinese frontier AI lab — and all three placements this cycle’s briefing documents sit at institutional layers this series’ prior briefings have not documented at before:

  1. Bloomberg’s October 6, 2026 reporting plus Sina/Phoenix Tech’s October 8, 2026 piece on DeepSeek’s pre-IPO round at the DeepSeek-80-billion-yuan-pre-IPO-round-plus-CATL-Tencent-dual-lead-plus-100-billion-yuan-target-plus-June-2026-limited-partnership-plus-99-percent-voting-rights-plus-five-year-lockup-plus-CITIC-Securities-STAR-board-preparation institutional-form layer (October 6, 2026 - October 8, 2026). This is a first-documented instance in this series of a Chinese-frontier-AI-lab pre-IPO round being operationalized at the limited-partnership-plus-founder-retained-99-percent-voting-rights-plus-external-investor-five-year-lockup-plus-CITIC-Securities-STAR-board-preparation institutional-form layer simultaneously with this series’ October 7 briefing’s CNBC-15-billion-dollar-funding-round-consideration layer, this series’ October 7 briefing’s Reuters-12-billion-funding-round layer, and this series’ October 8 briefing’s Crypto Briefing DeepSeek-plus-Huawei-Ascend-950DT-160000-chip-1-gigawatt layer.

  2. Hongxing/Red Star News’s October 6, 2026 reporting plus the Sina/Hongxing documentation on Moonshot AI’s pre-IPO round at the Moonshot-50-billion-dollar-pre-IPO-round-plus-business-registration-change-plus-6-rounds-valuation-progression-plus-Kimi-K3-Hugging-Face-rank-1-plus-ARR-300-million-dollars-plus-dual-track-Hong-Kong-and-A-share-pre-IPO-signal-plus-Yang-Zhilin-founder-profile institutional-form layer (October 6, 2026). This is a first-documented instance in this series of a Chinese-frontier-AI-lab pre-IPO round being operationalized at the limited-liability-company-to-joint-stock-company-conversion-plus-dual-track-pre-IPO-signal-plus-6-rounds-valuation-progression institutional-form layer simultaneously with this series’ October 7 briefing’s Euronews-Moonshot-50-billion-valuation-Hong-Kong-IPO-eyes layer and this series’ October 1 briefing’s National-AI-Industry-Investment-Fund-consolidation-of-Moonshot-at-35-billion-dollar layer.

  3. Sina/Phoenix Tech’s October 8, 2026 documentation of DeepSeek Harness’s experimental Claude Code Mods compatibility layer plus Cui Tianyi’s October 4, 2026 statement at the DeepSeek-Harness-Claude-Code-Mods-experimental-compatibility-layer-plus-everything-is-a-plugin-architecture-as-reference-plus-subset-testing-experiment-plus-Cui-Tianyi-recruiting-and-interviewing-plus-DeepSeek-Harness-group-goal-of-advancing-AGI-from-the-Harness-angle institutional-form layer (October 4, 2026 - October 8, 2026). This is a first-documented instance in this series of a Chinese-frontier-AI-lab agent-infrastructure product line being operationalized at the experimental-compatibility-layer-plus-everything-is-a-plugin-architecture-plus-subset-testing-experiment-plus-agentic-infra-plus-organizational-scale-transition institutional-form layer simultaneously with this series’ October 7 briefing’s CNBC-DeepSeek-15-billion-dollar-funding-round-consideration layer, this series’ October 7 briefing’s Reuters-DeepSeek-12-billion-funding-round layer, this series’ October 8 briefing’s Crypto Briefing DeepSeek-plus-Huawei-Ascend-950DT-160000-chip-1-gigawatt layer, and this series’ September 20 briefing’s Tencent-WeChat-OpenClaw-framework-1-billion-users-and-Alibaba-Agent-Native-Cloud layer.

