晚安咖啡GoodNightCoffee

晚安咖啡
  • 5.0 (1)
  • Leisure
  • Daily

"晚安咖啡" 一杯陪你从清晨到深夜的神奇咖啡。 早晨，来杯清醒提神的商业浓缩， 读懂全球科技脉搏， 创业风云、商业变革， 让你的早安问候都带着点儿专业范。 午间，来杯香醇的娱乐拿铁， 影视剧追踪，明星动态， 让你的话题库永远新鲜， 办公室八卦，从此不用怕。 夜晚，来杯温暖特调， 治愈故事，生活感悟， 让一天的疲惫， 在耳机里慢慢融化。 每一杯咖啡的温度， 都能让你的碎片时间闪闪发光。 洗漱、通勤、午休、健身， 每个平凡时刻， 都能遇见不平凡的分享。 这是一杯， 让你在生活中保持清醒， 但不会让你失眠的咖啡。 你的随身资讯管家， 每天都为你精心冲泡， 一杯又一杯， 不见不散。

  1. 1h ago

    《晚安咖啡》科技夜话：DeepSeek 千亿融资落地，宝马推城市智驾，微软语音模型降价

    嗨~~大家晚上好呀！今天是2026年11月3日，星期二，欢迎收听晚安咖啡，我是主播安安。今晚咱们一起聊聊新鲜热乎的商业科技动态，轻松又有料，快搬好小板凳，我们马上开始啦！ @@@ 据The Verge报道，宝马计划从2027年起通过软件更新在部分车型推出导航引导的城市驾驶辅助，率先在德国本土市场上线，后续逐步拓展至欧洲其他国家及全球更多市场。这套系统定位于L2级驾驶辅助，可同时覆盖高速与城市道路场景：依托车辆内置的高精度GPS与环境感知硬件，实现自动转向、智能加减速，在城市路况下还能协助驾驶员完成变道、路口转弯、礼让行人和车辆，以及安全通过环岛等操作。 按照设计规范，驾驶员全程仍需保持对道路的关注并随时准备接管车辆；高速场景下允许短时脱手驾驶，而城市等地方道路则要求驾驶员手握方向盘。系统采用人机协同的逻辑，驾驶员随时主动介入转向、刹车或加速操作，都不会触发功能关闭，驾驶体验更连贯自然。目前宝马3系、7系、X5、iX3和iX4等车型已出厂预装所需的摄像头与传感器硬件，车主后续可直接通过OTA远程升级激活功能，无需到店额外加装。 这套系统的落地也遵循了欧洲最新的道路安全法规，该法规首次允许车辆在城市道路场景下主动提出并执行自动化驾驶动作。宝马方面表示，已有外部专家团队完成了大量真实道路测试，确认系统达到既定技术标准，最终的政府监管批准预计在2027年初落地。行业普遍认为，此举是宝马对标特斯拉FSD城市辅助功能的核心布局；与此同时，安全研究者也表达了普遍担忧：长期使用驾驶辅助可能导致驾驶员注意力下降、产生技术依赖，且已有多项研究数据显示，现阶段多数驾驶辅助系统的综合安全表现，仍不及专注驾驶的人类驾驶员。 @@@ 据VentureBeat报道，微软AI于周四正式发布全新语音识别模型MAI-Transcribe-2，官方称该模型在识别速度、准确率与使用成本上，均优于OpenAI、谷歌和ElevenLabs的同类现有产品，定价低至每音频小时10美分。 就在五个月前，该系列首款模型的收费还为每小时0.36美元，此次早鸟价降幅约72%。以企业场景测算，若一家企业每年处理10万小时的呼叫中心音频，使用新模型后年成本将从3.6万美元直接降至1万美元，降本效应十分显著。 能力层面，新模型支持的语言数量提升至60种，相较于6月的43种、4月的25种实现了快速扩容，并且专门针对背景噪声、低质量录音、多人语音重叠等企业真实场景做了深度优化。基础功能已集成语者分离、词级时间戳、关键词偏置和自动语言识别；同时提供两种输出模式：逐字模式会完整保留语气词、口误、错误开头和口吃等细节，满足法务合规等场景的精准转录需求；干净模式则自动去除填充词与冗余语气，直接输出通顺文稿，适配会议纪要、内容整理等日常场景。针对多语言混用场景，模型支持代码切换能力，可流畅处理同一句话里的语言切换，微软专门举例了印地语+英语的Hinglish、西班牙语+英语的Spanglish等典型混合语场景。 报道同时指出，微软正按模态逐步自研前沿AI模型，替换此前产品中使用的OpenAI技术，而语音转录是目前进展最快、落地最成熟的领域。 @@@ 据Readhub报道，国内大模型厂商DeepSeek已完成规模达1055亿元的第二轮融资，投后估值达到6155亿元，本轮共有近30家投资方参与，内容社区小红书也出现在投资方名单中。事实上小红书近年一直在持续加码AI领域布局，此前已先后参投了月之暗面Kimi和MiniMax等大模型公司。 本轮融资的投资方结构也备受关注：国资占比升至27%，腾讯与宁德时代为最大的两家投资方，同时覆盖了芯片、制造、车企、互联网等多个领域的产业资本，以及多家市场化基金，体现出全行业对AI大模型赛道的长期看好。 值得注意的是，本轮融资设置了特殊的投资门槛：除国家人工智能产业投资基金外，其余外部投资者的资金都需要通过由梁文锋管理的有限合伙企业进入，同时外部投资者需放弃投票权，并承诺股份锁定五年。这一设计既保障了创始团队对公司的控制权与长期发展节奏，也侧面反映出AI头部标的的稀缺性——各方争相入局，核心都是为了抢占AI时代的未来发展机遇，绑定长期技术红利。 @@@ 新浪科技消息，Anthropic于11月2日美股盘前更新了S-1/A招股文件，披露了2026年第三季度的核心财务数据。截至2026年9月30日的第三季度，公司总营收达169.6亿美元，去年同期仅为13.8亿美元，同比大幅增长1127%；其中旗下Claude大模型业务的营收规模已达到170亿美元。 高速增长的同时，亏损也在同步扩大：同期净亏损233亿美元，上年同期净亏损为121亿美元。前三季度累计来看，公司总营收332.4亿美元，同比增长约1200%；净亏损1040亿美元，而2025年同期净亏损为251亿美元。 盈利层面，第三季度公司调整后营业利润为25亿美元，调整后营业利润率达15%，展现出不错的商业化健康度；但按通用会计口径计算，当期营业亏损仍有5.67亿美元，差额主要来自股权激励、算力折旧等非现金支出。 文件同时披露了多项运营数据：截至9月30日及此后，公司在算力采购承诺及租赁协议项下的承诺支出总额约8954亿美元，彰显了长期的算力投入规划；第三季度年化收入不低于10万美元的企业客户超过11900家，截至9月30日的合同义务总额达3200亿美元，企业端付费能力与长期订单储备充足。此外公司也在文件中明确承诺，Claude产品体系内不会投放广告，坚持以订阅与企业服务为核心的商业化路径。 @@@ 据晚点LatePost独家报道，字节跳动旗下的火山引擎已将业务重心全面转向商业化，其核心的模型即服务（MaaS）业务在2026年上半年毛利率超过55%。这一毛利率水平在国内云服务与AI大模型服务赛道处于前列，也意味着其MaaS业务已经跑通了商业化模型，从技术投入期正式进入规模化盈利阶段。依托字节跳动内部的技术沉淀与大规模落地验证，火山引擎在企业级AI服务市场的竞争力正持续凸显。 @@@ 据虎嗅24小时消息，马斯克将与台积电重新展开谈判。市场普遍推测，本次谈判大概率围绕先进制程芯片的产能分配展开，可能涉及特斯拉自研智能驾驶芯片，或是马斯克旗下xAI所需的高端算力芯片。台积电作为全球先进制程芯片的核心供应商，其产能一直是科技巨头争夺的焦点，此次重启谈判也侧面反映出，当前AI与智能汽车赛道对高端芯片的需求仍处于高位，产能紧张的局面尚未得到明显缓解。 @@@ 好啦，今天的节目到这里就结束啦。如果您喜欢我们的节目，也非常欢迎订阅和转发给你的小伙伴，这对我们非常有帮助。那我们下期再见吧！ 在小宇宙查看该单集文稿

