单词学习：![[2026032243550536.mp3]]

新闻播报：![[2026032212895267.mp3]]

OpenAI推进首款自研AI芯片，三星确认供应HBM4内存

简报：

• 三星电子计划向 OpenAI 供应下一代高带宽内存（HBM4）芯片，支持其“星门项目”（Stargate），以摆脱对英伟达的算力依赖。
• OpenAI 首款定制 AI 芯片由博通（Broadcom）参与设计并由台积电代工，预计今年第三季度投产，目标于2026年底正式推出。
• 根据排期，三星计划在今年下半年向 OpenAI 供应最高达8亿Gb的12层 HBM4 芯片；此外，三星还与 AMD 签署了备忘录，将为其即将推出的 AI GPU 供应 HBM4。
• 伴随硬件与业务扩张，OpenAI 计划在2026年底前将员工规模从4500人提升至8000人，新增人员集中在产品开发、工程、研究、销售及“技术大使”等岗位。

相关链接：

• 全球最强 AI 算力呼之欲出！三星截胡英伟达，将为 OpenAI 自研芯片供应 HBM4
• 消息称 OpenAI 加速扩充员工规模，目标年底前从 4500 人增员至 8000 人

OpenAI Advances Its First In-House AI Chip, Samsung Confirms HBM4 Memory Supply 🚀

Brief:

• Samsung Electronics plans to supply OpenAI with next-generation high-bandwidth memory (HBM4) chips to support its “Stargate Project,” helping reduce its reliance on Nvidia for computing power.
• OpenAI’s first custom AI chip is being designed with Broadcom’s involvement and manufactured by TSMC. Mass production is expected to begin in the third quarter of this year, with an official launch targeted for the end of 2026.
• According to the schedule, Samsung plans to supply OpenAI with up to 800 million Gb of 12-layer HBM4 chips in the second half of this year. In addition, Samsung has signed a memorandum of understanding with AMD to supply HBM4 for its upcoming AI GPUs.
• Alongside its hardware and business expansion, OpenAI plans to grow its workforce from 4,500 to 8,000 employees by the end of 2026, with new hires focused on product development, engineering, research, sales, and “technical ambassador” roles. 📈🤖

Related links:

• The world’s most powerful AI computing capacity is on the horizon! Samsung beats Nvidia to supply HBM4 for OpenAI’s in-house chips
• OpenAI is reportedly accelerating hiring, aiming to expand from 4,500 to 8,000 employees by year-end

马斯克将推 Grok Computer 智能体，xAI 开启激进上门部署策略

简报：

• 马斯克确认 xAI 即将推出“Grok Computer”智能体，作为特斯拉与 xAI 联合开发的“巨硬（Macrohard）”项目尝鲜版。
• 该智能体将指挥“数字擎天柱”执行电脑端实时操作，主攻企业自动化、处理海量重复性任务及模拟公司运作。
• 为与 OpenAI 和 Anthropic 等争夺企业客户，xAI 采取直接派工程师上门部署的激进策略加速商业落地。
• 上门部署策略已见成效，支付公司 Shift4 在引入 xAI 团队后，决定逐步停用 ChatGPT，转而在日常运营和客服自动化中采用 Grok。

相关链接：

• 为“巨硬”铺路：马斯克宣布 Grok Computer 智能体即将上线
• 马斯克 xAI 为抢夺客户祭出激进打法：直接派工程师上门部署

请问这份简报是否符合您的排版预期？是否还需要为您处理其他新闻资料？

Musk to Launch the Grok Computer Agent, as xAI Adopts an Aggressive On-Site Deployment Strategy 🚀

Brief:

• Musk has confirmed that xAI will soon launch the “Grok Computer” agent, serving as an early version of the “Macrohard” project jointly developed by Tesla and xAI.
• The agent will direct “digital Optimus” to perform real-time computer operations, focusing on enterprise automation, handling massive volumes of repetitive tasks, and simulating company operations.
• To compete for enterprise clients with OpenAI, Anthropic, and others, xAI is accelerating commercial adoption through an aggressive strategy of sending engineers directly on-site for deployment.
• The on-site deployment strategy is already showing results: after bringing in the xAI team, payment company Shift4 decided to gradually phase out ChatGPT and instead adopt Grok for daily operations and customer service automation 🤖.