One structural pattern across the three placements: the same unsolved institutional object — the pre-IPO governance form and agent-infrastructure architecture of a Chinese frontier AI lab — is being operationalized at three institutionally distinct layers simultaneously in this cycle’s briefing. The DeepSeek pre-IPO round operationalizes the boundary at the limited-partnership-plus-99-percent-voting-rights-plus-STAR-board-preparation layer; the Moonshot pre-IPO round operationalizes the boundary at the business-registration-change-plus-6-rounds-valuation-progression-plus-dual-track-listing layer; and the DeepSeek Harness Claude Code Mods compatibility layer operationalizes the boundary at the experimental-compatibility-layer-plus-everything-is-a-plugin-architecture-plus-subset-testing-experiment layer. A three-surface-simultaneous Chinese-frontier-AI-lab pre-IPO-institutional-form-and-agent-infrastructure-architecture boundary observation that this series’ prior fifty briefings have not documented at before.

One structural risk across the three placements: the same pre-IPO-institutional-form-and-agent-infrastructure-architecture institutional object this cycle’s briefing documents on three institutionally distinct layers may convert the Chinese-frontier-AI-lab pre-IPO institutional form from an institutional-form object into a specific operational outcome object — the DeepSeek pre-IPO round into a founder-control-plus-state-capital-influx-plus-platform-corporation-synergy-object rather than a state-capital-consolidation-object; the Moonshot pre-IPO round into a dual-track-listing-plus-business-registration-change-plus-founder-control-object rather than a state-capital-consolidation-object; and the DeepSeek Harness Claude Code Mods compatibility layer into a cross-lab-agent-infrastructure-architecture-reference-object rather than a cross-lab-open-source-ecosystem-partnership-object. The three-surface-simultaneous institutional-form-translation pattern is the same pattern this series’ October 8 briefing documented at the NDA-embodied-AI-standards-plus-agentic-AI-framework layer and the DeepSeek-plus-Huawei-Ascend-sovereign-hardware-stack layer.

  • The DeepSeek pre-IPO round placement pairs institutionally with this series’ October 7 briefing’s CNBC-DeepSeek-15-billion-dollar-funding-round-consideration layer, this series’ October 7 briefing’s Reuters-DeepSeek-12-billion-funding-round layer, this series’ October 8 briefing’s Crypto Briefing DeepSeek-plus-Huawei-Ascend-950DT-160000-chip-1-gigawatt layer, and this series’ October 1 briefing’s National-AI-Industry-Investment-Fund-consolidation-of-DeepSeek-at-74-billion-dollar layer.
  • The Moonshot pre-IPO round placement pairs institutionally with this series’ October 7 briefing’s Euronews-Moonshot-50-billion-valuation-Hong-Kong-IPO-eyes layer and this series’ October 1 briefing’s National-AI-Industry-Investment-Fund-consolidation-of-Moonshot-at-35-billion-dollar layer.
  • The DeepSeek Harness Claude Code Mods compatibility layer placement pairs institutionally with this series’ September 20 briefing’s Tencent-WeChat-OpenClaw-framework-1-billion-users-and-Alibaba-Agent-Native-Cloud layer and this series’ October 8 briefing’s Forkast-Deployment-State-may-8-July-15-agent-framework layer.

The institutional-economics question — which this cycle’s briefing documents but does not answer — is whether the Chinese-frontier-AI-lab pre-IPO-institutional-form-and-agent-infrastructure-architecture institutional-form object this cycle’s briefing documents on three institutionally distinct layers simultaneously — the DeepSeek pre-IPO-round layer, the Moonshot pre-IPO-round layer, and the DeepSeek Harness Claude Code Mods compatibility-layer layer — can hold the same institutional-form object at three institutionally distinct layers without one converting the other, and whether the same three-surface-simultaneous institutional-form-translation pattern this cycle’s briefing documents signals a broader Chinese-frontier-AI-lab pre-IPO-institutional-form-and-agent-infrastructure-architecture object being operationalized at the state-capital-influx-plus-voting-control-separation-plus-agent-infrastructure-architecture-reference layer.

One perspective, not a verdict.