  2. 1h ago

    《晚安咖啡》科技夜话（英文版）：DeepSeek 千亿融资落地，宝马推城市智驾，微软语音模型降价

    Hi there, good evening everyone! Today is Tuesday, November 3rd, 2026. Welcome to Good Night Coffee, I'm your host An An. Tonight we're catching up on the latest business and tech updates – light, engaging and full of insights. Grab a seat and let's dive right in! @@@ According to The Verge, BMW plans to roll out a navigation-guided urban driver assistance feature on select models via software updates starting in 2027. It will launch first in Germany, followed by other markets around the world. Classified as Level 2 (L2), the system assists with steering, acceleration and deceleration on both highways and urban roads using built-in GPS. In city scenarios, it also supports lane changes, turns, yielding and roundabout navigation. Drivers must remain attentive to the road and be ready to take control at all times: hands-off driving is permitted on highways, while drivers are required to keep hands on the wheel on local roads. Built for human-machine collaboration, the system will not deactivate if the driver manually steers, brakes or accelerates at any point. Models including the 3 Series, 7 Series, X5, iX3 and iX4 already come factory-equipped with the required cameras and sensors, meaning owners can activate the feature via over-the-air (OTA) updates without additional hardware installation. The system complies with new European safety regulations that, for the first time, allow vehicles to initiate and execute automated driving maneuvers on urban roads. BMW states that external experts have completed extensive real-world road testing and confirmed the system meets technical standards, with final regulatory approval expected in early 2027. The move is widely viewed as BMW's direct response to Tesla's Full Self-Driving (FSD). Safety researchers, however, have raised concerns about reduced driver attention and over-reliance on the technology, noting that evidence shows most current driver assistance systems are less safe than a fully focused human driver. @@@ According to VentureBeat, Microsoft AI launched its new speech recognition model MAI-Transcribe-2 on Thursday, claiming it outperforms existing offerings from OpenAI, Google and ElevenLabs in speed, accuracy and cost, priced at just 10 US cents per audio hour. Five months ago, the first model in the MAI-Transcribe line cost $0.36 per hour. This early-bird price marks a roughly 72% reduction. For an enterprise processing 100,000 hours of call center audio per year, annual costs would drop from $36,000 to $10,000, delivering significant savings. The new model supports 60 languages – up from 43 in June and 25 in April – and is optimized for real-world enterprise audio conditions including background noise, low-quality recordings and overlapping speech. Core built-in features include speaker diarization, word-level timestamps, keyword biasing and automatic language identification. It offers two output modes: a verbatim mode that preserves filler words, false starts and stutters for legal and compliance use cases, and a clean mode that strips out filler language for polished transcripts. It also handles code-switching seamlessly for mid-sentence language shifts, with Microsoft specifically highlighting Hinglish (Hindi-English) and Spanglish (Spanish-English) as common mixed-language use cases. Reports note that Microsoft is building state-of-the-art in-house AI models across modalities to replace OpenAI technology previously used in its products, with speech transcription being the fastest-moving and most mature area to date. @@@ According to Readhub, Chinese large model developer DeepSeek has closed its second financing round totaling 105.5 billion yuan, bringing its post-money valuation to 615.5 billion yuan. Nearly 30 investors participated in the round, including content and lifestyle platform Xiaohongshu. Xiaohongshu has been steadily ramping up its AI investments in recent years, having previously backed Moonshot AI (developer of the Kimi assistant) and MiniMax. State-backed capital accounts for 27% of this financing round. Tencent and CATL are the two largest investors, with additional participation from industrial capital and market funds across chips, manufacturing, automotive, internet and other sectors. Notably, the round comes with special investment terms: with the exception of the National Artificial Intelligence Industry Investment Fund, all other external investors must contribute capital through a limited partnership managed by Liang Wenfeng, surrender voting rights, and agree to a five-year share lock-up period. The intense investor interest reflects a broader race to secure long-term opportunities in the AI era. @@@ Sina Tech reports that Anthropic filed an updated S-1/A prospectus before US market open on November 2nd, disclosing its third quarter 2026 financial results. For the quarter ended September 30, 2026, the company generated total revenue of $16.96 billion, up 1,127% year-over-year from $1.38 billion in the same period last year. Revenue from its Claude model business reached $17 billion. Net loss for the quarter stood at $23.3 billion, compared with $12.1 billion a year earlier. For the first three quarters combined, total revenue hit $33.24 billion, up roughly 1,200% year-over-year, while net loss reached $104 billion, up from $25.1 billion in the first nine months of 2025. On an adjusted basis, Q3 operating profit came in at $2.5 billion with an adjusted operating margin of 15%, signaling healthy commercialization momentum. On a GAAP accounting basis, however, operating loss was $567 million, with the gap driven largely by non-cash items such as stock-based compensation and computing infrastructure depreciation. The filing also revealed key operational metrics: total committed spending under procurement and lease agreements amounted to approximately $895.4 billion as of September 30 and beyond, reflecting the company's massive long-term compute investment plans. It had over 11,900 enterprise customers with annualized revenue of at least $100,000 in Q3, and total contract obligations reached $320 billion as of September 30, pointing to strong enterprise demand and a robust order backlog. The company also explicitly pledged that Claude will never carry advertising, sticking to a subscription and enterprise services monetization model. @@@ According to an exclusive report by LatePost, Volcano Engine, ByteDance's enterprise cloud and AI arm, has fully shifted its business focus to commercialization. Its core Model-as-a-Service (MaaS) business achieved a gross margin of over 55% in the first half of 2026, putting it among the top tier in China's cloud and AI service sector and marking its transition from technology investment to scaled profitability. @@@ According to Huxiu 24/7, Elon Musk is set to restart negotiations with TSMC. Market observers widely expect the talks to center on advanced process chip capacity allocation, potentially covering Tesla's in-house autonomous driving chips or high-end compute chips for Musk's xAI. As the world's leading advanced node chipmaker, TSMC's capacity has long been contested by major tech firms. The renewed negotiations also underscore that demand for high-end chips remains intense across AI and smart vehicle sectors, with supply tightness yet to ease significantly. @@@ Alright, that wraps up today's show. If you enjoyed it, feel free to subscribe and share it with your friends – it really helps us out. See you next time! 在小宇宙查看该单集文稿

