New leader board: Citi refreshes AI compute semiconductor picks - Yahoo Finance

Citigroup has refreshed its rankings of AI compute semiconductor stocks following the latest quarterly earnings reports, highlighting that demand for AI computing remains undersupplied. The bank empha... Read original →

Memory scarcity—not compute—is now the true bottleneck in AI infrastructure. Citi’s prioritization of DRAM allocation signals that HBM-integrated packaging has shifted from a competitive edge to a market entry requirement. NVIDIA’s deep integration with HBM3E via CoWoS cements its lead, while AMD leverages dual-sourced SRAM/HBM support from Samsung and Micron for supply resilience. Qualcomm’s LPDDR-based approach, constrained by limited allocations until 2029, reveals a fundamental mismatch for training workloads. Geopolitically, U.S.-led export controls are driving hyperscalers to lock in decade-long memory contracts, forcing chipmakers to treat supply chain security as a core technical spec. Over the next 18 months, IDM players or heterogeneous integration platforms—like Intel’s Foveros or TSMC’s SoIC—will dominate ecosystem control; logic-only designers without guaranteed HBM access risk exclusion from the high-end AI market.