The three placements this cycle’s briefing documents — Bloomberg’s and Sina/Phoenix Tech’s documentation of DeepSeek’s pre-IPO round, Hongxing/Red Star News’s and Sina/Hongxing’s documentation of Moonshot AI’s pre-IPO round plus business-registration change, and Sina/Phoenix Tech’s documentation of DeepSeek Harness’s experimental Claude Code Mods compatibility layer — are best read as observations of institutional movement in progress, not as verdicts on institutional direction.

Editorial note on perspective: This briefing presents one institutional-economics reading of Chinese open-source developments, not a verdict. The “institutions” in this story — DeepSeek’s pre-IPO round governance structure, Moonshot AI’s pre-IPO round plus business-registration change, and DeepSeek Harness’s experimental Claude Code Mods compatibility layer — are treated as objects of observation, not targets of critique. The Great Divergence 2.0 framework (FLOSS vs. State-Chartered Codebase vs. Intranet Shared Source vs. Cyber-Eestate) and the Williamson L1→L4 institutional-economics reading (L1 social embedding → L2 institutional environment → L3 governance mechanisms → L4 resource allocation) are lenses, not universal answers. One perspective, not a verdict.

Deduplication note: The three stories this cycle’s briefing documents — Bloomberg’s and Sina/Phoenix Tech’s documentation of DeepSeek’s ¥80 billion pre-IPO round (October 6, 2026 - October 8, 2026); Hongxing/Red Star News’s and Sina/Hongxing’s documentation of Moonshot AI’s US$50 billion pre-IPO round plus business-registration change (October 6, 2026); and Sina/Phoenix Tech’s documentation of DeepSeek Harness’s experimental Claude Code Mods compatibility layer plus Cui Tianyi’s October 4, 2026 statement (October 4, 2026 - October 8, 2026) — were not covered in any of the prior three briefings (October 7, October 8, October 9). The October 7 briefing covered CNBC’s DeepSeek-15-billion-dollar-funding-round-consideration documentation, Reuters’s DeepSeek-12-billion-funding-round documentation, Euronews’s Moonshot-50-billion-valuation-Hong-Kong-IPO-eyes documentation, the South China Morning Post HKIC-BAAI-global-AI-talent-partnership documentation, and ChinaTalk’s China-AI-safety-money-problem documentation. The October 8 briefing covered The Diplomat’s October 2026 essay on China’s AI-governance push and the race to shape global AI rules, Forkast’s The Deployment State essay on China’s state-directed AI-agent industrial-policy apparatus, and Crypto Briefing’s DeepSeek-plus-Huawei-Ascend-950DT-plus-Ulanqab-160000-chip-1-gigawatt-plus-Ascend-960DT-launch-acceleration documentation. The October 9 briefing covered the OpenHarmony community’s AI-tool-use rulebook, the 国产信创外传’s Xinchuang-blind-replacement-slowdown-and-value-based-rollout-reframe, and the 自主可控新鲜事’s selective-headline-framing-plus-individual-vendor-vs-national-industry-dichotomy layer. None of the prior three briefings covered the Bloomberg-October-6-approaching-80-billion-yuan-DeepSeek-pre-IPO-round-plus-CATL-Tencent-dual-lead layer, the Hongxing-October-6-Moonshot-50-billion-dollar-pre-IPO-round-plus-business-registration-change layer, or the DeepSeek Harness experimental Claude Code Mods compatibility-layer-plus-Cui-Tianyi-everything-is-a-plugin-subset-testing-experiment layer.

WeChat input note: The raw/2026-10-09.md file exists in the china-daily-english-input repository but contains only section headers with no article content (all 16 monitored WeChat public-account sections are empty), so the WeChat section is skipped silently for this cycle. The raw/2026-10-08.md file was already processed in the October 9 briefing (OpenHarmony AI-tool-use rulebook plus Xinchuang slowdown plus 自主可控新鲜事 selective-headline-framing). No new WeChat-channel articles are being added in this cycle’s briefing.