Related links:

• Paving the Way for “Macrohard”: Musk Announces Grok Computer Agent Coming Soon
• Musk’s xAI Uses an Aggressive Tactic to Win Customers: Sending Engineers Directly On-Site for Deployment

Does this briefing meet your formatting expectations? Would you also like me to help process any other news materials? ✨

谷歌内测系统级Mac版Gemini助理，同时收紧Pro模型免费调用权限

简报：

• 谷歌正为Mac开发原生的Gemini人工智能应用，目前已进入内部测试阶段，旨在与苹果系统深度协同以反击桌面端竞争。
• Mac版Gemini具备极高的系统级权限，未来有望直接调取用户的日历行程、提醒事项及相册，并支持拖拽方式无缝上传本地文件。
• 受益于近期苹果与谷歌达成的AI合作，该原生应用在响应速度与能耗管理上将显著优于传统的浏览器端交互。
• 此外，谷歌宣布将于2026年3月25日生效新规，免费用户将失去Gemini CLI中Gemini Pro模型的访问权，未来仅能使用轻量级的Gemini Flash模型。
• 为打击开发者利用漏洞“零成本”调用AI算力的滥用行为，继续使用Pro及以上级别模型需订阅Google AI付费计划，月费19.99美元起。

相关链接：

• 正面刚 ChatGPT！谷歌原生版 Gemini 杀入 Mac：深度集成苹果系统生态
• 薅羊毛薅秃了？Google 宣布封杀 Gemini 免费福利：Pro 模型不再随意用

Google Internally Tests a System-Level Gemini Assistant for Mac, While Tightening Free Access to Pro Models

Brief:

• Google is developing a native Gemini AI app for Mac, which has now entered internal testing. The move is aimed at deeply integrating with Apple’s system ecosystem and countering competition on the desktop front. 💻
• The Mac version of Gemini is expected to have extensive system-level permissions, with the potential to directly access users’ calendar events, reminders, and photo libraries, while also supporting seamless local file uploads via drag-and-drop.
• Benefiting from the recent AI partnership between Apple and Google, the native app is expected to deliver significantly better response speeds and power efficiency than traditional browser-based interactions. ⚡
• In addition, Google has announced a new policy that will take effect on March 25, 2026: free users will lose access to the Gemini Pro model in Gemini CLI and will only be able to use the lightweight Gemini Flash model going forward.
• To curb abuse by developers exploiting loopholes for “zero-cost” access to AI computing power, continued use of Pro-level models and above will require a subscription to a paid Google AI plan, starting at dJyH6kDkfC5RlM16.answerText9.99 per month. 🔒

Related links:

• Taking on ChatGPT Head-On! Google’s Native Gemini Arrives on Mac with Deep Integration into Apple’s Ecosystem
• Did Users Overdo the Free Perks? Google Announces a Crackdown on Gemini’s Free Benefits: Pro Model Will No Longer Be Freely Available

阿里Qwen3.5-Max登顶全球盲测榜，中国大模型矩阵崛起且调用量反超美国

简报：

• 3月20日LMArena盲测榜单显示，阿里Qwen3.5-Max-Preview以1464分位居前列，超越GPT5.4、Claude4.5及Grok4.1等海外顶尖模型。
• 中国AI力量呈现集群化崛起，全球大模型公司综合实力前十强中中国企业占据五席（包含阿里、字节跳动、智谱AI、月之暗面及百度）。
• OpenRouter最新数据显示，中国AI大模型周调用量（3月9日至15日）达到4.69万亿Token，已连续两周超越美国，并包揽全球调用量排名前三。
• 阿里通义千问全端月活跃用户已突破3亿，平头哥自研GPU芯片实现规模化量产交付。
• 阿里集团CEO吴泳铭提出战略目标，未来五年内包含MaaS在内的云和AI商业化年收入将冲击1000亿美元。

相关链接：

• Qwen3.5-Max预览版登顶LMArena，国产大模型矩阵集体跻身全球前列
• 中国 AI 大模型周调用量达 4.69 万亿 Token，连续两周超越美国
• 阿里五年新目标：云与 AI 商业收入冲刺千亿美元，吴泳铭亮出全栈家底

Alibaba’s Qwen3.5-Max Tops the Global Blind Benchmark, as China’s Large-Model Ecosystem Rises and Surpasses the U.S. in Usage 🚀

Briefing:

