Morgan Stanley has analyzed the growing interest in application-specific integrated circuits (ASICs) within the AI industry, highlighting their advantages in specific scenarios but also emphasizing their limitations compared to commercial GPUs. ASICs, while cost-effective in development, incur higher system and software deployment costs, making their total ownership cost potentially higher than that of GPUs. NVIDIA's (NVDA, Financial) CUDA ecosystem remains robust and widely used in global cloud services, maintaining its solid market position.
Recently, the AI trading momentum has shifted towards custom ASICs, perceived to have growth potential surpassing commercial GPUs. This shift has stalled NVIDIA's stock growth and affected AMD's performance. However, Morgan Stanley believes that expectations for ASICs are overly optimistic, and these chips will struggle to displace GPUs in the long term. While ASICs excel in narrow applications and offer performance advantages, their deployment costs are higher, limiting their broader market appeal.
NVIDIA continues to optimize its GPUs to regain market share, particularly in cloud computing, where commercial GPUs outperform ASICs. The firm's significant investments in R&D, projected at $16 billion this year, and its mature ecosystem give it a competitive edge. Despite the allure of lower-cost processors, the lack of a mature ecosystem often leads customers back to NVIDIA's high-performance graphics cards.
