X（推特）上的财经、科技与宏观账号
.@JensenHuang wrapped up his Paris visit at the Élysée with President @EmmanuelMacron, discussing NVIDIA’s investments in France’s AI ecosystem, open source, and AI safety. Our road to #NVIDIAGTC Berlin continues. https://t.co/CGYw6B8UFl
Try @Starlink Internet! 引用 @cb_doge：BREAKING: Starlink ranks #1 for both download speed and Consistent Quality in rural UK, beating EE, Three and Vodafone in Opensignal’s study. Starlink also leads rural download speeds in all 14 European markets studied. https://t.co/Cp9Hcw31mW
Starlink brings Internet access to remote places 引用 @cb_doge：BREAKING: Vietnam’s Hai Phong government partners with Starlink to bring FREE internet to remote islands. The pilot includes 10 Starlink terminals, with FPT handling deployment and technical support: 6 in Cat Hai and 4 in Bach Long Vi.
All Grok @Bots get their own email address and can help analyze and summarize your email box too 引用 @ericzakariasson：grok @bot 0.70.2 is rolling out! - each bot can claim its own email address at @mail.grokbot.com and draft emails for you to send - apps like Figma, Notion and Canva show up as live
Assign Grok @Bot to your phone button for instant access 引用 @liam_fallen：I gave Grok Bot its own button → Press the Action Button → Open Grok in voice mode → Give it tasks and deadlines → Grok delegates them to Bots I feel like James Bond. https://t.co/yPX9CViyQy
updating a bit on the labs (Anthropic and OpenAI to be specific) full piece: https://t.co/5e75aXG0zy https://t.co/U6DXfJR9sM
Cushman & Wakefield 的《2026指南》新增了关于建设成本、劳动力、设备交付周期、通胀和选址的外部参考。现代绿地设施平均建设成本为每兆瓦 1760 万美元，不含芯片和 GPU；每兆瓦建设成本自 Q4 2024 以来上涨 21%。
Bloomberg puts a theoretical 1GW AI data center in Abilene at ~$30bn. Its ten cost categories actually sum $17.1bn for chips and servers, and ~$13bn for everything else. My read is that the hardware assumption is likely closer to Hopper-era economics vs. the latest generation. https://t.co/KjTa26XLj
Thought about this more: Open weights could support attractive INFRA returns even as LAB margins compress. If training is funded elsewhere and open models deliver enough value, inference providers can charge less for tokens while retaining a spread over compute costs. 引用 @ShanuMathew93：This is a b
图表分析不同数据中心容量假设对实验室收入和支出的含义，依据包括公司披露、公开报道和示例情景。文中引用的估算显示，2026年美国数据中心增长规模约为10—21 GW（按IT口径），不同估值跨度超过两倍。
英伟达黄仁勋表示，将在一周内宣布20亿美元的欧洲投资，过去一年已投资80亿美元。黄仁勋在巴黎点名四家法国初创公司；英伟达今年前四个月承诺投入超过400亿美元用于AI股权投资。
Guess like we found out why Elon has been going on a crusade against memory in his recent posts. $SKHY and Samsung have been price hiking memory too much? https://t.co/r3sTYkd7TX
这两天美股一个迹象就是部分资金开始流切往价值股、医疗保健、金融、公用事业板块，一方面是防守另一方面也有上周五这里https://t.co/HyuQZKyn26聊“关注美债收益率筑顶的可能性”，收益率筑顶回落，之前一个多月被压制的这些板块都能往上冒头。昨晚纳指、标普跌，道指小涨也是这个逻辑。正好跟美银的报 https://t.co/Zmj26DGTpA 引用 @qinbafrank：现在看美银上周这份《FLOW Show》时机很好，Hartnett认为高利率已经开始影响市场内部结构，而中期选举可能进一步改变盈利、资本开支和监管预期。在这个窗口里，现金有收益，投资者也就有耐心等待。他给的建议是：
From satellites to skincare, France is building with NVIDIA. 🇫🇷 Developers, researchers, and businesses across France are using NVIDIA technology and open models to bring AI into the real world. Explore France’s AI ecosystem: https://t.co/sL6GVrYOE0 Meet two of the people https://t.co/A2tlwIUgZ
AVGO和INTC在DBS会议材料中表示，AI领域的突破将使芯片投资需求至少在未来两年保持旺盛；发言还称当前供应不足以满足市场需求。引用内容提到韩国市场的$SKHY和三星下跌，并批评AVGO CEO称AI投资者是赌徒、企业AI尚未带来投资回报。
Meta发布Muse Wonder和Muse Wonder Full两款模型，官方称其在个人和专业智能体任务上达到行业前沿水平，这是Meta超级智能实验室（MSL）去年重组以来推出的第一款前沿大模型。Muse Wonder即日起通过Meta Model API和编程智能体Muse提供。
每隔一段时间我可能都得重复一下我自己的理念，仅供参考。 板块，个股的基本面不值得天天聊，短期除非重大事件，不然不大会有基本面改变，价格的波动会出现一些减仓加仓的机会。 引用 @adakk2716701：@TJ_Research talk君，请问cpu您目前怎么看，amd和arm
Correlation vs causation vs intent - as we argue in the #GenevaReport, it is mostly about the inflation differential driven by China’s housing bust vs the West inflation surge, at the time of rapid tech progress. A Reverse Dutch Disease that benefits China @cepr_org https://t.co/XCDNy6qrki 引用 @Brad
To quantify how the models handle unfamiliar tasks, we compared test-time compute scaling on tasks trained on during post-training against other tasks. Both models improve similarly out-of-distribution, but given the same in-distribution score, Muse Wonder Full performs better https://t.co/SES4eMwL
