⚡️@octopusycc 对话 168X：AI 半导体进入困难模式：MU、GLW、Meta Compute、光通信与敏捷交易 这次我们请到 168X 的老朋友大老师，一起聊最近 AI supply chain 里最难做的几条线！ MU 暴力洗盘后，到底是深度回调，还是趋势转弱？GLW / 康宁从光互连核心叙事走到高估值区间，什么时候该继续拿，什么时候该止盈？Meta 出售 AI compute 的新闻，是真正的算力过剩，还是别有他意？ 这场 Space，我们会聊 MU、GLW、Meta Compute、GB200 / Rubin、光通信、DOCN，以及 AI 核心股降温后，资金是否正在轮动到 SaaS、防御、工业和其他板块？ 大老师善于用产业逻辑、期权结构、资金流和技术位做敏捷交易，我们会来实战拆解 AI 半导体进入高波动阶段后，如何加仓、止盈和风控！ 主持：@168MrZ @vcmktasa 嘉宾：@octopusycc 直播时间：7/7 15:00 东八区 X Space 链接：

一個國小四年級就輟學去做家具的木工，最後靠「抽屜滑軌」，一路滑進 AI 供應鏈，成為台灣首富。 川湖創辦人林聰吉早年就是做家具出身。 1970 年代開始做家具五金，1986 年成立川湖，最主要的產品就是大家家裡抽屜旁邊那兩條不起眼的「滑軌」。 真正的轉折發生在 2000 年。 當時中國製造崛起，川湖一個佔營收約 40% 的家具大客戶，直接要求他們降價 30%。 林聰吉沒有跟著打價格戰，而是選擇把這個大客戶丟掉。 結果隔年營收幾乎腰斬。 但也因為這次危機，川湖開始思考： 既然都是做「滑軌」，為什麼一定只能裝在家具裡？ 於是他們把幾十年累積的金屬加工、模具、機構設計能力，轉去做難度高很多的「伺服器滑軌」。 2000 年底，康柏原本的供應商出問題，緊急找到川湖。 川湖只花大約 一週就交出設計、通過測試，拿下第一張約 10 萬美元的試產訂單；後來 IBM、Dell、Sun 等科技大廠陸續成為客戶。 二十多年後，當 Server 變成 AI Server，這個當年「被迫轉型」的決定直接吃到史上最大一波算力建設。 因為今天一台塞滿 GPU 的 AI Server，重量、價格、散熱與維修需求都比以前更誇張。 GPU 再貴，最後還是要靠兩條滑軌撐住。 川湖現在累積超過 3,000 項滑軌專利，打進 NVIDIA 與大型 Cloud Service Provider 的供應鏈；本土券商甚至推估，川湖在 NVIDIA 新世代產品的滑軌市佔率超過 70%。 所以川湖其實不是「從家具突然轉型做 AI」。 它做的東西，從頭到尾都是滑軌。 只是： 家具抽屜 → Server → Cloud → AI Server 同一個核心能力，市場價值完全不同。 這也是我最近研究 AI Supply Chain 覺得最有趣的地方。 大家都在找下一個 NVIDIA、下一個台積電， 但 AI CapEx 真正爆發後，可能有一大批以前根本沒人注意的「傳統產業」，突然因為自己掌握了某個不可取代的零組件，變成 AI 公司。 AI 最大的機會，不一定是發明 AI。 也可能只是把你做了 40 年的東西，賣給 AI。 🚀

#为什么是中国# #WhyChina# The Global Logic of China's Economic Growth in the First Half of 2026: A 4.7% GDP Increase 2026年上半年GDP增长4.7%：中国经济增长的全球逻辑 According to the latest semi-annual report, China's gross domestic product (GDP) reached 69.6 trillion yuan, representing a year-on-year increase of 4.7% at constant prices, in line with the annual growth target. Against a backdrop of intertwined international complexities and volatilities, China's economic performance is commendable. Yet, a closer look at the data reveals that the logic underpinning China's economic growth is being reshaped. (I) At the Industry Level, Notable Highlights Emerge: First, new quality productive forces are being cultivated and strengthened at an accelerated pace.** In the first half of the year, industrial production grew robustly. The value-added output of the equipment manufacturing sector increased by 9.3% year-on-year, and that of high-tech manufacturing grew by 13.3%, both outpacing the overall growth rate of industrial output above a designated scale. This underscores a clear trend toward a high-end, intelligent, green, and integrated industrial structure. Looking at specific products, the output of 3D printing equipment, lithium-ion batteries, industrial robots, and other products emblematic of new quality productive forces surged. The average daily token call volume has reached hundreds of trillions, showcasing the vitality and potential of the digital and intelligent economies. Second, new growth drivers are accelerating to take on a leading role. Preliminary estimates suggest that new growth drivers—encompassing high-end manufacturing, the digital economy, and modern services—contributed over 40% to economic growth in the first half of the year. The economy's distinct shift toward a higher quality and more optimized structure is evident, and this overall trend is accelerating. For instance, industries related to artificial intelligence, such as integrated circuit manufacturing and intelligent vehicle equipment manufacturing, have all maintained high growth rates exceeding 30%, vividly illustrating the pace of China's industrial upgrading. Third, confidence on the investment front is on the rise. In the first half of the year, investment