Apple is reshaping the AI hardware market with its "local inference" logic, challenging NVIDIA's potential incremental share in AI inference. The latest M5 Ultra Mac Studio began shipping on September 22, supporting up to 512GB of unified memory and capable of running distributed inference tasks in multi-machine clusters. Apple previously demonstrated four Mac Studios collaboratively running a trillion-parameter model, with its hardware chief Johny Srouji stating that local computing power might be more economical for teams with stable workloads than long-term cloud GPU rentals. Goldman Sachs analysts believe that if more inference tasks migrate to end devices or enterprise-owned machines, NVIDIA's share in incremental AI inference activities might be lower than market expectations. Therefore, they recommend expressing a bullish view on Apple and a bearish view on NVIDIA using options.