Two things make Muse Wonder suited for long-running, multi-step work an agent faces every day: -> Active Memory carries your context across conversations and surfaces it when it's relevant. -> Active Focus keeps the models working toward your goal when a task gets challenging, https://t.co/4t3lIYq
Great to see what the Fin team has built with NVIDIA Nemotron 🙌 They fine-tuned Nemotron 3 Super and Nano for their new Fin Apex customer service models. Congrats on the launch! 引用 @fergal_reid：We’re pleased to announce the launch of our 2.0 series of Fin Apex models! We think these are the bes
Over the last few months, we’ve made significant progress in reducing the effectiveness of distillation attacks against Claude. In this post, we explain how: https://t.co/uWiRSLBuOS
US has ~2x China's data center capacity, and 3x+ Europe's. Let's not lose that competitive advantage. https://t.co/jGw5qEHFYH
Something that blows my mind is the variance of the estimates we're still seeing for data center capacity. We’re already in mid-October and the selected estimates for 2026 US data-center growth still span roughly 10–21 GW on an IT basis. A range more than 2x for the year we’re https://t.co/VL1dm0sQw
Why I believe Muse (and Muse-like agents) are game changers for SMBs $META full piece: https://t.co/rQBpOpGBKj https://t.co/VbDyCAJj9J
Meta超级智能实验室成立一年多，推出Muse Wonder前沿模型家族，面向个人和专业智能体工作，性能达到最先进水平。
Super interesting: "Anthropic's Claude Fable 5.1 scored 41% in Claude Code and 61% in OpenAI's Codex. It made 28 tool calls at the median in one and 66 in the other." "No model did best in its maker’s own harness" https://t.co/BONmxl9hiE 引用 @chamath：Model Benchmarks Are Worthless How can a model
Love people opening up the convo to debate/learn in public and Matt is great at it. Sharing for broader awareness. I'll leave the chip-based GW forecasts to others, but on power, I've presented several frameworks for thinking about US IT capacity and the very real limits that https://t.co/fTp3ZzdW5
Great episode. Meta's $100B+ AI bet is built around PERSONAL superintelligence: an agent that knows your context, remembers it and gets things done for you. Zuckerberg and Wang on Cheeky Pint: >They see room for a consumer-focused frontier lab while much of the industry's 引用 @collision：Cheeky Pi
作者称，DTE与Google的交易符合大型负荷客户承诺期限延长、最低付款要求提高的更广泛变化。文中列举AEP Ohio的安排为爬坡后8年、最长4年爬坡期，且该规模客户需求底线为85%；Dominion的安排为14年，输配电需求底线为85%，发电需求底线为60%。
DTE与Google在Van Buren Township签订1 GW、期限为20年的电力合同。作者称，密歇根州10月1日的相关命令披露了超大规模云服务商的最低电力承诺，并提到为公用事业提供下行保护的基础设施投资安排。
媒体称微软自研 Maia 200 AI 芯片已投产，下一代 Maia 300 计划明年开始制造。Maia 300 的计算芯粒将由台积电采用 2nm（N2）制程制造；原文末句关于微软部署计划未完整。
Sure... Random train of thoughts on markets now, optics like $LITE / memory + $IREN / Anthropic. On Memory: - Demand visibility on $MU / Samsung / SK Hynix extend into 2030-2031. Samsung has LTAs covering 70% of their capacity into 2031 (likely with SK Hynix type price 引用 @Wolfex_Yeat：@aleabi
高通CEO: “什么是AI手机？当OPENCLAW面世时，AI手机的未来就清晰了” 💯💯💯 引用 @TJ_Research：以前我们买手机看哪家的照相功能好，未来我们买手机看哪家的端侧AI强
The latest on Apple’s HomeView, new Apple TV and HomePod mini ahead of Tuesday. https://t.co/4vuopByarz 引用 @markgurman：Power On: Apple spent a decade losing the smart home. Now it has a plan to compete: HomeView, a new Apple TV and HomePod mini, and LG-built accessories. Tuesday’s unveiling is abou
The AI boom is increasingly shifting from parabolic to linear growth. Most investors are totally unprepared for that shift. This week, free on the blog app. https://t.co/EUji9tuxSA https://t.co/qQk0lvaJcR
2000年互联网泡沫破裂，科技股持续了很长时间的低迷。核心不是泡沫，而是2000年前后的互联网都没找到商业模式，从00年一直到06、07年 引用 @market_sleuth：“Too big to fail. Active trading isn’t needed anymore. In fact it’s more of a waste of time.” Those were the words I heard as I was speaking with a finance person yesterday who is way smarter than I & who mana
I hear all the time that the semis & hyperscalers see plenty of AI-related demand ahead, with no connection to who will pay to meet those lofty expectations. In short, it would require all non-tech companies to bet their entire free cashflow on AI by 2030. Good luck. https://t.co/tllPSskzj8
2/ Her take: Yes, rate hikes won't open the Strait of Hormuz or immediately slow the data-center buildout. But they might make other firms "think twice" about expanding in this environment, leaving scarce parts or supplies to data centers and easing pressures at the margins.
红杉两年前提出的6000亿美元问题，如今已变成3万亿美元问题。相关支出规模扩大之际，实际收入仅增加略多于1000亿美元，前瞻性需求与现实收入之间的差距已扩大。