跳到正文
11月2日周一
  1. a16z62

    a16z 的 Moses Sternstein 和 Alex Danco 提出 rack spread 概念,指 GPU 价格上涨与 token 价格下跌之间的缺口。自 7 月中旬以来 B300 GPU 租金上涨 106%,B200 上涨 66%,H100 上涨 17%,平均 token 价格下跌 58%;两人将原因归为开放竞争与模型实验室降价。

    AI 生成摘要 · 以原文为准

    引用a16z@a16z

    The "Rack Spread" is unfolding, token prices are falling while GPU prices climb Change since mid-July: - B300 GPU rentals up 106% - B200 up 66% - H100 up 17% - Average token prices down 58% Charts of the Week: https://www.a16z.news/p/charts-of-the-week-job-fud-fades

  2. The Quantum Insider52

    Nord Quantique CEO 访谈:单腔实现纠错量子比特与路线图

    Nord Quantique 联合创始人兼 CEO Julien Camirand Lemire 接受 Yuval Boger 访谈,介绍基于超导腔内微波光子的玻色编码路线,用单腔实现量子纠错以降低比特开销。公司 2024 年发布 GKP 单腔全纠错演示(无后选择,纠错增益约 1.1,现已超过 2),2025 年发布 Tesseract 多模编码,四比特系统今年投入运行。

    AI 生成摘要 · 以原文为准

10月31日周六
  1. Quantum Computing Report48

    Quantinuum 首席科学家:物理保真度比门速度更决定容错量子扩展竞争

    Quantinuum 首席科学家 Harry Buhrman 在《Fidelity First in the Quantum Race》一文中主张,物理保真度而非纳秒级门速度才是决定实用求解时间的核心参数。其 Helios 系统 98 量子比特离子阱架构实现 1Q 门保真度 99.9975%、2Q 99.921%,并在距离 7 码上实现每周期 5.0×10⁻⁶ 逻辑错误率;而超导平台依赖误差缓解。

    AI 生成摘要 · 以原文为准

  2. Ben Bajarin29

    Avicena 让我从头号质疑者变成了超级多头。等不及写完整篇分析。他们共进行了三场会议、总计约 6 小时,对每一个问题/疑虑——电气、光学、色散、FEC、光纤成本、对准、封装、DFE 反向传播——都给出了出色的回答。

    AI 生成摘要 · 以原文为准

    引用Irrational Analysis@insane_analyst

    Avicena has converted me from #1 hater to an ultra bull. Cant wait to finish writeup. Seriously there were three meetings totaling ~6 hours. They had excellent answers to every question/concern. Electrical, optic, dispersion, FEC, fiber cost, alignment, packaging, DFE backprop.

  3. The Humanoid Hub35

    马斯克否认了 Optimus 将于今年 12 月开放预订的报道。 该报道源自 Humanoids Summit 组织者 Modar Alaoui 接受 The San Francisco Standard 采访时所说。

    AI 生成摘要 · 以原文为准

    引用The Humanoid Hub@TheHumanoidHub

    Humanoids Summit organizer Modar Alaoui has spilled the beans on Tesla's secret Optimus plans in an interview with The San Francisco Standard. What he says Tesla told him: Optimus reservations will be timed to launch with his summit (Dec 1-2), or in the days right after. Tesla is a key partner of the event this year and plans to demo the Optimus V3 and showcase its self-driving cars there. What Alaoui expects on top of that: consumer reservations open in early December (a first for the public), the first shipment being the V3 since the prior gen was mainly a prototype, and most robots going to companies with only a small number set aside for consumers, at an expected $20,000+.

  4. Ben Bajarin39

    我们对话中关于 DPU 的这一部分很值得理解。我一段时间以来一直在看好 DPU,但随着安全/网络安全计算成为基本要求,DPU 将扮演更大的角色。

    AI 生成摘要 · 以原文为准

    引用The Diligence Stack@DiligenceStack

    When agents become users, operators lose the assumption that every workload is trusted. Mohamed Awad explains why DPUs and hardware isolation move deeper into the AI control plane. Full paid interview: https://www.thediligencestack.com/p/executive-interview-arms-mohamed

  5. Morgan Stanley · Thoughts on the Market45

    Morgan Stanley:信贷市场平静表象下的“猴爪效应”

    Morgan Stanley 固定收益研究主管 Andrew Sheets 分析当前信贷市场的波动:央行加息缩表、收益率上升,AI 热潮推动今年近 9000 亿美元 CapEx、明年超大规模厂商资本开支或达约 1.5 万亿美元,科技板块发债推高收益率与利差并扩大信用分化。他认为这些条件正符合投资者过去十几年的愿望,当前局面整体是健康正常的,无需过度恐慌。

