Morgan Stanley: GPT-6 Astra Reshapes AI Demand, Compute Remains the Top Priority Allocation Focus

TechFlow reports, according to Chaowang Research, Morgan Stanley's September 7 research note indicates that the launch of GPT-6 Astra is bolstering the AI infrastructure investment narrative. Training large models on 1 million GPUs and extending them to broad capabilities such as inference, engineering, and physical world tasks will enlarge the AI addressable revenue pool. Morgan Stanley argues that the risk of underestimating AI investment currently exceeds that of overestimation. After investors drastically reduced AI holdings, the market correction offers a window to reposition.

Morgan Stanley segments AI investment into three verticals: AI compute is the highest conviction idea, storage offers structural opportunities but demands greater selectivity, and networking ranks as a secondary option. On the compute front, TSMC has raised its 2026 revenue growth guidance to over 40%, with AI semiconductors expected to account for more than 30%. ABF substrates and MLCCs are key derivative beneficiaries, with ASE Technology and Murata named as top picks in their respective segments. The memory cycle is moving into its later stages, as manufacturing paradigms shift due to the increased complexity of HBM4E backend processes, leaving ChangXin Memory Technologies as the sole overweight storage exposure. Networking catalysts center on rapid scale-up expansion (addressing a market exceeding $70 billion), while mass CPO adoption is not forecasted until after 2029; Corning, Lumentum, and Coherent are highlighted. Analog semiconductors are in the early recovery phase, with STMicroelectronics, NXP Semiconductors, and Renesas Electronics offering non-AI diversification channels.