The 'Next Nvidia' Trade? Why Investors Are Suddenly Watching Advanced Micro Devices, Arm Holdings, and Marvell Technology - The Motley Fool

NVIDIA has dominated the AI chip market for years, controlling an estimated 80% share, thanks to its powerful GPUs ideal for training large language models. However, despite strong revenue growth, its... Read original →

The AI workload shift from training to inference is triggering a structural reconfiguration of hardware architectures. While GPUs excel in high-bandwidth, floating-point-intensive training, inference prioritizes energy efficiency and cost-per-operation—favoring AMD’s EPYC CPUs, Arm’s Neoverse platforms, and Marvell’s custom ASICs. Technologically, this accelerates adoption of chiplets, CXL interconnects, and near-memory computing, reshaping server SoC design. Geopolitically, tightening U.S. export controls push hyperscalers toward non-U.S. architectures like Arm to mitigate supply chain risks, raising NVIDIA’s compliance overhead. In response, NVIDIA may counter with Grace CPUs and GB200 superchips, but its CUDA moat weakens outside GPU-centric workloads. Over the next 12–24 months, heterogeneous computing will dominate AI infrastructure, with AMD and Marvell capturing over 30% of new deployments through customization. Arm, if tightly integrated with TSMC and packaging partners in Taiwan, China, could emerge as the pivotal disruptor.