    AI 生成摘要 · 以原文为准

10月30日周五
  1. Ben Bajarin69

    Ben Bajarin转发@pequityresearch的Nebius研究要点,并称自己几周前所说的算力价格如今基本成了竞价。要点包括:Nebius采用竞价驱动的集群接入定价,通过竞争性抬价使价格水平提高约20%;要求60–65%预付款,相比高杠杆的CoreWeave(新债务利率高达15%)风险更低;短期现货容量每兆瓦ACV达4000万–5000万美元,竞价加持下或达6000万美元,中期合约则为2000万–2500万美元;最新一代GPU供应紧张将持续到2028年中,公司计划把2027年产能留给高毛利现货负载。研究还称Meta和SpaceX各自控制超过100万张GPU,每月部署2.5万–5万张,正从neocloud手中抽走裸金属合约;Gartner预测到2030年超过80%的独立neocloud将倒闭;Nebius核心客户Microsoft、Meta为AA/AAA评级而受保护;GB200/GB300正转向毛利高出4–8倍的推理负载;Micron预计HBM3/HBM4短缺将延续到2028年以后,Microsoft内部30万颗Maia ASIC替代了30万张以上Nvidia GPU的需求。

    AI 生成摘要 · 以原文为准

    引用P Equity Research 📰@pequityresearch

    Expert Takeaways: Nebius > Nebius operates an auction-driven cluster-access pricing model that drives up pricing levels by ~20% through competitive bidding. > Nebius requires 60–65% upfront contract payments, providing strong protection compared to debt-heavy peers like CoreWeave (facing up to 15% interest on new debt). > Short-term spot capacity fetches $40–50M ACV per megawatt (potentially up to $60M with auction uplift) vs. $20–25M for medium-term deals. > Tight supply for latest-generation GPUs through at least mid-2028 supports reserving 2027 capacity for high-margin spot workloads. > Meta and SpaceX control >1M GPUs and plan to deploy 25k–50k units monthly, pulling bare-metal contracts away from neoclouds. > Open-source models deliver comparable results at 70–80% lower cost, eroding OpenAI and Anthropic enterprise revenue. > Gartner projects >80% of independent neoclouds will fail by 2030 due to a K-shaped market favoring GB300/Vera-class hardware over older GPUs. > Nebius is insulated from cascading neocloud defaults because its core clients (Microsoft, Meta) are AA/AAA-rated, unlike BB-rated frontier labs. > High-end hardware (GB200/GB300) is shifting to inference workloads, which offer 4–8× higher margins than training. > Micron predicts HBM3/HBM4 memory shortages beyond 2028, while Microsoft’s 300,000 internal Maia ASICs displace demand for 300,000+ Nvidia GPUs. $NBIS $GOOGL $META $MSFT $AMZN

  2. Utility Dive60

    州监管机构应以审慎态度审视以数据中心需求为由的新建燃气电厂

    环保防御基金和 CURE 的两位作者撰文称, utilities 正以不确定的数据中心需求预测为由申请建设昂贵的新燃气电厂,若需求落空,普通用户将承担长达 30 年的付费。文章举例称,明尼苏达监管机构要求 Minnesota Power 探索风电、光伏和储能等更低成本替代方案,北卡监管机构以负荷预测不足、风险转嫁给居民用户为由拒绝了 Duke Energy 一座 5.84 亿美元的燃气轮机项目。

    AI 生成摘要 · 以原文为准

  3. 华尔街见闻 · 高盛61

    高盛发布美国中期选举交易指南,按监管压力给出分行业多空建议

    高盛发布针对美国中期选举的交易指南,博彩市场目前将民主党横扫两院的概率定价在67%。核心建议为做多医疗保健ACA受益标的、可再生能源与电网基础设施、非国防政府支出受益标的及住宅建筑商,做空大型药企、国防股、大型科技股、投资银行与加密货币,并对超大规模云计算服务商及数据中心设备买入保护性头寸。

    AI 生成摘要 · 以原文为准

  4. 华尔街见闻 · 美银66

    美银Hartnett:超大规模云厂商2026年发债1820亿美元,科技债回报跌至30年低点

    美银Hartnett团队在10月29日《Flow Show》周报中指出,AI超大规模云厂商2026年发债规模已达1820亿美元,远高于2024年的170亿美元,资本开支预计从2024年2400亿美元升至2027年预估1.2万亿美元,自由现金流同期从2400亿美元降至2027年负1800亿美元。美国投资级科技债相对整体投资级债券指数的总回报跌至30年低点。

