Open-weight models broaden demand for AI infrastructure despite compressing frontier-model margins. After DeepSeek R1, Tencent, Alibaba and ByteDance significantly increased Nvidia H20 orders, while smaller firms bought H20s for in-house use. Dell's AI customer count rose from over 5,000 to over 6,500 last quarter, roughly 30%, mainly on enterprise acceleration per management; its neocloud, sovereign and enterprise customers offer a read on self-hosting demand. x.com  x.com 

AI infrastructure bottlenecks could give eight stocks as much as 18x upside. Nvidia supplies frontier-model GPUs; Broadcom and Marvell supply custom silicon and networking; TSMC makes advanced chips; Micron supplies sold-out HBM; Vertiv provides power and cooling; Palantir monetizes enterprise AI; Alphabet spans TPUs, research, cloud and apps. The list tracks where AI infrastructure spending is flowing. x.com  x.com 

China's domestic market lets YMTC and CXMT threaten Micron, Samsung Electronics and SK hynix without entering US products. China receives 15-25% of smartphones and PCs and is on a path to at least 20-30% memory share over the next several years; current leaders are much too optimistic. History suggests at least a near-term semiconductor rally even if the AI trade has topped. It likely has not: agentic AI should give the trade room to run into 2027. x.com  x.com  x.com 

Baidu looks ready for a second leg higher as President Trump meets President Xi this week. It has built a five-year base, holds $43B of cash, grew AI Cloud 34% year over year and has a Kunlunxin AI chip spin-off coming. x.com 

China's entry into chip design and fabrication is a structural change: stop fading China as Western AI chip and cloud leaders get steeply discounted. China dominates nearly every manufacturing category; with multi-patterning its DUV tools perform near newer nodes and could scale massively or reach EUV. Chinese open models co-designed for Chinese chips could pull sovereignty-minded countries onto its tech stack. As Ford fared after China joined the WTO, so may Nvidia, ASML, TSMC and AWS by 2050. … Show more

Alibaba's full AI stack is not yet working economically: it loses $4 for every $1 of revenue, and that is the whole thesis risk. AI Labs lost RMB13.9B, cloud profit covers only 41% of that hole, and commerce is still funding the entire AI build. The cloud business works; the full AI stack does not yet. x.com 

Agentic AI burns 10-100x more compute per session than a chatbot, lifting the whole chip supply chain. Anthropic's ARR rose from $9B to $30B in five months. Alphabet's TPU v8, with Broadcom and Marvell exposure, runs through TSMC, the unavoidable foundry chokepoint, and CoWoS shortages make Intel's EMIB packaging relevant. Micron, SanDisk and SK hynix gain on on-package memory demand, and Applied Optoelectronics, Lumentum, Coherent, Sivers and POET on interconnect demand. x.com 

Alibaba is the cleanest public-market bet that Chinese AI labs are innovating rather than just distilling, and that cheaper models drive more usage. If China is innovating, Qwen drives Alibaba Cloud share; if Jevons paradox holds, Alibaba captures the spend across models, compute, data, chips and apps. AI products are already 30% of Alibaba Cloud's external revenue, above a $5B run rate, and the commerce cash engine can carry it through the capex/revenue mismatch. x.com 

Alibaba and Baidu stand to benefit because China will not slow its AI development even if the US slows its own. x.com 

Huawei's Ascend 950 suppliers gain as China's AI compute demand shifts from policy-driven buying to commercial absorption of capacity. Per the FT, Ascend revenue may rise 60% to $12B, with ByteDance, Tencent and Alibaba chasing the 950PR, and DeepSeek ties its V4 Pro price cuts to Ascend 950 supernode volume, implying tens of thousands of chips. HG Tech is a lead 800G optical module supplier, Shennan Circuits covers PCBs and substrates, Fastprint ABF substrates and Shengyi upstream laminates. … Show more