  3. 1d ago

    《晚安咖啡》科技夜话（英文版）：巨头AI筹资，OpenAI阿布扎比签约，美团LongCat-2.5发布

    Hi~~ Good evening, everyone! Today is Monday, November 2, 2026, and welcome to Goodnight Coffee. I’m your host, An’an. Today we’re going to chat about the latest and hottest business and tech stories—relaxed, fun, and you’ll pick up a thing or two. So come along with me! @@@ According to Readhub, with AI investment running high, tech giants are taking diverging paths to raise capital. Meta launched a new round of equity financing on October 26, increasing the size from $35 billion to $37.5 billion. It also plans a share sale program of up to $60 billion, bringing the total to as much as $97.5 billion, to invest in AI infrastructure and personal superintelligence. At the same time, Meta issued non-voting Class C shares, paused shareholder buybacks, and raised the lower end of its full-year capital expenditure forecast. Its third-quarter report showed quarterly revenue of $64.601 billion, up 26% year on year. On October 22, Meta’s stock fell about 7%, with investors worried about the timeline for AI returns and dilution from additional Class C shares. Oracle announced on October 28 that it is adding up to $20 billion in capital markets financing for this investment cycle through fiscal 2030, calling it the “last tranche of funding” to complete the remainder. It plans to turn free cash flow positive in the second half of fiscal 2028 and to rely on customer prepayments and supplier financing to share upfront investment. Oracle’s stock closed up 10.73% that day, then fell 3.45% and 6.37% over the following two trading days. Amazon is maintaining its full-year cash capital expenditure plan of about $220 billion, while Alphabet is maintaining $195 billion to $205 billion and expects a clear increase in investment in 2027. @@@ Sina Tech reports that, according to a statement from the Abu Dhabi Media Office, OpenAI CEO Sam Altman met with Crown Prince Khaled in Abu Dhabi on Monday and witnessed the signing of four agreements between relevant institutions and OpenAI, covering government, energy, and investment. In energy, Abu Dhabi National Oil Company signed with Inception42 and OpenAI, becoming the first energy company to deploy OpenAI’s frontier technology. ADNOC said OpenAI agents will predict crude oil supply and demand, prices, and margins, and anticipate impacts in advance. In investment, Mubadala and Inception42 will work with OpenAI to apply advanced AI to investment analysis, opportunity identification, and risk assessment. In government, the two sides launched the world’s first AI-native “machine-readable government” initiative, with the first pilot, focused on food imports, now in production with the Abu Dhabi Agriculture and Food Safety Authority. Inception42 also signed a five-year agreement with OpenAI to promote the application of frontier AI in government, enterprise, and education. OpenAI is seeking billions of dollars in financing from UAE state-backed investors. Earlier reports said it was in talks with several UAE funds, including MGX, and hopes MGX will lead a $30 billion funding round. @@@ According to Huxiu 24 Hours, Samsung plans to shut down all production lines of its 8-inch wafer foundry business by the end of 2030. @@@ cnBeta, citing Reuters, reports that Binance, the world’s largest cryptocurrency exchange, resumed withdrawals on Sunday after a hack. Binance customers then withdrew $3.64 billion, or 2.1% of its assets. Binance said that after the initial withdrawal peak, customer withdrawal volume stabilized, while deposits remained at “substantial levels.” It also said it has recovered 95% of the $6.3 billion in stolen assets. Binance said the net outflow figure reflects customer withdrawals and deposits as of 15:30 GMT. Crypto data analytics platform DefiLlama showed a higher outflow of $4.36 billion. A Binance spokesperson responded that this figure does not include funds already deposited into the company’s new hot wallets, so it is inaccurate; the latest update on X is closest to the actual situation, but deposit and withdrawal changes after that data was published have not yet been included. Binance also said withdrawals will resume in phases, prioritizing user needs. The first phase covers 15 tokens, including Bitcoin, Ether, some USDC- and USDT-related stablecoin tokens, and Dogecoin. @@@ According to Huxiu 24 Hours, Apple has responded to claims that “the Chinese version of iPhone Duo cuts corners.” @@@ Sina Tech reports that Meituan released its large model LongCat-2.5 on November 2. The model is designed for long-horizon agent tasks and is a native multimodal model with 1.6 trillion total parameters; its entire training was completed on domestic compute. LongCat-2.0, released in June this year, was the first to complete training and inference of a trillion-parameter model using domestic compute. LongCat-2.5 converts visual input into uniformly processed tokens and supports tasks such as interactive web pages and 3D display pages. Its agentic coding uses large-scale long-horizon reinforcement learning to improve end-to-end engineering capabilities. Its training sequences are up to 1M tokens, with cumulative training volume reaching trillions of tokens. Meituan worked with domestic compute vendors to jointly optimize memory, inter-card communication, and compute and transfer efficiency, and optimized task allocation around supernodes. Meituan said the model focuses on strengthening business analytics and software development capabilities, equipping local-services merchants with an AI helper. In one case, the model identified a growth-quality issue—payment GMV rose 23.2% while channel profit fell 1.7%—and proposed differentiated resource allocation recommendations. Evaluations show its scores fall between Claude-Opus-4.8 and Claude-Opus-5, and it surpasses the latter on some evaluations. Wang Xing previously said that models and AI products will support the core business and improve user and merchant experience and operating efficiency. @@@ Readhub reports that BYD sold 490,200 new energy vehicles in October, up 10.97% year on year; cumulative sales in the first 10 months reached 3.6217 million units, down 2.16% year on year. In October, battery electric vehicle sales were 305,700 units, up 37.38% year on year; plug-in hybrid electric vehicle sales were 177,400 units, down 17.22% year on year. In the first 10 months, cumulative BEV sales were 1.9357 million units, up 5.86% year on year; cumulative PHEV sales were 1.626 million units, down 11.02% year on year. In addition, BYD exported 170,200 new energy vehicles in October. @@@ Sina Tech reports that Xinlian Power, a controlled subsidiary of United Nova Technology, recently completed a nearly RMB 2 billion Series C round. The round was led by the China Mobile Chain Leader Fund, with joint investment from Inspur Information, Yuanjing Capital, Sinopec Hydrogen Energy Fund, NAURA Nohua Capital, Jiangfeng Electronics, Changzhou Xingyu, Cathay Capital, Puchao Capital, CCB Equity, CITIC Goldstone, CITIC Securities, Galaxy Yuanhui, and others. After the financing, its shareholders span communications, computing power, power grids, new energy, and automotive sectors, with several AI infrastructure industrial investors joining, marking AI server power supplies as a strategic priority for the next stage. Yuan Feng said the company will consolidate its global leadership in automotive silicon carbide while building AI server power supplies into a new growth curve. Xinlian Power focuses on silicon carbide, gallium nitride, and module packaging. Its products cover 650V to 3300V, with above 6.5kV under development. Its 3300V SiC MOSFET, launched in June this year, targets scenarios such as solid-state transformers and has been sampled and validated with core customers. The comprehensive solution can reduce system-level BOM costs by 20% to 35%, filling a key gap in domestic high-voltage SiC devices for core power-stage applications in solid-state transformers. @@@ Sina Tech reports that on October 28, Sunwoda announced that its subsidiary Sunwoda Power, in its C++ financing round, had 5.34% of its shares subscribed by a wholly owned Xiaomi subsidiary for RMB 1.7 billion in cash, making Xiaomi the third-largest shareholder of Sunwoda Power. Its second-largest shareholder, Li Auto, had already invested RMB 2.65 billion in September to subscribe for 8.79%. After Xiaomi’s investment, Li Auto’s stake was diluted to 8.32%, and together with affiliates it exceeds 10%. Sunwoda no longer has absolute control over Sunwoda Power, so it will no longer consolidate it, and both Li Auto and Xiaomi will receive several board seats. Reports say Xiaomi and Sunwoda cooperate in consumer electronics, and Sunwoda’s stacked-cell process capacity is one reason automakers are interested in it. In Li Auto’s supplier lineup, Sunwoda accounts for the largest share, CATL has fallen below 30%, and CALB is just getting started. The next day, Sunwoda terminated its H-share listing plan; the plan may be rearranged, or Sunwoda Power may be listed in Hong Kong. @@@ According to Huxiu 24 Hours, a Unitree robot withdrew from the Mount Tai climbing competition due to overheating and was later carried down the mountain on a stretcher. @@@ Alright, that’s all for today’s show! Thank you for spending time with us. Remember to head out with a good mood. If you enjoy our program, we’d love for you to subscribe and share it with your friends—it really helps us. See you next time! 在小宇宙查看该单集文稿

