Goldman Sachs Privorotsky's View on AI - FJElite

AI: Four stories to unpack. First, CXMT is a legitimate It concern for memory economics. Second, Microsoft’s Maia 300 custom chips, moving to TSMC’s 2nm process. Incrementally negative for merchant GPU suppliers (NVDA), positive for the custom silicon supply chain and aggregate compute consumption. Third, Hock l’an’s comments comparing A1 investors to gamblers have been taken wildly out of context. Both Broadcom and Intel reaffirmed exceptionally strong demand for another 2-3 years, potentially through 2030. Tan sees compute demand exceeding supply and explicitly acknowledged that frontier labs generating substantial revenue can afford expensive financing. LT ROl is a legitimate question, but hardly a revelation. Fourth, Meta’s Muse Wonder and Wonder Full look genuinely competitive at the frontier, potentially adding another serious player alongside OpenAI, Anthropic and Google. Some questions around benchmarking, but impressive nonetheless. Prediction markets also anticipating another major Google release imminently. What do I make of all this? Competition fuels competition, and competition requires more compute. The frontier keeps advancing, driven by Western closed source labs trying to outdo one another. Hardly bearish for aggregate compute demand, even if the economics shift around the supply chain. Think the legitimate concerns are around the ability to complete capacity/deliver power relative to expectations and that is fueling further de-rating in the capex vertical.