科技日报-美股盘前-20260601

TMT BREAKOUTJUN 01, 2026∙PAID Good morning. Futures flattish as software +3.5% continues to squeeze higher at theexpense of semis -80bps - software just had its best month in May since 2002 up 21% (we had some thoughts on the move and what’s next inour weeklyin case youmissed it). 早上好。期货表现平稳，软件板块上涨3.5%继续挤压⾛⾼，⽽半导体板块则下跌80个基点⸺软件板块刚刚经历了⾃2002年以来表现最好的5⽉，涨幅达21%（如果您错过了，我们在周报中对这⼀⾛势及后续展望发表了⼀些看法）。 NVDA +2%announced they are entering the market with a new RTX chip, which hasARM +11%early andINTC -5%/AMD-4%/QCOM -9%.Memory the bright spot insemis while opticals continue to lag.Hyperscalers weak withGOOGL -1%followingthrough on the neg 3p data from Friday while AMZN -1%. NVDA上涨2%，宣布将凭借新款RTX芯⽚进⼊市场，这带动ARM早盘上涨11%，⽽INTC下跌5%、AMD下跌4%、QCOM下跌9%。存储芯⽚是半导体板块中的亮点，⽽光模块板块继续落后。超⼤规模云⼚商表现疲软，GOOGL下跌1%，延续了周五负⾯第三⽅数据的跌势，同时AMZN下跌1%。 Asia mixed butKoreacontinued to ripand many memory/robotics AI names upsignificantly: TPX -0.42%, NKY +0.91%, Hang Seng +0.86%, HSCEI +0.97%, SHCOMP-0.27%, Shenzhen -0.78%, Taiwan TAIEX +1.35%, Korea KOSPI +3.68%.Softbank+14%. Samsung +12%. Kioxia +8%15 亚洲市场涨跌互现，但韩国股市继续飙升，许多存储/机器⼈AI相关个股⼤幅上涨：东证指数(TPX) -0.42%，⽇经指数(NKY) +0.91%，恒⽣指数+0.86%，恒⽣中国企业指数(HSCEI) +0.97%，上证综指(SHCOMP)-0.27%，深证成指-0.78%，台湾加权指数(TAIEX) +1.35%，韩国综合指数(KOSPI) +3.68%。软银上涨14%。三星上涨12%。铠侠(Kioxia)上涨8%。 On the macro front, yields ticking up 2-3bps across the curve while Oil popping 3% asan Iran agreement continues to prove elusive. 宏观⽅⾯，各期限收益率⼩幅上升2-3个基点，⽽由于伊朗协议依然难以达成，原油价格跳涨3%。 In Tech, we have a busy week coming up with Earnings from AVGO, CRWD, CRDO,HPE, GTLB, PANW, CIEN, DOCU and more. Computex continues through the weekwith keynotes INTC’s Lip Bu Tan, and MRVL CEO.BAML’s and ISI’s TMT conferencesstart Tuesday as well as the William Blair Growth conf. We get SNOW and KioxiaInvestor Days on Tuesday. 科技⾏业即将迎来繁忙的⼀周，AVGO、CRWD、CRDO、HPE、GTLB、PANW、CIEN、DOCU等公司将陆续发布财报。Computex台北电脑展将贯穿本周，INTC的陈⽴武（Lip Bu Tan）和MRVL⾸席执⾏官将发表主题演讲。美银美林（BAML）和Evercore ISI的TMT⼤会以及William Blair成⻓股⼤会均于周⼆开幕。此外，SNOW和Kioxia的投资者⽇也将于周⼆举⾏。 Lots to get to this morning, so let’s get right to it… 今天早上有很多内容要处理，让我们直接进⼊正题…… NVDA: Jensen speech at GTC quick notes NVDA：⻩仁勋在GTC演讲的简要笔记 Key points are he said Vera Rubin is in full production, talked up CPU tailwind (VeraCPU 1.8x agentic sandbox performance vs. X86),and unveiled RTX Spark 3nm chipwith Mediatek which offers up to 1 petaFLOP of local AI performance and 128GB ofunified memory. 核⼼要点包括：他表示Vera Rubin已进⼊全⾯⽣产阶段；强调了CPU的助⼒作⽤（Vera CPU的智能体沙盒性能是X86的1.8倍）；并发布了与联发科（Mediatek）合作的RTX Spark 3nm芯⽚，该芯⽚可提供⾼达1 petaFLOP的本地AI性能和128GB的统⼀内存。 