  4. 1d ago

    《晚安咖啡》科技夜话：巨头AI筹资，OpenAI阿布扎比签约，美团LongCat-2.5发布

    嗨~~大家晚上好呀！今天是2026年11月2日，星期一，欢迎收听晚安咖啡，我是主播安安。今天咱们一起聊聊新鲜热乎的商业科技动态，轻松又涨知识，快跟我出发吧！ @@@ 据Readhub报道，AI投入高企下，科技巨头筹资路径分化。Meta于10月26日启动新一轮股权融资，规模由350亿美元增至375亿美元，另拟设最高600亿美元股票出售计划，合计最高975亿美元，用于AI基础设施与个人超级智能投资；同时发行无投票权C类股，暂停股东回购并上调全年资本开支预期下限。其三季报显示，季度营收646.01亿美元，同比增26%。10月22日Meta股价跌约7%，投资者担忧AI回报时间及C类股增发摊薄。甲骨文10月28日宣布，截至2030财年本轮投资周期新增资本市场融资上限200亿美元，称其为完成余下部分的“最后一笔资金”；计划2028财年下半年自由现金流转正，并靠客户预付款、供应商垫资分担前期投入。当天甲骨文股价收涨10.73%，随后两个交易日分别跌3.45%和6.37%。亚马逊维持约2200亿美元全年现金资本开支计划，Alphabet维持1950亿至2050亿美元，并预计2027年投入明显增加。 @@@ 新浪科技消息，据阿布扎比媒体办公室声明，OpenAI首席执行官萨姆·阿尔特曼周一在阿布扎比与王储哈立德会谈，并见证相关机构与OpenAI签署四项协议，涉及政府、能源和投资。能源领域，阿布扎比国家石油公司与Inception42、OpenAI签约，成为首家部署OpenAI前沿技术的能源企业；ADNOC称，OpenAI智能体将预测原油供需、价格和利润率，并提前预判影响。投资领域，穆巴达拉和Inception42将与OpenAI把先进AI用于投资分析、机会识别和风险评估。政府领域，双方启动全球首个AI原生“机器可读政府”倡议，首个试点与阿布扎比农业和食品安全局围绕食品进口投入生产。Inception42另与OpenAI签署五年协议，推动前沿AI在政府、企业和教育领域应用。OpenAI正寻求阿联酋国有背景投资方提供数十亿美元融资；此前报道称，其正与包括MGX在内的多家阿联酋基金洽谈，希望由MGX牵头一轮300亿美元融资。 @@@ 据虎嗅24小时消息，三星计划在2030年底前关闭其8英寸晶圆代工业务的所有产线。 @@@ cnBeta援引路透社报道，全球最大加密货币交易所币安在遭黑客攻击后于周日恢复提现。币安客户随后提走36.4亿美元，占其资产的2.1%。币安表示，在最初的提现高峰过后，客户提现规模已趋于稳定，存款金额则一直保持在“可观水平”，并称目前已补回被盗63亿美元资产的95%。币安称，净资金流出数据反映的是截至格林尼治标准时间15时30分客户提现和存款的情况。加密货币数据分析平台DefiLlama显示，资金流出规模更高，达到43.6亿美元。币安发言人回应称，这一数字不包含公司已存入自身新热钱包的资金流，因此并不准确；币安在X上的最新更新最接近实际情况，但该数据发布后发生的存款和提现变动尚未计入。币安还表示，提现服务将分阶段恢复，并优先满足用户需求，第一阶段涉及包括比特币、以太币、部分USDC和USDT稳定币相关代币以及狗狗币在内的15种代币。 @@@ 据虎嗅24小时消息，苹果已就“iPhone Duo中国版偷工减料”的说法作出回应。 @@@ 据新浪科技消息，美团11月2日发布大模型LongCat-2.5。该模型面向长程智能体任务，为原生多模态模型，总参数1.6万亿，训练全程由国产算力完成。今年6月发布的LongCat-2.0已首次以国产算力完成万亿参数模型训练和推理。LongCat-2.5将视觉输入转化为统一处理的token，支持交互网页、3D展示页面等任务；智能体编程采用大规模长程强化学习，提升端到端工程能力。其训练序列最长1M token，累计训练量达数万亿token，美团与国产算力厂商协同优化显存、卡间通信和计算传输效率，并围绕超节点优化任务分工。美团称，该模型重点强化商业分析和软件开发能力，为本地生活商家配备AI帮手；案例中，模型识别支付GMV增长23.2%而渠道利润下降1.7%的增长质量问题，并提出差异化资源配置建议。评测显示，其成绩介于Claude-Opus-4.8与Claude-Opus-5之间，部分评测超过后者。王兴此前表示，模型和AI产品将支持主业，提升用户、商户体验和经营效率。 @@@ Readhub消息，比亚迪10月新能源汽车销量为49.02万辆，同比增长10.97%；前10个月累计销量362.17万辆，同比下降2.16%。其中，10月纯电动汽车销量30.57万辆，同比增长37.38%；插电式混合动力汽车销量17.74万辆，同比下降17.22%。前10个月，纯电动汽车累计销量193.57万辆，同比增长5.86%；插混汽车累计销量162.60万辆，同比下降11.02%。此外，比亚迪10月出口新能源汽车17.02万辆。 @@@ 新浪科技消息，芯联集成控股公司芯联动力近日完成近20亿元C轮融资，中国移动链长基金领投，浪潮信息、元璟资本、中石化氢能基金、北方华创诺华资本、江丰电子、常州星宇、凯辉基金、普超资本、建信股权、中信金石、中信证券、银河源汇等联合投资。融资后股东覆盖通信、算力、电网、新能源及汽车等领域，多家AI基础设施产业资本入局，标志AI服务器电源成为下一阶段战略重点。袁锋称，公司将在巩固汽车碳化硅全球领先地位的同时，把AI服务器电源打造成新增长曲线。芯联动力专注碳化硅、氮化镓及模组封装，产品覆盖650V至3300V，6.5kV以上在研；今年6月推出的3300V SiC MOSFET面向固态变压器等场景，已向核心客户送样验证，综合方案可使系统级BOM成本降低20%至35%，填补国内高压碳化硅器件在固态变压器核心功率级应用的关键空白。 @@@ 新浪科技消息，10月28日欣旺达发布公告称，子公司欣旺达动力在C++轮融资中，由小米全资子公司以17亿元现金认购5.34%股份，小米由此成为欣旺达动力第三大股东。其第二大股东理想今年9月已出资26.5亿元认购8.79%，小米本轮出资后，理想持股稀释为8.32%，加上关联公司超过10%。欣旺达在欣旺达动力不再拥有绝对控制权，因此不再并表，理想和小米都将获得若干董事席位。报道称，小米与欣旺达在消费电子领域有合作，欣旺达的叠片工艺产能也是其被车企看中的原因之一；在理想供应商序列中，欣旺达占大头，宁德时代已降至30%以下，中创新航刚起步。第二天，欣旺达终止H股上市计划，该计划可能重新安排，或让欣旺达动力到香港上市。 @@@ 据虎嗅24小时消息，宇树机器人因过热退出登泰山大赛，随后被担架抬下山。 @@@ 好啦，今天的节目到这里就结束啦！谢谢大家的陪伴，记得带着好心情出发哦。如果您喜欢我们的节目，也非常欢迎订阅和转发给你的小伙伴，这对我们非常有帮助。那我们下期再见吧！ 在小宇宙查看该单集文稿

