NVIDIA Corporation (NVDA.US) CEO Jensen Huang sounds the horn! Prepare for the new battle in the CPU market, targeting Intel Corporation (INTC.US) and AMD (AMD.US).

Although NVIDIA Corporation's massive wealth is currently mainly built on specialized graphical processing units (GPUs) for artificial intelligence servers, its CEO Huang Renxun is increasingly showing favoritism towards general-purpose central processing units (CPUs). As the "brain" of computers in the traditional sense for decades, CPUs were previously synonymous with Intel Corporation (INTC.US) and sometimes associated with AMD (AMD.US). Huang Renxun has pointed out that in the past, 90% of computing tasks were handled by CPUs, with only 10% being completed by GPUs, but in recent years this ratio has reversed. However, as AI companies shift their focus from model training to model deployment, CPUs are once again being valued and even considered as a choice not inferior to GPUs. NVIDIA Corporation plans to play an important role in this transition. Huang Renxun stated during the company's fourth quarter earnings call, "We love CPUs, just as we love GPUs." He assured analysts that NVIDIA Corporation is not only prepared for the resurgence of CPUs, but its data center CPU products, first launched in 2023, will have capabilities surpassing those of competitors. At the Consumer Electronics Show held in Las Vegas last month, Huang Renxun further predicted that the usage of high-performance NVIDIA Corporation CPUs in data centers will experience explosive growth, stating "I wouldn't be surprised if NVIDIA Corporation becomes one of the world's largest CPU manufacturers." Battle between CPUs and GPUs For decades, CPUs and GPUs have each had their roles. CPUs, as general-purpose chips, are designed to handle various mathematical tasks at a reasonable speed that software programmers may assign. In contrast, GPUs specialize in executing a set of simpler mathematical operations but can perform thousands of them in parallel. In video games, this means calculating the values of thousands of pixels on the screen multiple times per second; in the field of AI, it involves multiplication and addition operations on large matrices of numbers used by developers to represent real-world data such as text and images. AI companies are increasingly deploying "agents" that can autonomously perform tasks like writing code, filtering documents, and writing research reports. Creative Strategies analyst Ben Bajarin pointed out that these computations are "increasingly, sometimes predominantly, being run on CPUs." He believes that the configuration of NVIDIA Corporation's flagship AI server NVL72 (with 36 CPUs and 72 GPUs), in handling the workload of these "agents," may evolve into a 1:1 ratio or even potentially not require GPUs at all. NVIDIA Corporation aims to prove its strength To showcase its ambitions in the CPU field, NVIDIA Corporation recently announced an agreement with Meta Platforms, which will heavily utilize NVIDIA Corporation's Grace and Vera CPU chips as standalone products. This marks a relatively new development compared to the current pattern of each CPU in NVIDIA Corporation's AI servers being paired with multiple GPUs. However, this does not mean that Meta is switching CPU suppliers but is more likely expanding its supplier mix. Just days later, AMD also announced a large deal with Meta involving its CPUs, as Meta has been purchasing AMD's CPUs for several years. During an analyst conference call, Huang Renxun outlined the fundamental differences in CPU design adopted by NVIDIA Corporation. He explained why NVIDIA Corporation minimized the approach used by Intel Corporation and AMD to break chips down into smaller units, emphasizing that NVIDIA Corporation's CPUs are capable of continuous processing of numerous simple tasks with good access to large amounts of memory. "Its design focus is on extremely high data processing capabilities," Huang Renxun said. "This is because most of the computing problems we focus on are data-drivenartificial intelligence being one of them." Dave Altavilla, Chief Analyst at HotTech Vision and Analysis, believes that NVIDIA Corporation aims to prove that the traditional CPUs that were primarily supplied by Intel Corporation are "no longer the assumed baseline for modern computing infrastructure, but just one of many architecture choices." Huang Renxun revealed that NVIDIA Corporation will disclose more information about its CPUs at the annual developer conference next month in Silicon Valley.