    AI 生成摘要 · 以原文为准

  5. The Quantum Insider44

    富士康目标成为全球最大量子计算机制造商

    富士康董事长刘扬伟向 CNBC 表示,公司将为量子技术开发商代工制造整机,目标是“不拥有核心技术也成为全球最大量子计算机制造商”。他认为量子计算距商业可行至少还需两年,且是全行业共同的挑战。他将量子计算机定位为加速特定计算的专用设备,与 CPU、GPU 互补而非取代,但指出其组件成本高昂,一根符合温度要求的线缆就可能超过 1000 美元。

    AI 生成摘要 · 以原文为准

  6. Goldman Sachs · Exchanges43

    Brookfield CEO Connor Teskey 谈 AI 建设潮的真正瓶颈

    Brookfield Asset Management CEO Connor Teskey 在 Goldman Sachs 播客中表示,AI 建设的瓶颈既非资本也非需求,而是拿地、并网和建设电力供应的运营能力。他还指出市场认为数据中心需求“脆弱”是一种误解,并谈到长期投资理念和私人信贷平台的发展。

    AI 生成摘要 · 以原文为准

  7. Ben Bajarin16

    CPU 设计中最重要的经验之一:今天做出的架构决策,决定了你明天的性能提升空间。真正的进步不只是把频率推得更高或消耗更多电力,而在于做出正确的取舍。💎

    AI 生成摘要 · 以原文为准

    引用Gerard Williams@gerard3805

    One of the most important lessons in CPU design is that the architectural decisions you make today determine the performance headroom you have tomorrow. Real progress isn't just about pushing frequencies higher or consuming more power. It's about making the right tradeoffs.

  8. a16z41

    a16z 的 Connor Love 表示,面向国防建设就是为其他一切领域建设:军用系统所需的执行器、电机、阀门和开关等子系统,与未来机器人系统完全相同,一旦能“按质量和按成本”制造这些部件,原本不可行的市场就会变得可行。他认为 AI 终将能“转动扳手”,届时美国将迎来机器人浪潮,且无需依赖中国工厂。他强调美国工业基础在组件层面严重不足,必须补强。

    AI 生成摘要 · 以原文为准

    引用a16z@a16z

    The American Dynamism 50 is a16z's annual list of companies building for America's future. This year's edition covers the industrial base end to end, from the inputs that come out of the ground to factories that make things and systems that deter conflict. a16z's Erin Price-Wright and Connor Love join Elena Burger to break down this year's list and its theme, Production is Deterrence, built around a simple idea: "You don't have a defense industrial base without an industrial base." 0:50 From mines to factories and systems 3:05 How the internet and nuclear was built 5:05 Drones built DJI into a robotics giant 6:50 Defense dollars pay for new factories 8:30 SpaceX and Tesla lessons on scale 10:15 How Mariana runs a mine on software 12:20 "AI cannot turn a wrench yet" 14:10 Retooling in 20 mins, not 3 months 17:45 Startups build before Pentagon asks 20:35 Why hard tech can be more valuable 24:45 The race for the biggest American flag 28:25 How defense parts unlock US robotics 32:05 The rise of the "manufacturing middle" 37:25 Anduril and SpaceX founder mafias 41:05 Private capital: USA's edge over China 46:30 The buildout must hit Middle America YouTube: https://youtu.be/cRJ7CfaK1Vc @espricewright @ConnorLoveCA @VirtualElena

  9. a16z37

    a16z 的 Connor Love 与 Erin Price-Wright 谈创业公司能从 SpaceX 和 Tesla 学到什么:两家公司因美国缺乏可支撑的产业基础而被迫高度垂直整合、自建一切。两人认为,有高质量供应商时应使用供应商;只有导弹这类需要控制精度与成本的独特系统,才值得全面垂直整合,核心是“面向制造设计”的理念。

    AI 生成摘要 · 以原文为准

    引用a16z@a16z

    The American Dynamism 50 is a16z's annual list of companies building for America's future. This year's edition covers the industrial base end to end, from the inputs that come out of the ground to factories that make things and systems that deter conflict. a16z's Erin Price-Wright and Connor Love join Elena Burger to break down this year's list and its theme, Production is Deterrence, built around a simple idea: "You don't have a defense industrial base without an industrial base." 0:50 From mines to factories and systems 3:05 How the internet and nuclear was built 5:05 Drones built DJI into a robotics giant 6:50 Defense dollars pay for new factories 8:30 SpaceX and Tesla lessons on scale 10:15 How Mariana runs a mine on software 12:20 "AI cannot turn a wrench yet" 14:10 Retooling in 20 mins, not 3 months 17:45 Startups build before Pentagon asks 20:35 Why hard tech can be more valuable 24:45 The race for the biggest American flag 28:25 How defense parts unlock US robotics 32:05 The rise of the "manufacturing middle" 37:25 Anduril and SpaceX founder mafias 41:05 Private capital: USA's edge over China 46:30 The buildout must hit Middle America YouTube: https://youtu.be/cRJ7CfaK1Vc @espricewright @ConnorLoveCA @VirtualElena

