英伟达 Vera CPU 来袭，英特尔 (INTC.US) 祭出至强 6+：Agentic AI 能否带来 CPU 价值重估？

The Zhitong Finance APP noted that at the 2026 Taipei International Computer Show (Computex), the competitive focus in the semiconductor industry is shifting. If the core of the AI battle in recent years was "who has the strongest GPU," then the signal conveyed by this year's Computex is that as AI transitions from the training era to the inference era, competition is expanding from single-chip performance to entire system architecture. On the first day of the exhibition, NVIDIA (NVDA.US) launched the Vera CPU for the next-generation Rubin platform, attempting to further enhance its full-stack AI infrastructure capabilities from CPU, GPU to network interconnect; while on the second day's keynote speech, Intel (INTC.US) unveiled the Xeon 6 Plus processor using 18A process technology and systematically showcased 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 propelled GPUs to become the most important infrastructure in data centers.

During this phase, server architecture has shown a clear "GPU centralization" characteristic. CPUs mainly undertake auxiliary scheduling tasks, while the vast majority of capital expenditures flow towards GPUs.

However, as Agentic AI begins to enter real business scenarios, the computing demands of data centers are changing.

In the keynote speech at Computex, Intel CEO Pat Gelsinger stated that in the future, AI will no longer just train models, but will enable agents to continuously perform tasks. Compared to traditional chatbots, an agent needs to undergo a cycle of "thinking, planning, acting, and reflecting," while frequently calling databases, APIs, and external tools.

This means that AI systems are no longer just simple matrix calculations, but require extensive task scheduling, resource management, and workflow orchestration.

Ben Bajarin, Chief Analyst at Creative Strategies, predicts that during the training era, the ratio of CPU to GPU in AI deployments is close to 1:4; as Agentic AI becomes more widespread, this ratio is expected to gradually approach 1:1.

This is also an important background for Intel's emphasis on launching the Xeon 6+.

According to data disclosed by Intel, the Xeon 6+, built using 18A process technology, has 288 E-Core cores optimized for high-density inference and agent hosting scenarios. Intel even claims that a single rack Xeon 6+ platform can support up to 150,000 AI Agents running.

For Intel, this does not mean the end of the GPU era, but rather that the role of the CPU in AI systems is being redefined—from a traditional general-purpose computing platform to the scheduling and orchestration hub within AI infrastructure.

Intel's true goal: shifting from selling CPUs to selling AI systems Compared to the Xeon 6+ itself, what is more noteworthy at this Intel press conference is a series of layouts centered around Rackscale infrastructure.

In the past, Intel primarily sold CPUs to customers; now, customers increasingly hope to obtain complete AI system solutions.

To this end, Intel announced partnerships with Foxconn, SambaNova, and others to jointly create rack-scale AI infrastructure and launched Rackscale Blueprints.

At the same time, Vector Core Compute, supported by Vista Equity Partners and Cambium Capital, showcased a fully disaggregated AI inference architecture for the first time:

In this architecture, Intel Xeon processors are responsible for task orchestration and execution, SambaNova RDU handles token decoding, and NVIDIA Blackwell GPUs are responsible for prefill computation.

The logic behind this architecture is that future AI systems may not require all tasks to be completed by GPUs; different computing units can take on different workloads based on their characteristics, thereby improving overall efficiency.

In a sense, this is also Intel's alternative answer to the current "GPU-centric" AI architecture.

Vera is not Intel's biggest threat

However, if we were to depict the loss of Intel's server market share as a responsibility allocation pie chart, the market may need to recognize a reality: the proportion of Vera CPUs may be far less significant than the market imagines.

From an industry perspective, Vera seems more like an important component of NVIDIA's efforts to 完善 its AI infrastructure landscape, rather than a product specifically targeting the traditional general-purpose server market.

The real forces that are continuously eroding Intel's server CPU share are actually two other powers.

First, the competitive strength of AMD (AMD.US) EPYC processors continues to improve.

In recent years, AMD has been expanding its penetration in the cloud computing and enterprise server markets, becoming Intel's most direct competitor within the x86 camp.

Second, the self-developed Arm routes continuously promoted by cloud giants such as Amazon AWS, Microsoft, and Google.

Compared to traditional enterprise customers, these large-scale cloud service providers have stronger software adaptation capabilities and are more motivated to reduce long-term operating costs through self-developed chips.

Therefore, from a long-term perspective, the challenges Intel faces do not only come from NVIDIA, but rather the entire data center architecture is gradually moving towards diversification.

The x86 moat still exists, but it is no longer impregnable

Nevertheless, the x86 ecosystem remains Intel's most important asset.

According to IDC's forecast, by 2030, over 80% of servers worldwide will still run on x86 architecture.

In critical business scenarios such as finance, industrial manufacturing, and government databases, a large number of software systems have been built around x86 for many years, and the costs and risks of migration are extremely high This is also an important reason why, even though the Arm architecture continues to expand, x86 can still maintain its dominant position.

However, it is important to note that the advantages of x86 are more reflected in the existing market, while the new AI infrastructure is showing a more open and diverse development trend.

For Intel, the real question that needs to be answered in the future is not whether it can prevent Vera from entering the server market, but whether it can redefine the value of CPUs in the AI era.

Conclusion

From the signals released at Computex 2026, Intel is attempting to complete an important transformation: from a CPU supplier to a key participant in AI system infrastructure.

This means that future competition will no longer be just about "who has the strongest chip," but rather "who can define the next generation of AI system architecture."

For investors, a more important indicator to judge whether Intel can maintain its data center position, beyond Vera, is:

Which main control layer will be chosen in the future new AI clusters of AWS, Azure, and Google Cloud: Xeon, EPYC, or the cloud vendors' own Arm processors?

This answer may determine Intel's fate in the coming years more than any speech on the Computex stage