  5. 2d ago

    《晚安咖啡》科技夜话（英文版）：苹果市值反超英伟达，OpenAI融资下调，谷歌卫星在轨AI

    Hi everyone~~ Good evening! Today is Sunday, November 1, 2026, and welcome to Goodnight Coffee. I’m your host, An’an. Tonight, let’s talk about new developments in business and technology—relaxed, fun, and full of substance—to keep you company for a pleasant evening. Come along with me! @@@ According to Readhub, Apple has reclaimed the title of the world’s most valuable company, overtaking Nvidia. Amid concerns over AI spending, investors turned to Apple for relative safety. Apple shares fell 1.2% that day, giving it a market value of $5.03 trillion, while Nvidia fell 4.2% to a market value of $4.93 trillion. This is the first time Apple has overtaken Nvidia again since briefly surpassing it in late July. @@@ According to Readhub, OpenAI has lowered its latest fundraising target from an original $30 billion by $10 billion to about $20 billion, with the corresponding valuation adjusted from $1.4 trillion to $1.25 trillion, and has asked investors to complete their funding commitments by Friday. Affected by the selloff in AI stocks in public markets and the impending listing of top rival Anthropic, the fundraising process is under pressure. Some investors have reduced their subscriptions, and the first closing is expected to bring in less than $20 billion. A UAE-related investment consortium will be the largest investor in this round, while Founders Fund, Coatue, Sequoia Capital, Andreessen Horowitz and SoftBank also plan to participate. A retail allocation for wealthy individual investors will follow, and total fundraising is expected to reach around $20 billion. @@@ According to 36Kr, Meta plans to raise nearly $100 billion, and Oracle has put in the “final offer.” @@@ According to 36Kr, Google engineers recently sent a prompt to a prototype satellite in orbit via a ground station: “Write a haiku about being in space.” The satellite’s TPU started up and ran the pre-deployed Gemma 4 to complete inference. As the satellite passed near Svalbard, Norway, the ground station received the returned haiku. Google CEO Pichai called it Project Suncatcher’s “Hello World” moment, and the first time a data-center-class TPU had generated and sent back a complete AI response in orbit. The satellite was developed by Google in partnership with Planet and launched into orbit aboard SpaceX Transporter-18 on October 1. The team completed debugging and started the TPU within weeks, and only this week carried out the actual generative AI task. The satellite carries four Google TPUs and can draw up to about 1 kilowatt of power, but only one TPU can be started at a time, for about 20 minutes. Google originally planned to launch two prototype satellites in early 2027 to test in-orbit TPUs and inter-satellite optical communications, but adjusted the schedule and sent the TPU into orbit ahead of time in October. Next, it will move to more practical system tests to observe how the TPU performs under orbital thermal stress and radiation. @@@ According to cnBeta, the FBI has confirmed that a North Korean hacker group carried out the Binance theft, involving $6.3 billion. @@@ According to Readhub, the official DeepSeek Harness team published an article introducing the design approach of its Super-agent architecture for engineering practice. The article notes that traditional single-session agents have limitations such as handling multiple tasks, and summarizes the new autonomous agent paradigm with four characteristics: continuous, efficient, proactive and persistent. The architecture consists of five parts, including a dual-layer architecture and event-driven design, with multiple optimization mechanisms. Experiments show that the architecture outperforms traditional single-session solutions in task completion efficiency and routing throughput. At present, an internal instance based on this approach is already in operation. The team has also released a prototype implementation for technical validation, while noting that the architecture still has several unresolved issues. @@@ According to Sina Tech, ChangXin Memory Technologies said at its Q3 results briefing that its sixth-generation process technology platform based on the 4F² vertical-channel DRAM architecture has completed engineering sample validation. This is the first time ChangXin has announced that its sixth-generation platform adopts the 4F² architecture. Completion of engineering sample validation means the platform’s core performance indicators have met design expectations on real wafers, and after client validation it can move into mass production. ChangXin previously said in a public speech that it expects to launch a new-generation DDR5 RDIMM product incorporating 4F² technology for the server sector by the end of the year. According to publicly disclosed progress, Samsung and SK hynix have listed 4F² as a key direction below the 10nm class, and ChangXin is the first to complete engineering sample validation. If subsequent progress goes smoothly, ChangXin is expected to become the world’s first company to implement 4F². A 4F² cell area is about two-thirds that of a mainstream 6F² cell, and is seen as a key path to breaking the DRAM scaling bottleneck and moving toward 3D DRAM. The core of ChangXin’s approach is a self-developed VSG vertical surrounding-gate transistor, which can alleviate short-channel effects and leakage. The architecture decouples DRAM and CMOS manufacturing processes and uses wafer-level hybrid bonding for high-speed interconnection, allowing logic performance, memory density and reliability to be optimized separately. @@@ According to Huxiu 24 Hours, the People’s Bank of China will hold a symposium on relending for stock buybacks and increased holdings. @@@ According to Readhub, Xiaomi Auto delivered more than 50,000 vehicles in October. @@@ According to QuantumBit, XHEART made its first public appearance and launched the X7 Ultra chip, saying this is the first time a world model has guided chip design. The X7 Ultra delivers 700 TOPS of computing power, with NPU utilization of 98.7%; the output of quantized models has a similarity of over 99.99% to the FP32 baseline. Compared with industry flagships, it reduces memory bandwidth consumption by 80% and doubles energy efficiency. Its “large-model compute” evaluation directly uses production intelligent-driving models such as the Momenta R7 world model for testing, rather than relying only on laboratory benchmarks. The chip adopts a super-core NPU architecture and provides dedicated engines for high-frequency operators such as LayerNorm and Softmax. Its AI toolchain reduces data movement through computational graph fusion and parallel execution; the lidar accelerator processes 20 million points per second, and camera pipeline stability is improved by 100%. The CPU adopts the Arm v9 automotive architecture and is implemented in China for the first time; the SoC reaches ASIL-B, and the integrated MCU reaches ASIL-D. XHEART has no direct equity affiliation with Momenta; Momenta founding team members Cao Xudong and Xia Yan hold indirect stakes. The X7 has already entered mass production in new SAIC Volkswagen vehicles. Cao Xudong said that “large models define chips” is Momenta’s third important judgment. @@@ Alright, that’s all for today’s show! Thank you all for staying with us, and may good moods always be with you. If you enjoy our program, you’re very welcome to subscribe and share it with your friends—that helps us a lot. See you in the next episode! 在小宇宙查看该单集文稿

