Original Insight Summary

🧠 AI Commercialization & Model Economics

• AI commercialization is moving beyond closed-model ARR. Closed-model ARR from Anthropic/OpenAI still matters because it shows willingness to pay for model capability. But the cleaner forward indicator may shift to the earnings performance of hyperscale CSPs, as infrastructure-scale monetization becomes the bigger scoreboard. 1

• Large-model commoditization is accelerating. The key watch is whether SpaceXAI/Cursor Grok 4.5 can avoid under-delivering after heavy expectations. The setup implies model capability alone is becoming less differentiated; execution, speed, token efficiency, and cost matter more. 2

💾 Semiconductors & AI Infrastructure

• Micron is not cheap enough yet. After Samsung’s earnings hit sentiment, Micron fell 5% pre-market and returned to $930. Even with forward PE back to single digits, the preferred add zone is still $850, around the 100-day moving average. The perceived floor price remains $700, so there is still downside room before a stronger dip-buy setup. 3

• China’s AI chip push may be signaling progress in domestic HBM. If DeepSeek and Zhipu are considering their own chips, that could imply China’s HBM self-sufficiency has already reached a meaningful level, or is at least on track. The bigger second-order question: the EDA moat may also be starting to crack. 4

• EUV supply is still the bottleneck. The 2027 capex estimate of US$73bn, below consensus US$80-85bn, is likely capped by EUV supply. Translation: demand may not be the constraint; tool availability is. 5

• Advanced packaging competition is forcing TSMC to move faster. Intel EMIB pressure could push TSMC to accelerate CoPoS adoption to 2028, potentially with Feynman. The reported use of D2W for GPU stacking and a SiC carrier points to packaging becoming a sharper battleground for $TSM and $INTC. 6

💻 Software Rotation & AI Beneficiaries

• Software is splitting hard, but the laggards may become the next dip-buy trade. The setup is fund rotation: when profit-taking comes out of DRAM and semis, weak software names could get bought as Buy the Dip targets. The focus is not the winners, but the beaten-down losers with rebound optionality. 7

• Microsoft, Palantir, and Zeta are positioned as AI beneficiaries, not AI victims. The market is worried AI will replace software-company revenue, but these three names are framed as companies eating the AI dividend instead. In the stack, conviction and market-cap ranking are Microsoft, Palantir, then Zeta. For Microsoft, fiscal 2027 has already become the valuation lens, with a reasonable price range of 550-600. 8

• Zeta’s story is shifting from marketing software to AI infrastructure. The key upside surprise is Palantir integration. The market expected Foundry integration to drag into year-end or even early next year, but progress appears faster than feared. That changes the narrative from ordinary marketing tech to AI infra leverage. 9

🛢️ Geopolitics, Oil & Risk Appetite

• The U.S.-Iran flare-up looks like volatility, not a systemic risk event. Trump saying the previous ceasefire is over adds pressure to an already fragile market. Base case: renewed friction creates short-term swings, but not a sustained escalation path. The key watch is how quickly both sides return to de-escalation. 10