NVIDIA Corporation's Vera CPU is coming, and Intel Corporation (INTC.US) unveils Xeon 6+: Can Agentic AI bring a reevaluation of CPU value?

Notice that at the Taipei International Computer Show in 2026, the focus of competition in the semiconductor industry is changing. If in the past few years the core of the AI war was "who has the strongest GPU", then the signal transmitted at this year's Computex is that competition is shifting from single-chip performance to the entire system architecture as AI moves from the training era to the inference era. On the first day of the exhibition, NVIDIA Corporation (NVDA.US) released the Vera CPU for the next generation Rubin platform, attempting to further improve its full-stack AI infrastructure capabilities from CPU, GPU to network interconnection; and in the second day's keynote speech, Intel Corporation (INTC.US) unveiled the Xeon 6 Plus processor using the 18A process and systematically demonstrated its rack-level AI infrastructure strategy built around Agentic AI. The essence of this competition is no longer a simple confrontation between CPU and GPU, but a redistribution of system control in the AI era. Agentic AI is changing the computing logic of data centers. In the past two years, large model training has driven GPU to become the most important infrastructure in data centers. At this stage, server architecture shows a clear "GPU-centralized" feature. CPUs mainly handle auxiliary scheduling tasks, while the vast majority of capital spending flows to GPUs. But as Agentic AI begins to enter real business scenarios, the computing needs of data centers are changing. At Computex keynote speech, Intel Corporation CEO Chen Liwu proposed that in the future, AI will no longer only train models, but will allow intelligent agents to perform tasks continuously. Compared to traditional chat Siason Robot & Automation, an intelligent agent needs to go through a cycle of "thinking, planning, acting, and reflecting", while frequently accessing databases, APIs, and external tools. This means that AI systems are no longer just simple matrix operations, but require a large number of task scheduling, resource management, and workflow orchestrations. Chief Analyst Ben Bajarin from Creative Strategies predicts that in the training era, the ratio of CPU to GPU in AI deployment is close to 1:4; with the popularization of Agentic AI, this ratio is expected to gradually approach 1:1. This is also an important background for Intel Corporation's focus on the Xeon 6+ this time. According to the data disclosed by Intel Corporation, the Xeon 6+ built with the 18A process has 288 E-Core cores optimized for high-density inference and Agent hosting scenarios. Intel Corporation even claims that a single Xeon 6+ platform can support up to 150,000 AI Agents running. For Intel Corporation, this does not mean the end of the GPU era, but rather the redefinition of the role of CPUs in AI systemsfrom a traditional general-purpose computing platform to the central hub for scheduling and orchestrating AI infrastructure. Intel Corporation's true goal: shifting from selling CPUs to selling AI systems. Compared to Xeon 6+ itself, what is more noteworthy at Intel Corporation's release conference is actually the series of layouts around the Rackscale infrastructure. In the past, Intel Corporation sold more CPUs to customers; now, customers are increasingly looking for complete AI system solutions. To this end, Intel Corporation announced a collaboration with partners such as Foxconn and SambaNova to jointly build rack-level AI infrastructure and launch Rackscale Blueprints. At the same time, Vector Core Compute, supported by Vista Equity Partners and Cambium Capital, demonstrated a fully disaggregated AI inference architecture for the first time: In this architecture, the Intel Corporation Xeon processor is responsible for task orchestration and execution, the SambaNova RDU is responsible for Token decoding, and the NVIDIA Corporation Blackwell GPU is responsible for prefill computation. The logic behind this architecture is that in the future, AI systems may not necessarily need all tasks to be completed by GPUs, and different computing units can take on different workload characteristics to improve overall efficiency. In a sense, this is also another answer that Intel Corporation has proposed to the current "GPU-centralized" AI architecture. Vera is not Intel Corporation's biggest threat. However, if the loss of market share in the server market by Intel Corporation is depicted in a pie chart, the market may need to realize a reality: Vera CPU may not account for as much of a proportion as the market imagines. From an industry perspective, Vera is more like an important component of NVIDIA Corporation's AI infrastructure layout enhancement, rather than a product specifically targeting the traditional general server market. The real forces continuously eroding Intel Corporation's server CPU market share are actually two other forces. First, the continuously improving competitive power of AMD (AMD.US) EPYC processors. Over the past few years, AMD has steadily expanded its penetration rate in the cloud computing and enterprise server markets, becoming the most direct competitor to Intel Corporation within the x86 camp. Second, the self-developed Arm route pushed forward by cloud giants such as Amazon.com, Inc. AWS, Microsoft Corporation, and Alphabet Inc. Class C. Compared to traditional enterprise customers, these super large-scale cloud service providers have stronger software adaptation capabilities and are more inclined to reduce long-term operating costs through self-developed chips. Therefore, from a long-term perspective, the challenge facing Intel Corporation does not only come from NVIDIA Corporation, but rather the entire data center architecture is gradually diversifying. The x86 moat is still there, but no longer impregnable. Even so, the x86 ecosystem remains Intel Corporation's most important asset. According to IDC forecasts, by 2030, over 80% of servers worldwide will still be running on the x86 architecture. In key business scenarios such as finance, industrial manufacturing, and government databases, a large number of software systems have been built around x86 for many years, with the cost and risk of migration extremely high. This is why, even as the Arm architecture continues to expand, x86 can still maintain its dominant position. However, it needs to be understood that the advantage of x86 is more reflected in the stock market, while the new AI infrastructure is showing a more open and diverse development trend. For Intel Corporation, the real question that needs to be answered in the future is not whether they can stop Vera from entering the server market, but whether they can redefine the value of CPUs in the AI era. Conclusion From the signals released at Computex 2026, Intel Corporation is attempting to complete a major transformation: moving from a CPU supplier to a significant participant in AI system infrastructure. This means that future competition is no longer just about "who has the strongest chip", but about "who can define the next generation of AI system architecture". For investors, the more important indicator than Vera to watch for Intel Corporation's future in the next few years is: In the future AI clusters of AWS, Azure, and Google Cloud, whether they choose Xeon, EPYC, or the cloud vendor's own Arm processor as the main control layer. This answer may determine the fate of Intel Corporation in the coming years more than any speech on the Computex stage.