in high-tech industries grew by 4.6% year-on-year. Notably, investment in the manufacturing of aircraft, spacecraft, and equipment, computer and office equipment manufacturing, and information services grew by 23.3%, 8.1%, and 15.5%, respectively. Investment structure best reflects market expectations. The increasing "new economy content" in investments signals an acceleration in the replacement of old growth drivers with new ones. Investment in intellectual property products grew by 9.4%, indicating that enterprises are placing greater emphasis on R&D and innovation. This suggests that technological advancement is not simply about capacity expansion but about qualitative change driven by innovation. (II) Observing a Major Economy Requires Looking Beyond the Immediate Figures to the Long-term Trajectory. Behind the "new economy content" of the semi-annual report lies China's ongoing transformation from a global manufacturing hub to a global center of innovation. At the 17th Annual Meeting of the New Champions (Summer Davos), observers noted a new phenomenon: a host of unicorn companies are heading to China. They are establishing R&D centers, regional headquarters, and deeply integrating into China's innovation and industrial chains—shifting from "produced in China" to "created in China." So, why China? Economist Justin Yifu Lin, in his book *Demystifying the Chinese Economy, touched upon the theory of the "speed of technological change." He argues that the essence of the industrial revolution is not just the application of new technologies, but more fundamentally, the ever-accelerating pace of technological change. Since the mid-18th century, starting with the steam engine reshaping the textile industry, the snowball of technological change has grown, rapidly sweeping through industries like chemicals and automobiles, ultimately redrawing the geographical map of great power competition. Looking at China today, the trajectory of accelerating technological change is equally clear. A leading enterprise can drive an entire industry, which in turn can boost a whole region. These burgeoning industrial clusters, growing from saplings to forests, not only enhance production efficiency and invigorate market vitality but also effectively improve development quality and resilience. For example, specialized and sophisticated "little giant" enterprises above a designated scale in Beijing, through deep cultivation of innovation chains, supply chain collaboration, and international expansion, have become "connecting points" and "accelerators" for the dual circulation strategy. More importantly, emerging industrial clusters possess powerful spillover effects. The rapid rise of new energy vehicles is not only reshaping the automotive industry but also driving transformations in chips, software, and energy networks, allowing more sectors to gain value from efficiency improvements. The swift advancements in AI and biomedicine are sparking a "gentle qualitative change" in people's livelihoods, significantly enhancing the sense of fulfillment and well-being through smarter, more affordable products and more livable environments. (III) Looking from the First Half to the Full Year, China's Development Momentum Remains Positive. Of course, during this critical period of transitioning between old and new growth drivers, China's economy still faces lingering issues and new challenges. Some core areas are still grappling with "bottleneck" technologies, certain high-tech industries face external risks of "decoupling" and supply chain disruptions, and "involution"-style