He also announced a new opensource model Nemotron 3 Ultra, which is 5x faster and30% cheaper than other leading open-source models including Kimi K2.6, Qwen 3.5,and GLM 5.1.Also said cost of AI DC will grow from$50B per GW to$80B per GW incoming years. 他还发布了⼀款全新的开源模型Nemotron 3 Ultra，其速度⽐包括KimiK2.6、Qwen 3.5和GLM 5.1在内的其他领先开源模型快5倍，成本低30%。此外，他还表示AI数据中⼼的成本将在未来⼏年从每GW 500亿美元增⻓到每GW 800亿美元。 NVIDIA highlighted tokenshave now become profitable, which is driving a significantdemand for compute. Jensen further noted GitHub commits have increased nearly 3xin 2026 due to AI Codingtools improving the productivity of SW engineers. NVIDIAargues this will result in hiringmore SW engineers as output per engineer is increasing NVIDIA强调Token现在已经实现盈利，这正驱动着对算⼒的巨⼤需求。⻩仁勋进⼀步指出，由于AI编程⼯具提⾼了软件⼯程师的⽣产⼒，2026年GitHub的提交量增加了近3倍。NVIDIA认为，随着每位⼯程师产出的增加，这将导致企业雇佣更多的软件⼯程师。 CDNS +7%: Cadence Unveils Industry’s First FullyAutonomous Virtual Engineer for Chip Design, poweredby NVIDIA CDNS +7%：Cadence发布业界⾸个由NVIDIA提供⽀持的芯⽚设计全⾃动虚拟⼯程师 PR: At Computex 2026, Cadence (Nasdaq: CDNS) announced the industry’s first fullyautonomous virtual agentic AI design engineer, extending the ChipStack™ AI SuperAgent to Level-5 autonomy. Built on Cadence’s AI-driven electronic designautomation (EDA) portfolio withNVIDIA Nemotronmodels, and secured byNVIDIAOpenShellruntime, the new agentic capabilities enable customers to run dynamicsimulations in automated workflows 在Computex 2026上，Cadence（纳斯达克股票代码：CDNS）发布了业界⾸个全⾃动虚拟代理AI设计⼯程师，将ChipStack™ AI超级代理扩展⾄Level-5级⾃主。该新型代理功能基于Cadence的AI驱动电⼦设计⾃动化（EDA）产品组合及NVIDIA Nemotron模型构建，并由NVIDIA OpenShell运⾏时提供安全保障，使客户能够在⾃动化⼯作流中运⾏动态仿真。 The company noted ChipStack can reduceverification cycles from weeks to hours(~40% reduction). NVIDIA noted it will hire 100sof thousands of ChipStack Agents 该公司指出，ChipStack可以将验证周期从数周缩短⾄数⼩时（缩减约40%）。NVIDIA表示将雇佣数⼗万个ChipStack智能体。 INTC: Intel targets new AI data centre chip by year endINTC：英特尔计划在年底前推出新型AI数据中⼼芯⽚ FinancialTimes:⾦融时报： Intel plans to ship an AI chip by the end of this year that uses cheaper memory andcooling technology than rival offerings from Nvidia and AMD, as the US chipmakerseeks to capitalise on a sharp turnaround in its fortunes. 英特尔计划在今年年底前出货⼀款AI芯⽚，该芯⽚采⽤⽐竞争对⼿Nvidia和AMD更廉价的内存和冷却技术。这家美国芯⽚制造商正寻求利⽤其业绩⼤幅好转的契机。 Kevork Kechichian, who leads Intel’s data centre group, told the FT that thecompany is “starting with the basics” as it tries to challenge its rivals in thebooming market for semiconductors th