  6. 2d ago

    《晚安咖啡》科技夜话：苹果市值反超英伟达，OpenAI融资下调，谷歌卫星在轨AI

    嗨~~大家晚上好呀！今天是2026年11月1日星期日，欢迎来到晚安咖啡，我是主播安安～今天咱们一起聊聊商业科技新动态，轻松又有料，陪你度过愉快夜晚，快跟我出发吧！ @@@ 据Readhub报道，苹果公司重新夺回全球市值最高公司头衔，反超英伟达。在人工智能支出引发担忧的背景下，投资者转向苹果寻求相对安全感。当日苹果股价下跌1.2%，市值为5.03万亿美元；英伟达股价下跌4.2%，市值为4.93万亿美元。这是苹果自7月底短暂超过英伟达后，首次再次反超。 @@@ 据Readhub报道，OpenAI已将最新一轮融资目标从原定的300亿美元下调100亿美元至约200亿美元，对应估值由1.4万亿美元调整为1.25万亿美元，并要求投资者在周五前完成出资承诺。受公开市场AI板块抛售以及头号竞争对手Anthropic即将上市等因素影响，本轮融资进程承压，部分投资者缩减认购额度，首次交割预计到位资金不足200亿美元。阿联酋相关投资财团将成为本轮最大投资方，Founders Fund、Coatue、红杉资本、Andreessen Horowitz及软银等机构也计划参与。后续还将开启面向富裕个人投资者的零售配售，融资总额有望达到200亿美元左右。 @@@ 据36氪报道，Meta计划筹集近千亿美元，甲骨文报出了“最后一笔”。 @@@ 36氪消息，谷歌工程师近日通过地面站向轨道上的原型卫星发送“写一首关于身处太空的俳句”提示词，卫星TPU启动，运行预先部署的Gemma 4完成推理；卫星飞经挪威斯瓦尔巴群岛附近时，地面站收到回传俳句。谷歌CEO皮查伊称这是Project Suncatcher的“Hello World”时刻，也是数据中心级TPU首次在轨生成并回传完整AI响应。该卫星由谷歌与Planet合作研制，10月1日搭乘SpaceX Transporter-18入轨；团队数周内完成调试并启动TPU，本周才执行真正的生成式AI任务。卫星搭载四颗Google TPU，最高提供约1千瓦电力，但每次只能启动一颗，运行约20分钟。谷歌原计划2027年初发射两颗原型卫星测试在轨TPU和星间光通信，后调整进度，提前于10月将TPU送入轨道。后续将转向更实际的系统测试，观察TPU在轨道热应力和辐射环境下的表现。 @@@ 据cnBeta消息，美国联邦调查局确认，朝鲜黑客组织发动了币安盗窃案，涉案金额为63亿美元。 @@@ 据Readhub消息，DeepSeek Harness官方团队发布文章，介绍面向工程实践的Super-agent架构设计思路。文章指出，传统单会话Agent存在多任务处理等局限，并将新型自主Agent形态归纳为连续、高效、主动、持久四项特征。该架构由双层架构、事件驱动等五部分组成，并配套设计了多项优化机制。相关实验显示，该架构在任务完成效率和路由吞吐能力上优于传统单会话方案。目前，基于该思路的内部实例已投入运行，团队同步公开了用于技术验证的原型实现，同时指出该架构仍存在多方面待解决的问题。 @@@ 新浪科技消息，长鑫科技在第三季度业绩交流会上表示，基于4F²垂直沟道DRAM架构的第六代工艺技术平台已完成工程样片验证，这是长鑫首次宣布第六代平台采用4F²架构。工程样片验证完成意味着该平台核心性能指标已在真实晶圆上达到设计预期，在客户端完成验证后可进入规模量产。长鑫此前在公开演讲中称，预计年底面向服务器领域推出融合4F²技术的新一代DDR5 RDIMM产品。按公开披露进度，三星和SK海力士已将4F²列为10nm级以下关键方向，长鑫率先完成工程样片验证；若后续进展顺利，长鑫有望成为全球首个落地4F²的企业。4F²单元面积约为主流6F²单元的三分之二，被视作破解DRAM微缩瓶颈、迈向3D DRAM的关键路径。长鑫方案的核心是自研VSG垂直包围栅晶体管，可缓解短沟道效应和漏电；该架构将DRAM与CMOS制造工艺解耦，并通过晶圆级混合键合实现高速互联，使逻辑性能与存储密度、可靠性得以分别优化。 @@@ 据虎嗅24小时消息，中国人民银行将召开股票回购增持再贷款座谈会。 @@@ 据Readhub报道，小米汽车10月交付量超过5万台。 @@@ 据量子位报道，新芯航途XHEART首次公开亮相并发布X7 Ultra芯片，称这是世界模型首次指导芯片设计。X7 Ultra算力700 TOPS，NPU利用率98.7%，量化模型输出与FP32基准相似度超99.99%；对比行业旗舰，内存带宽消耗降80%，能效提升100%。其“大模型算力”评测直接用Momenta R7世界模型等量产智驾模型测试，而非仅依赖实验室基准。芯片采用超大核NPU架构，为LayerNorm、Softmax等高频算子配专用引擎，AI工具链通过计算图融合、并行执行减少数据搬运；激光雷达加速器每秒处理2000万点，摄像头链路稳定性提高100%。CPU采用Arm v9车规架构并首次在中国落地，SoC达ASIL-B、集成MCU达ASIL-D。新芯航途与Momenta无直接股权隶属，Momenta创始团队成员曹旭东、夏炎间接持股；X7已在上汽大众新车上量产，曹旭东称“大模型定义芯片”是Momenta第三个重要判断。 @@@ 好啦，今天的节目到这里就结束啦！谢谢大家的陪伴，愿好心情常伴你左右。如果您喜欢我们的节目，也非常欢迎订阅和转发给你的小伙伴，这对我们非常有帮助。那我们下期再见吧！ 在小宇宙查看该单集文稿