• On March 20, the LMArena blind benchmark leaderboard showed that Alibaba’s Qwen3.5-Max-Preview ranked among the very top with a score of 1464, surpassing leading overseas models such as GPT5.4, Claude4.5, and Grok4.1.
• China’s AI strength is showing a clustered rise. Among the world’s top 10 large-model companies by overall capability, Chinese firms account for five spots, including Alibaba, ByteDance, Zhipu AI, Moonshot AI, and Baidu.
• According to the latest data from OpenRouter, Chinese AI large models recorded a weekly usage volume of 4.69 trillion tokens from March 9 to 15, surpassing the United States for two consecutive weeks and taking the top three spots globally by usage volume 📈.
• Alibaba’s Tongyi Qianwen has surpassed 300 million monthly active users across all platforms, while T-Head’s self-developed GPU chips have achieved large-scale mass production and delivery.
• Alibaba Group CEO Eddie Wu has set a strategic goal for the next five years: annual revenue from cloud and AI commercialization, including MaaS, will target dJyH6kDkfC5RlM16.answerText00 billion.

Related links:

• Qwen3.5-Max Preview Tops LMArena, as China’s Large-Model Lineup Collectively Enters the Global Front Ranks
• China’s Weekly AI Large-Model Usage Reaches 4.69 Trillion Tokens, Surpassing the U.S. for Two Consecutive Weeks
• Alibaba’s New Five-Year Goal: Cloud and AI Commercial Revenue to Sprint Toward dJyH6kDkfC5RlM16.answerText00 Billion, as Eddie Wu Showcases Its Full-Stack Strength 🌐

阿里明确全栈AI战略：云与人工智能业务剑指五年千亿美元营收

简报：

• 阿里集团CEO吴泳铭提出最新战略“军令状”，未来五年内包含MaaS（模型即服务）的云和AI商业化年收入目标突破1000亿美元。
• 阿里云业绩呈现加速态势，本季度收入同比增长36%，AI相关产品收入连续十个季度实现三位数增长，经调整EBITA达39亿元。
• 阿里首次披露全栈AI布局：底层平头哥自研GPU芯片已实现规模化量产交付，应用层通义千问全端月活跃用户正式突破3亿。
• 阿里千问系列旗舰预览版模型Qwen3.5-Max-Preview表现强劲，以1464分在LMArena权威盲测中实现对海外顶尖模型的性能超越。
• 中国AI生态整体呈爆发趋势，3月9日至15日中国AI大模型周调用量达4.69万亿Token，已连续两周超越美国。

相关链接：

• Qwen3.5-Max预览版登顶LMArena，国产大模型矩阵集体跻身全球前列
• 中国 AI 大模型周调用量达 4.69 万亿 Token，连续两周超越美国
• 阿里五年新目标：云与 AI 商业收入冲刺千亿美元，吴泳铭亮出全栈家底

Alibaba Clarifies Its Full-Stack AI Strategy: Cloud and AI Business Targets dJyH6kDkfC5RlM16.answerText00 Billion in Revenue Within Five Years 🚀

Brief:

• Alibaba Group CEO Eddie Wu has issued a new strategic “mission order,” setting a goal for the company’s cloud and AI commercialization business—including MaaS (Model as a Service)—to exceed dJyH6kDkfC5RlM16.answerText00 billion in annual revenue within the next five years.
• Alibaba Cloud’s performance is showing clear acceleration. Revenue for this quarter rose 36% year over year, revenue from AI-related products achieved triple-digit growth for the tenth consecutive quarter, and adjusted EBITA reached RMB 3.9 billion.
• Alibaba has, for the first time, disclosed its full-stack AI布局: at the foundational layer, T-Head’s self-developed GPU chips have entered scaled mass production and delivery; at the application layer, Tongyi Qianwen’s total monthly active users across all platforms have officially surpassed 300 million.
• Alibaba’s Qwen flagship preview model, Qwen3.5-Max-Preview, has delivered strong results, scoring 1464 points on the authoritative blind benchmark LMArena and outperforming leading overseas models.
• China’s AI ecosystem as a whole is entering an explosive growth phase 📈. From March 9 to March 15, weekly API calls to China’s large AI models reached 4.69 trillion tokens, surpassing the United States for the second consecutive week.

Related Links:

• Qwen3.5-Max Preview Tops LMArena as Chinese Large-Model Lineup Collectively Rises into the Global Top Tier
• China’s Weekly AI Large-Model Calls Reach 4.69 Trillion Tokens, Surpassing the U.S. for Two Consecutive Weeks
• Alibaba’s New Five-Year Goal: Cloud and AI Commercial Revenue to Sprint Toward dJyH6kDkfC5RlM16.answerText00 Billion, Eddie Wu Reveals Full-Stack Strength 🤖