🤖 AI Infrastructure & Semis

• Meta selling compute is not a sign of excess capacity. The better read: enterprise demand is strong enough that buyers are willing to pay high prices, and selling compute may offer better economics than internal use. Bullish angle: more diversified revenue and a shorter payback cycle. 1

• $TSM growth looks under-modeled. Revenue is projected to grow roughly 35% in 2027 versus 30% consensus, with AI data center revenue approaching $100B. Key demand drivers: AI CPUs, networking, HBM base dies, accelerators, and CoWoS. 2

• TSMC June numbers show acceleration, not just strong headline growth. June revenue hit 4426.8亿元新台币, up 67.9% YoY; 2026H1 revenue reached about 2.40万亿元新台币, up 35.6% YoY. The gap between 35.6% H1 growth and 67.9% June growth implies revenue growth already stepped up sharply from last year’s second half. 3

• CXMT closing the gap with the memory Big 3 just by reaching 1c-class 3D DRAM is too aggressive. Performance is not the only hurdle, and the idea that it could smoothly sell into Western markets looks like a stretch. 4

🧠 Mega-Cap Tech

• $AAPL is trading like its own beast. It was $1 away from fresh ATHs, and the tape suggests it can ignore macro and broader-market pressure better than most names. 5

• $MSFT has one of the cleaner AI monetization paths among hyperscalers. FCF is expected to nearly triple from ~$55B in 2027 to ~$165B by 2030 as Copilot scales across the existing enterprise base. The key edge: turning heavy AI infrastructure spend into high-margin AI revenue. 6

📈 Trading & Market Structure

• Gap-down opens often offer better setups than gap-up opens. It feels counterintuitive at first, but after enough screen time, the edge becomes clearer. 7

• This market still looks range-bound and volatile. With earnings, inflation, oil risk, and other catalysts stacked together, preparation matters more than prediction. 8

• Chasing high-beta names with leverage is a risk-management problem before it is a return problem. Before hitting buy, the one-day puke scenario must already be priced into the decision. If that drawdown is unacceptable, the trade size is wrong. 9

• SPX 6900 by August OpEx is being framed as a high-convexity upside bet. Index options could pay 600-900% if that type of move lands. 10