  7. 3d ago

    《晚安咖啡》科技夜话（英文版）：英特尔暴跌，苹果AI入华

    Hi there~~ Good evening, everyone! Today is Saturday, October 31, 2026. Welcome to Goodnight Coffee. I’m your host, An’an! Tonight, let’s chat about the latest, piping-hot business and tech stories—relaxed, fun, and full of substance—to keep you company through a lovely evening. So come along with me! @@@ According to The Verge, Apple announced that it will launch Apple Intelligence in the Chinese market with iOS 27.3. Some of the first features will arrive first in the iOS 27.3 beta planned for release on November 4, while the full version of iOS 27.3 is scheduled for December. This rollout will not include the new Siri AI; instead, it will offer earlier Apple Intelligence features such as notification summaries, live translation, rewriting and proofreading tools, Visual Intelligence, and object removal in photos. Apple also said it has partnered with Alibaba to support its Tongyi Qianwen chatbot in some experiences, similar to how iOS integrated OpenAI’s ChatGPT in 2024. Apple Intelligence was approved in China in July; Reuters reported in August that Apple had also trained its own large language model specifically for China. @@@ According to Readhub, after the market opened on October 30 local time, Intel’s stock plunged sharply, falling more than 30% at one point during the session, with more than $137 billion in market value wiped out, on track for its largest single-day drop on record. The report said the plunge was mainly due to the company’s fourth-quarter guidance coming in below market expectations, along with warnings that weakness in the PC market and rising memory prices could persist into the first half of 2027. The earnings report showed that Intel’s third-quarter 2026 revenue rose 21% year over year to $16.51 billion, but its fourth-quarter revenue guidance was far below analysts’ expectations, and the company also said it would not achieve positive cash flow this year. In addition, an explosion at its factory in Israel is expected to cause about $600 million in revenue losses within two quarters; the company has invested heavily in AI chips and manufacturing but is constrained by capacity and supply chain limits, making it difficult to fully meet data center customer demand. The decline also affected several semiconductor companies, including AMD and Nvidia. @@@ According to Huxiu 24 Hours, Anthropic is sprinting toward an IPO, with the report describing the listing process as part of its race with OpenAI in valuation, revenue, and computing power. @@@ According to The Verge, Google’s Project Suncatcher satellite has generated an AI response from space for the first time. In a blog post, Google said the team successfully booted up the tensor processing unit, or TPU, aboard the satellite, enabling the onboard Gemma 4 model to compose a haiku about being in space in response to a prompt: “Blue Earth lies below, / data streams in silence flow, / starlight feeds my core.” Google said it also tested other space-themed limericks and telemetry prompts. The test is part of the Project Suncatcher experimental program, which could eventually send AI data centers into space; earlier this month, Google launched the satellite into orbit. Next, Google will conduct a series of tests to assess how the TPU performs under orbital thermal stress and radiation exposure. Google said that verifying whether TPUs can reliably run workloads in space will pave the way for its next milestone in 2027: two interconnected satellites. @@@ TechWeb reports that ChangXin Technology’s third-quarter 2026 financial results show quarterly revenue of RMB 118.714 billion, up 19.30% quarter over quarter; net profit was about RMB 88.983 billion, up about 19.38% quarter over quarter. For the first three quarters, revenue was RMB 269.024 billion, up 738.51% year over year; net profit attributable to parent was RMB 132.04 billion, turning from a loss to a profit year over year. The company said rising global demand for computing power and capacity allocation by major manufacturers have left DRAM in short supply; capacity construction and product mix optimization drove sharp increases in sales volume and selling prices; higher prices, combined with economies of scale and lean production, lowered unit costs and sharply increased gross profit. First-three-quarter R&D expenses were RMB 11.071 billion, up 73.03% year over year, accounting for 4.12% of revenue. The company disclosed that DRAM based on its sixth-generation platform has completed engineering sample validation; the report also noted that G6 continues to use DUV, with cell area reduced from 0.00128205 square micrometers in G5 to 0.000896 square micrometers, a reduction of about 30%. G5 was announced for official mass production on September 20, 2026, with the active-area half-pitch reduced to 11.95 nanometers. @@@ According to an exclusive report by LatePost, Volcano Engine has fully shifted its business focus to commercialization, and its MaaS, or Model-as-a-Service, business posted a gross margin of over 55% in the first half of the year. @@@ According to 36Kr, on October 28, Unitree Robotics, which has been listed for just over two months, announced that it has jointly established the “Shanyu Venture Program” with Sequoia China to enter early-stage investment in embodied intelligence. The program will focus on embodied AI, robotics applications, and core upstream and downstream technologies, providing funding, technical guidance, product validation, and industry resources through an “engineer + investor” approach. Its five principles are: “understand technology and industry, discover early, decide fast, value people, and stay out of the way.” The two sides have a seven-year partnership: in December 2019, Sequoia invested RMB 15 million in Unitree, and has since invested a cumulative total of about RMB 102 million. Unitree listed on the STAR Market in August this year; Sequoia holds a 7.11% stake. Based on the RMB 882 share price in early trading on its first day of listing, the paper return exceeds 220 times. On the same day, Unitree disclosed that in the first three quarters its revenue was RMB 1.817 billion, up 55.71% year over year; net profit attributable to parent was RMB 387 million, up 270.54% year over year; and R&D expenses were RMB 263 million, up 191.48% year over year. Wang Xingxing previously revealed that the “Yuchuang Program” received more than 200 projects in just over a month, and that in the next five years the company hopes to incubate more than 200 ecosystem companies. @@@ According to Sina Tech, 360 Group’s self-developed AI security attack-and-defense system, Tulongfeng, recently topped the international authoritative benchmark CyberGym with a 99.4% vulnerability reproduction success rate. Other contenders on the same leaderboard included models and agents from leading global AI companies such as OpenAI, Anthropic, and Google. CyberGym focuses on complex real-world challenges, covering a large number of mainstream open-source software programs and difficult historical vulnerabilities. It tests large models and security agents on autonomous reasoning, vulnerability reproduction, and PoC validation in massive source code, complex contexts, and multi-stage adversarial environments. Evaluation data shows that across all 1,507 tasks and three independent rounds of evaluation, Tulongfeng achieved success rates of 99.27%, 99.47%, and 99.40%, respectively, with an average success rate of 99.38%—equivalent to successfully tackling nearly 1,500 real vulnerabilities and setting a new record high on the leaderboard. To date, Tulongfeng has discovered more than 15,000 security vulnerabilities, covering operating systems, office software, AI applications, and other fields; the system has passed the China Academy of Information and Communications Technology’s capability assessment for vulnerability mining agents, and more than 500 enterprises have adopted it for specialized security testing, covering key industries such as finance, energy, government affairs, and manufacturing. @@@ TechWeb, citing Nikkei Chinese, reports that Toyota, Honda, Nissan, and Mazda produced a combined 1,020,579 vehicles in China from April to September this year, down 32% year on year, hitting the lowest level since 2008 and down 63% from the peak of about 2.8 million in the same period of 2020—less than 40% of that peak. Honda saw the largest drop, producing 131,716 vehicles, down 60% year on year and about 90% from its peak; its all-electric Ye GT was originally scheduled to launch in December 2025 but has been indefinitely delayed until after 2026. Nissan produced 163,192 vehicles, down 49% year on year and 80% from its peak; Toyota produced 663,079 vehicles, down 18% year on year and 23% from its peak; Mazda’s output rose 43% year on year, but its volume is limited. The main reason is shrinking demand for fuel-powered vehicles: from January to September, the new-car market fell 22% year on year, and new-car sales excluding new energy vehicles dropped by more than 30%. Honda plans to cut its annual fuel-vehicle production capacity in China from 1.49 million to 1 million. Mobility Global forecasts that the four automakers’ production in China in 2030 will fall another 47% from 2025 to about 1.6 million vehicles. @@@ Alright, that’s all for today’s show! Thank you for spending this time with us. I hope today’s little sharing has filled you with good vibes~ If you like our program, you’re very welcome to subscribe and share it with your friends—it really helps us a lot. See you next time! 在小宇宙查看该单集文稿

