Marvell Technology vs. Broadcom: Which Custom AI Chip Stock Has More Upside?

Graphics processing units (GPUs) have been the go-to chips for hyperscalers and artificial intelligence (AI) companies in the past three and a half years. That's not surprising, as GPUs pack massive parallel computing power, making them ideal for carrying out vast calculations simultaneously that are needed for AI model training and inference.However, GPUs have been losing ground to another type of chip known as application-specific integrated circuits (ASICs). These ASICs, popularly known as custom processors, are designed for specific tasks. As a result, they differ from GPUs, which are general-purpose chips. Since ASICs are custom-made to perform a specific task, they are reportedly 30% to 40% more power-efficient than GPUs while offering better performance.Not surprisingly, custom AI chips are better suited for AI inference applications, which don't require the horsepower needed to train AI models. As a result, TrendForce expects sales of ASICs to increase by 45% in 2026, well above the 16% anticipated increase in GPU shipments.Continue readingWeiter zum vollständigen Artikel bei MotleyFool Weiter zum vollständigen Artikel bei MotleyFool