Daily AINo. 40第 40 期
Alibaba and Huawei say the race is the whole system, not the chip
Washington lifted its freeze on UK pre-release model testing, and China's two biggest AI hardware players both pitched full systems instead of just faster chips.
UK testers get frontier models back after a 17-day freeze
The US reversed its Sept 24 order that told OpenAI and Anthropic to stop sharing unreleased models with Britain's AI Security Institute. Access to Anthropic, OpenAI and Google DeepMind models is restored as of Oct 11, though in the meantime Anthropic kept Claude Mythos 5.1 limited to US organizations.
Alibaba's T-Head unveils the Zhenwu V900
At the Apsara Conference in Hangzhou, Alibaba's chip arm launched a training-and-inference accelerator it says is three times faster than the Zhenwu M890, plus a supernode server built around it. Mass production and sales are expected in Q1 2027.
Huawei bets on optics with the Ascend 960 SuperPoD
The system links up to 4,096 NPU cards using near-packaged optics. Huawei says about 5,500 of its Hi-ONE optical engines replace 48,000 800G optical modules and cut power by more than 550 kW, which shows the fight is moving to interconnects and system design.
Italy finishes wiring up AI Act enforcement
A second implementing law, published Oct 8, names the authorities that will inspect and fine, sets the fine levels, opens a national regulatory sandbox, and restricts AI use in decisions about employees. That puts Italy among the first EU countries with a complete enforcement setup.
SourcesGaming Tech Law, Oct 2026
Every morning Andrew, the news-desk agent on my team, files this brief on what changed in AI. Every story links to its sources. How my AI team works
阿里和华为同一天出牌，真正比的已经不是单颗芯片，而是整套系统
今天最值得看的两件事：美国解除了对英国“提前测模型”的封锁；阿里和华为同一天发AI硬件，而且都在讲“系统”，不只是芯片。
英国又能提前测前沿模型了，封锁只持续了17天
9月24日，白宫让OpenAI和Anthropic暂停向英国AI安全研究所（AISI）提供未发布模型。10月11日这个决定被撤回，Anthropic、OpenAI、Google DeepMind的预发布模型重新对英国开放。中间这段时间，Anthropic的Claude Mythos 5.1只给美国机构用。一句话：AI安全合作，说到底还是看政治风向。
阿里平头哥发布真武V900
在杭州云栖大会上，平头哥发布了新一代训推一体AI芯片真武V900，官方说性能是上一代真武M890的3倍，还配了一整套超节点服务器。预计2027年一季度量产开售。
华为昇腾960超节点：把光模块“塞”到芯片旁边
最多支持4096张NPU卡，用的是近封装光学（NPO）。华为说，大约5500个Hi-ONE光引擎就能替代4.8万个800G光模块，省电超过550千瓦。重点不是又快了多少，而是卡和卡之间怎么连。
意大利的AI法执法体系正式到位
10月8日公布的第二部配套法律，明确了谁来检查、谁来罚、罚多少，开了国家级监管沙盒，还限制企业用AI做和员工相关的决定。意大利成了欧盟最早把AI法“落地执法”的国家之一。做出海、在欧洲招人的团队要注意了。
每天早上，我团队里负责新闻的智能体 Andrew 整理当天 AI 圈的变化，每条都附原始来源。 看看我的 AI 团队怎么运作