  8. 3d ago

    《晚安咖啡》科技夜话：英特尔暴跌，苹果AI入华

    嗨~~大家晚上好呀！今天是2026年10月31日，星期六，欢迎来到晚安咖啡，我是主播安安！今晚咱们一起聊聊新鲜热乎的商业科技动态，轻松又有料，陪你度过美好夜晚，快跟我出发吧！ @@@ 据The Verge报道，苹果宣布将随iOS 27.3在中国市场推出Apple Intelligence。首批部分功能将先行登陆计划于11月4日发布的iOS 27.3测试版，完整版iOS 27.3则计划于12月发布。此次上线不包括新的Siri AI，而是提供通知摘要、实时翻译、改写和校对工具、视觉智能以及照片物体移除等较早推出的Apple Intelligence功能。苹果还表示，已与阿里巴巴合作，在一些体验中支持其通义千问聊天机器人，类似2024年iOS整合OpenAI ChatGPT的方式。Apple Intelligence已于7月在中国获批；路透社8月报道称，苹果还专门为中国训练了自己的大语言模型。 @@@ 据Readhub报道，当地时间10月30日开盘后，英特尔股价大幅下挫，盘中跌幅一度超过30%，市值蒸发超过1370亿美元，有望创下有记录以来最大单日跌幅。报道称，暴跌主要因公司发布的第四季度业绩指引低于市场预期，并警告PC市场疲软、内存价格上涨等问题可能持续至2027年上半年。财报显示，英特尔2026年第三季度营收同比增长21%，达到165.1亿美元，但第四季度营收指引远低于分析师预期，公司还表示今年无法实现正现金流。此外，以色列工厂爆炸事故预计将在两个季度内造成约6亿美元营收损失；公司大力投资AI芯片和制造业务，却受产能和供应链限制，难以充分满足数据中心客户需求。此次下跌还波及AMD、英伟达等多家半导体企业。 @@@ 据虎嗅24小时消息，Anthropic正在冲刺IPO，报道将这一上市进程描述为其与OpenAI在估值、营收和算力上的竞赛。 @@@ 据The Verge报道，Google的Project Suncatcher卫星已首次从太空生成AI响应。Google在博客文章中称，团队成功启动卫星搭载的张量处理单元（TPU），使机载Gemma 4模型按照提示创作了一首关于身处太空的俳句：“蓝色地球在下方，数据流在静默中优雅涌动，星光滋养我的核心。”Google表示，还测试了其他太空主题打油诗和遥测提示。这项测试属于Project Suncatcher实验计划的一部分，该计划最终可能把AI数据中心送入太空；Google本月早些时候已将这颗卫星送入轨道。接下来，Google将开展一系列测试，评估TPU在轨道热应力和辐射暴露下的表现。Google称，验证TPU能否在太空可靠执行工作负载，将为2027年实现两颗互联卫星的下一个里程碑铺路。 @@@ TechWeb消息，长鑫科技2026年第三季度财报显示，当季营业收入1187.14亿元，环比增19.30%；净利润约889.83亿元，环比增约19.38%。前三季度营收2690.24亿元，同比增738.51%；归母净利润1320.4亿元，同比扭亏为盈。公司称，全球算力需求攀升及主要厂商产能调配令DRAM供不应求，产能建设和产品结构优化带动销量、售价大幅上涨；价格拉升叠加规模效应和精益生产使单位成本下降，毛利大幅上升。前三季度研发费用110.71亿元，同比增73.03%，占营收4.12%。公司披露，基于第六代平台的DRAM已完成工程样片验证；报道还提到，G6继续采用DUV，单元面积由G5的0.00128205平方微米降至0.000896平方微米，缩小约30%。G5已于2026年9月20日宣布正式量产，有源区半间距缩至11.95纳米。 @@@ 据晚点LatePost独家报道，火山引擎已将业务重心全面转向商业化，其MaaS（模型即服务）业务上半年毛利率超过55%。 @@@ 据36氪报道，10月28日，上市两个多月的宇树科技宣布与红杉中国共同成立“杉宇·创计划”，进军具身智能早期投资。该计划重点投资具身智能、机器人应用及上下游核心技术，以“工程师+投资人”方式提供资金、技术指导、产品验证及产业资源。五项原则为“懂技术和产业、发现早、决策快、看重人、不添乱”。 双方合作已有七年渊源：2019年12月红杉向宇树投入1500万元，此后累计投入约1.02亿元。宇树今年8月登陆科创板，红杉持股7.11%，按上市首日早盘882元股价测算，账面回报超220倍。 同日宇树披露，前三季度营收18.17亿元，同比增长55.71%；归母净利润3.87亿元，同比增长270.54%；研发费用2.63亿元，同比增长191.48%。王兴兴此前透露，“宇创计划”一个多月收到超200个项目，未来五年希望孵化200家以上生态企业。 @@@ 新浪科技消息，360集团自研的AI安全攻防系统图龙锋近日在国际权威基准测试CyberGym中以99.4%的漏洞复现成功率登顶榜单，同榜竞逐者包括OpenAI、Anthropic、Google等全球头部AI厂商的模型与智能体。CyberGym聚焦真实世界复杂挑战，覆盖大量主流开源软件与高难度历史漏洞，重点考验大模型和安全Agent在海量源码、复杂上下文及多阶段对抗环境下的自主推理、漏洞复现与PoC验证能力。评测数据显示，图龙锋在全部1507个任务中经过三轮独立评测，成功率分别为99.27%、99.47%和99.40%，平均成功率99.38%，相当于成功攻克近1500个真实漏洞，刷新该榜单最高纪录。截至目前，图龙锋累计发现安全漏洞超15000个，覆盖操作系统、办公软件及AI应用等领域；该系统已通过中国信通院漏洞挖掘智能体能力评估，超500家企业接入开展专项安全测试，覆盖金融、能源、政务、制造等重点行业。 @@@ TechWeb援引日经中文网报道，丰田、本田、日产、马自达四家车企今年4至9月在华合计产量102.0579万辆，同比降32%，创2008年以来新低，较2020年同期约280万辆峰值降63%、不足四成。本田跌幅最大，产量13.1716万辆，同比降60%，较峰值降约90%；纯电“烨GT”原定2025年12月上市，已无限期延至2026年后。日产16.3192万辆，同比降49%，较峰值降80%；丰田66.3079万辆，同比降18%，较峰值降23%；马自达同比增43%但体量有限。燃油车需求萎缩是主因，1至9月新车市场同比降22%，除新能源外新车销量降逾三成。本田计划将在华燃油车年产能从149万辆减至100万辆。Mobility Global预测，四家车企2030年在华产量比2025年再降47%至约160万辆。 @@@ 好啦，今天的节目到这里就结束啦！谢谢大家陪伴，愿今天的小分享给你满满好心情～如果您喜欢我们的节目，也非常欢迎订阅和转发给你的小伙伴，这对我们非常有帮助。那我们下期再见吧！ 在小宇宙查看该单集文稿

See All (1.4K)

About

"晚安咖啡" 一杯陪你从清晨到深夜的神奇咖啡。 早晨，来杯清醒提神的商业浓缩， 读懂全球科技脉搏， 创业风云、商业变革， 让你的早安问候都带着点儿专业范。 午间，来杯香醇的娱乐拿铁， 影视剧追踪，明星动态， 让你的话题库永远新鲜， 办公室八卦，从此不用怕。 夜晚，来杯温暖特调， 治愈故事，生活感悟， 让一天的疲惫， 在耳机里慢慢融化。 每一杯咖啡的温度， 都能让你的碎片时间闪闪发光。 洗漱、通勤、午休、健身， 每个平凡时刻， 都能遇见不平凡的分享。 这是一杯， 让你在生活中保持清醒， 但不会让你失眠的咖啡。 你的随身资讯管家， 每天都为你精心冲泡， 一杯又一杯， 不见不散。

Information

  • Creator
    晚安咖啡
  • Years Active
    2024 - 2026
  • Episodes
    1.4K
  • Rating
    Clean
  • Copyright
    © @晚安咖啡GoodNightCoffee
  • Show Website
    晚安咖啡GoodNightCoffee

You Might Also Like