10月29日周四
  1. Light Reading60

    T-Mobile 财报会称将在美国首个推出 6G,并否决 SpaceX MVNO 合作

    T-Mobile 在三季度财报电话会上表示,按行业预期 2030 年前后将建成 6G 网络,公司计划成为美国首家推出 6G 的运营商,CTO John Saw 介绍了 ISAC(通信感知一体化)等新应用。公司同时拒绝了与 SpaceX 在消费者及物理与边缘 AI 业务上的 MVNO 合作,称卫星只是互补产品。

    AI 生成摘要 · 以原文为准

  2. Light Reading56

    Nvidia 表示不打算用 GPU 攻入固网定制芯片,但看好 AI 与 DPU 机会

    Light Reading 报道称,Nvidia 明确表示不计划像 AI-RAN 那样把 GPU 引入固网接入(FAN)的定制芯片,称固定接入主要涉及成帧、复用、FEC 等确定性功能,专用硅片已高度优化。Nvidia 已通过 10 亿美元投资促使 Nokia 停止自研 RAN 定制芯片而转向 GPU 软件。

    AI 生成摘要 · 以原文为准

  3. 𝐷𝑟. 𝐼𝑎𝑛 𝐶𝑢𝑡𝑟𝑒𝑠𝑠37

    这是我对 Michael Kagan(Mellanox 创始人、NVIDIA 现任 CTO)的访谈。我们聊了他的人生经历、他职业生涯中网络技术的演进,以及 NVIDIA 目前构建“行星级”计算的工作。 (主推文:祝贺哥们!太棒了的一场对话。)

    AI 生成摘要 · 以原文为准

    引用Nick Dorsey@Midnight_Captl

    This is my conversation with Michael Kagan, the founder of Mellanox and current CTO of NVIDIA We speak about his life, the progression of networking over the span of his career, and how the work that NVIDIA is now doing to build "Planet scale" computing I hope you enjoy the conversation as much as I did 00:00 Highlights 01:16 Growing Up in Russia and Moving to Israel 03:42 Intel, the PC Revolution, and a Passion for Technology 06:14 Sponsor: Liquid 06:33 Lessons from Intel: Software and Networking 10:10 Founding Mellanox and the Vision for Cloud Computing 14:34 From Supercomputers to the Cloud 17:46 How GPUs Sparked the AI Revolution 20:52 Sponsor: Compute Exchange 21:14 NVLink and the Rise of Rack-Scale Computing 24:57 The Engineering Challenges Behind NVL72 27:39 Designing Hardware for Future AI Workloads 29:43 BlueField DPUs and the AI Factory 32:55 Making AI Agents Safer 35:37 What Mellanox Could Have Done Differently 37:03 NVLink, InfiniBand, and Spectrum-X 39:24 Connecting AI Factories at Planetary Scale 42:42 Keeping Massive AI Systems Running Through Failures

  4. 华尔街见闻 · 高盛62

    高盛建议用期权结构“租用”AI反弹,配置回溯看跌期权对冲

    高盛衍生品交易台建议投资者通过有限损失的期权结构参与AI板块反弹,而非直接持有现货,并推荐2027年3月到期、行权价随市上调、成本1.82%的SPX 90%回溯看跌期权作为对冲。报告中提到大型科技股隐含波动率处于过去四年第15百分位,而新兴云服务商篮子本月大跌超28%,高盛建议用CoreWeave可转债替代正股,并给出做空指数波动率、做多苹果和英伟达等19只成分股波动率的离散交易。

    AI 生成摘要 · 以原文为准

  5. 华尔街见闻 · 瑞银48

    瑞银再度下调KOSPI目标位至7500点,看好2027年复苏

    瑞银10月29日将韩国KOSPI 12个月目标价从8000点下调6%至7500点,为年内第二次下调,主因韩元自6月以来升值约15%及EPS自下而上修正。瑞银认为这是汇率驱动的盈利重置而非下行周期,KOSPI剔除三星电子和SK海力士后NTM市盈率仅6.6倍,估值已处历史偏低水平。瑞银预计年底前市场仍承压,上行目标位9300点,2027年EPS增速或达+33%。

    AI 生成摘要 · 以原文为准