Custom AI chips from AVGO and MRVL target cost performance gains over Nvidia GPUs in data center workloads.
Broadcom (NASDAQ: AVGO) and Marvell Technology (NASDAQ: MRVL) are expanding into custom AI chips, aiming to outperform traditional GPUs in cost efficiency. Both firms serve data center markets but now focus on application-specific integrated circuits (ASICs) for AI workloads, which promise superior performance at lower costs.
Nvidia (NASDAQ: NVDA) currently dominates AI training with GPUs, but their flexibility is underutilized in single-workload environments. Custom ASICs, designed for specific tasks, could deliver better power and cost metrics, driving adoption in data centers.
The shift toward custom AI chips may accelerate as demand for optimized, cost-effective solutions grows. Analysts highlight cost performance as the key differentiator between AVGO and MRVL in this emerging segment.