Focusing on the emerging trends of industry large models, deeply analyzing the implementation and practice of all-domain artificial intelligence in government and enterprise scenarios
Damao Technology and Pudong Environmental Energy will build a compute-grid coordination platform for the Dawn center — northern Pudong's first 10,000-GPU cluster with 18 MW designed capacity.
Google released TPU 8tx and 8i Max on Nov 9 for confidential AI computing and extra-long-context inference — the fourth 8th-gen system this year, sharpening its challenge to NVIDIA.
Anthropic's new flagships saw 1,000 agents formalize Fermat's Last Theorem in 30 hours, while API pricing drops to 8/40 dollars per million tokens ahead of its IPO.
Chongqing's state assets regulator published 52 AI opportunity scenes with planned investment over RMB 72 million plus 34 capability scenes, letting SOE demand and AI supply meet on one list.
EY launched its CaiKui YuanShu finance AI platform at the 9th CIIE, combining 15 AI agents, a three-layer capability architecture and one orchestration hub covering accounting, analytics and business insight.
USTC's open RNA model, Jensen Huang's $2B Europe investment, Wenzhou AI healthcare, Shenyang's zero-form service, AI agents running an 80k-star project, and more.
Yuanjie, a Jingyuan Environmental subsidiary, opened a 1,300 sqm chip-level GPU repair workshop in Nantong with X-ray inspection and BGA rework, serving Yangtze River Delta compute customers.
At the 9th CIIE, Novo Nordisk and JD Health upgraded their partnership, aiming AI and big data at disease education, patient services and chronic care across online and offline scenarios.
OpenRouter logged 210 trillion tokens for Nov 2-8; Chinese models took 116.18 trillion, up 16.71% and ahead of the US for a 28th week, with DeepSeek No.1 and Hunyuan Hy4 debuting fourth.
Guangdong's 12345 hotline reform made NDRC's 2026 best-practice list; Shenzhen handles 20M+ requests a year, AI dispatches 98% in seconds, and 18,000 issues were resolved before citizens complained.
Reuters reported 600,000 advanced Nvidia chips sit in Hong Kong pending mainland import approval; the foreign ministry said its position is consistent, as DeepInfer's open-sourcing accelerates domestic alternatives.
AI news for Nov 8 in 10 items: MiniMax founder on China-US compute paths; Zhang Yaqin and Yao Shunyu debate model moats; Topsstar bipedal robot; VR conference opens in Guangzhou; Frost & Sullivan compute-chip report; Lingchuan video-generation compute; Hitachi's factory-first physical AI; GBA AI jobs fair; and more.
Blackstone's Jon Gray says AI compute demand outruns supply; Beijing's token factory tops 1.4 trillion tokens a day and a $200M overseas token deal marks China's compute-services export. Analysts see a 260-billion-yuan leasing market in 2026 — with high-end capacity scarce and low-end crowded.
China Mobile's 2027 plan earmarks over 100 billion yuan for compute within 500 billion yuan of procurement, with capacity to double for two straight years; carrier AI revenue is accelerating — China Mobile's H1 AIDC revenue jumped 486.1% — and token operations are now the sector's common monetization path.
OpenAI's new security model GPT-6 Cyber scores 48.5% on ExploitGym, 16 points above its predecessor, and beats flagship GPT-6 Astra on most cybersecurity tasks; it has validated real bugs in Linux and Firefox and rolls out only to vetted Daybreak Red security teams.
Beijing's Chaoyang District Court ruled that an insurance-company counsel fired for using AI at her manager's recommendation should keep her job. Workers' Daily found banks, internet firms and publishers now encourage AI use; experts say employers must set compliance rules before punishing mistakes.
China Mobile and Zhipu (Z.ai) signed a partnership on Nov 6 covering domestic compute adaptation, token production efficiency, AI coding and finance/government applications; GLM models will join China Mobile's model service system for large-scale industry adoption.
E Fund launched a five-expert AI companion team; Changxin, Aegon-Industrial and Southern put agents on WorkBuddy covering 25 scenarios, with source-stamped replies.
Shanghai unveiled an AI + transportation plan built on three technical foundations — models and agents, industry datasets, computing clusters; its port-shipping AI pilot base already manages 54 business agents.
Tesla renames its chips around Super Intelligence and brings back Dojo D3 with chiplets and stacked HBM; SI6 claims 80x SI4 compute with 9x memory, arriving 2029–2030.
Microsoft is reportedly switching off Oracle-leased AI capacity built on older-generation Nvidia chips; IREN's capacity supporting over $4 billion annualized revenue slipped to January, and CoreWeave shares slid.
The first ~$11 billion tranche of OpenAI's new round landed Nov 6 at a $1.25 trillion valuation, below its original $1.4 trillion target. SoftBank, Sequoia, a16z and a UAE consortium joined; the CFO says data-center investment continues.
Anthropic detailed Claude's agent memory system: offline dreaming between sessions fixes errors and consolidates experience. Benchmarks show note-taking models close the learning loop, and an enterprise wiki lifts pass rates on a 420M-token corpus from 13% to 48%.
At the Nov 4 Wuzhen Summit, CAICT joined Tencent, Ant, Alibaba, Douyin and 20 ecosystem organizations to release the Agent Development Self-Discipline Pact — eight rules covering trust systems, interconnection infrastructure, interoperability, security and user rights. Verifiable agent identity, this commentary argues, is what turns agents into trusted digital employees.