competition affects the new energy market ecosystem. However, most of these are issues arising from development and transition, and they can be addressed with effort. The supporting conditions and fundamental trends for long-term economic improvement remain unchanged. China's economic journey toward a newer, higher-quality model is itself a process of encountering new problems and solving them along the way. By maintaining confidence, proceeding steadily, and balancing both qualitative improvements and quantitative growth, China's industries are poised to be brimming with dynamism, and the Chinese economy will continue to advance steadily and sustainably. #China# #Jiangxi# #JiangxiEconomy# #世界经济看中国# #赣出新精彩#

一篇不错的解读：《META出租H100与购买先进算力并不矛盾》 Meta 做 NeoCloud 与继续租 Crusoe 1.6GW，并不矛盾 昨天盘前，Meta 被报道正在考虑把多余 AI 算力对外商业化，甚至做成类似 NeoCloud 的业务。市场第一反应非常剧烈：Meta 盘前上涨接近 6%，但 AI算力和 neocloud 相关股票则受到负面 Narrative 影响, 市场担心的是：如果 Meta 也开始把 GPU 算力对外卖，是否会直接导致算力过剩？ 这个反应可以理解，但我们认为市场把问题想简单了。 首先，Meta 这件事本质上不是“AI 算力需求见顶”，也不是“Meta 不需要继续买算力”。相反，Meta 同时还在继续锁定非常大规模的新算力。根据 Bloomberg/Reuters 报道，Meta 最近与 Crusoe 签署了新的 AI computing capacity 协议，将从 Crusoe 位于 Texas Childress 和 Missouri Warrenton 的两个数据中心获得合计约 1.6GW 的容量。 同时，Meta还在向其他Neocloud购买算力。我们在去年3Q25 META Preview中就提到过META正在向NeoCloud寻求购买3GW算力。 所以表面上看，这里确实有一个矛盾：如果 Meta 自己已经有多余算力，为什么还要继续向 Crusoe 租 1.6GW？ 我们的理解是，这不是矛盾，而是算力代际切换。 过去两年，Meta 已经采购和部署了大量 H100/H200。这些 GPU 不是没价值，恰恰相反，它们对 inference、fine-tuning、企业模型服务、图像/视频生成、传统 ML workload 仍然非常有价值。但对于下一代 frontier model training，尤其是 3T+ 参数规模的 MoE、长上下文、多模态和 RL-heavy post-training，H100/H200 的训练经济性会明显下降。 关键不是 H100 不能训练，而是单位有效 token 成本变差。 当模型进入 3T+ 规模后，瓶颈不再只是单卡 FLOPS，而是 HBM 容量/带宽、GPU 间通信、scale-up 网络、checkpoint/restart、expert routing、sequence parallel、pipeline bubble、以及大规模 collective communication。H100 集群当然还能跑，但训练 wall-clock 更长，通信开销更高，集群利用率更难维持，最终表现为同样训练一个 frontier model，成本和时间都不如 GB200/GB300，未来更不如 Vera Rubin。 因此，Meta 现在面对的是一个很典型的资产配置问题： 最先进的 GB200/GB300/Rubin，要优先留给下一代模型训练；上一代 H100/H200，则应该尽量转成 inference 或外部商业化收入。 这也是为什么“做 NeoCloud”和“继续租 Crusoe 1.6GW”可以同时成立。 Meta 继续向 Crusoe 锁定 1.6GW，本质上是在为更长期、更先进、更大规模的 AI infrastructure 做准备。这种资源对于 Meta 来说，更多是未来 GB200/GB300/Rubin 时代的战略性产能，而不是简单补 H100 的缺口。 另一方面，Meta 既然已经买了大量 H100/H200，就不可能让这些资产在 frontier training 代际切换后闲置。Meta 内部当然有广告、推荐、内容排序等大量推理 workload，但这和 OpenAI/Anthropic 那种直接面向外部客户卖 token 的 LLM inference 业务并不完全一样。Meta 如果没有足够多可以直接 monetization 的外部 token demand，把 H100/H200 做成 cloud capacity 或 hosted model API 对外销售，是非常合理的资本回收方式。 这其实和 xAI / SpaceX 的思路有相似之处。xAI 今年公开宣布与 Anthropic 达成 compute partnership，向 Anthropic 提供 Colossus 1 算力；xAI 官方称 Colossus 1 包含超过 22 万张 NVIDIA GPU，包括 H100、H200 和 GB200，并可支持 training、fine-tuning、inference 和 HPC workload。(xAI) 这说明即使是 frontier AI 公司，也可能把一部分已有 GPU fleet 对外出租，同时把最新、最稀缺、训练效率最高的下一代集群保留给自己的 frontier model。 所以今天市场担心“Meta 进入 NeoCloud 会打垮所有 NeoCloud”，我们觉得有些过度。 更准确的判断应该是： AI 算力市场正在从单一的 GPU shortage，进入多代 GPU 分层定价和分层使用阶段。 第一层是最新训练算力：GB300、Rubin，以及后续更大 scale-up domain 的系统，主要服务 frontier model training。这部分供给仍然稀缺，客户仍然会向 Crusoe、CoreWeave、Nebius、Oracle、Microsoft 等各类供应商锁产能。 第二层是上一代高端算力：H100/H200/部分 GB200，更适合 inference、fine-tuning、enterprise AI、hosted model、agent workload 和中小模型训练。这部分不是没有需求，而是从“最稀缺的训练资源”变成“可以规模化商业化的推理资源”。 第三层是更通用的 GPU cloud 和 long-tail enterprise workload，对价格更敏感，但需求弹性也更大。 在这个框架下，Meta 的行为其实很合理：它不是停止建设 AI infrastructure，而是在把不同代际的 GPU 放到最适合的经济用途上。 因此，我们不认为这是 AI infrastructure 的大问题。真正重要的判断是：下一代 frontier model training 对 GB200/GB300/Rubin 的需求仍然非常强；同时，H100/H200 这类上一代 GPU 也不会被废弃，而会进入 inference monetization 和外部算力销售阶段。 这对整个 AI supply chain 的含义反而是： GPU fleet 开始变成多代际资产，而不是一次性训练工具。旧 GPU 不归零，新 GPU 继续稀缺。Meta 做 NeoCloud，不是需求崩了，而是算力资产终于开始金融化和商业化。