The competitive landscape within the semiconductor industry is undergoing a significant shift at the 2026 Taipei International Computer Show (Computex).

If the AI battle in recent years centered on "who has the most powerful GPU," the message from this year's event is that as AI transitions from the training era to the inference era, competition is expanding from single-chip performance to encompass the entire system architecture.

On the first day, NVIDIA (NVDA) unveiled the Vera CPU for its next-generation Rubin platform, aiming to further solidify its full-stack AI infrastructure capabilities spanning from CPU and GPU to networking.

During the keynote on the second day, Intel (INTC) countered with its Xeon 6 Plus processor, manufactured using the 18A process, and presented its systematic, rack-scale AI infrastructure strategy built around Agentic AI for the first time.

The nature of this competition is no longer a simple CPU versus GPU battle, but a reallocation of system control in the AI era.

The Computational Logic Shift Driven by Agentic AI

Over the past two years, large model training propelled GPUs to become the most critical infrastructure in data centers.

During this phase, server architecture exhibited a distinct "GPU-centric" characteristic, with CPUs primarily handling auxiliary scheduling tasks and the vast majority of capital expenditure flowing to GPUs.

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

In the Computex keynote, Intel's CEO highlighted that future AI will not just be about training models, but about enabling agents to continuously execute tasks.

Compared to traditional chatbots, an agent needs to go through a cyclical process of "thinking, planning, acting, and reflecting," while frequently calling upon databases, APIs, and external tools.

This means AI systems are no longer just about pure matrix computations but require extensive task scheduling, resource management, and workflow orchestration.

A leading analyst estimates that during the training era, the CPU-to-GPU ratio in AI deployments was close to 1:4.

With the proliferation of Agentic AI, this ratio is expected to gradually move towards 1:1.

This context is crucial for understanding Intel's focus on launching the Xeon 6+.

According to Intel's disclosed data, the Xeon 6+, built on the 18A process, features 288 E-Core cores and is optimized for high-density inference and Agent hosting scenarios.

Intel even claims a single rack of Xeon 6+ platforms can support up to 150,000 running AI Agents.

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

Intel's Strategic Pivot: From Selling CPUs to Selling AI Systems

More noteworthy than the Xeon 6+ itself at Intel's launch was a series of initiatives centered on Rackscale infrastructure.

Historically, Intel primarily sold CPUs to customers; today, clients increasingly seek complete AI system solutions.

To this end, Intel announced collaborations with partners like Foxconn and SambaNova to build rack-scale AI infrastructure and introduced Rackscale Blueprints as reference architectures.

Concurrently, a venture backed by Vista Equity Partners and Cambium Capital showcased a fully disaggregated AI inference architecture.

In this setup, Intel Xeon processors handle task orchestration and execution, SambaNova's RDU manages token decoding, and NVIDIA (NVDA) Blackwell GPUs are responsible for prefill computations.

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

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

Vera Poses a Lesser Threat Than Perceived

If one were to chart the responsibility for Intel's declining server market share, it's important to recognize that the portion attributable to the Vera CPU is likely far smaller than the market assumes.

From an industry structure perspective, Vera appears more as a key component for NVIDIA (NVDA) to complete its own AI infrastructure portfolio, rather than a product specifically targeting the traditional general-purpose server market.

The forces that have been consistently eroding Intel's server CPU share are actually two others.

The first is the continuously improving competitiveness of AMD (AMD) EPYC processors.

Over the past few years, AMD has steadily increased its penetration in cloud computing and enterprise server markets, becoming Intel's most direct competitor within the x86 camp.

The second is the ongoing push by cloud giants like Amazon AWS, Microsoft, and Google towards their own self-developed Arm-based silicon.

Compared to traditional enterprise clients, these hyperscale cloud service providers possess greater software adaptation capabilities and stronger incentives to reduce long-term operational costs through custom chips.

Therefore, from a long-term perspective, Intel's challenges are not solely from NVIDIA (NVDA), but stem from the broader trend of data center architecture moving towards diversification.

The x86 Moat Endures, But Is No Longer Impregnable

Despite these shifts, the x86 ecosystem remains Intel's most critical asset.

Industry forecasts suggest that by 2030, over 80% of global servers will still run on the x86 architecture.

In critical business scenarios such as finance, industrial manufacturing, and government databases, numerous software systems have been built around x86 for many years, making migration costs and risks extremely high.

This is a key reason why x86 maintains its dominance even as Arm architecture continues to expand.

However, it must be noted that x86's advantage is more evident in the existing installed base, while new AI infrastructure is showing a trend towards greater openness and diversity.

For Intel, the crucial question it must answer in the future is not whether it can prevent Vera's entry into the server market, but whether it can redefine the value of the CPU in the AI era.

Concluding Analysis

The signals from Computex 2026 indicate that Intel is attempting a significant transformation: from being a CPU supplier to becoming a key participant in AI system infrastructure.

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

For investors assessing Intel's ability to defend its data center position, an indicator more telling than Vera is the choice for the control layer in future AI clusters deployed by AWS, Azure, and Google Cloud.

Whether they opt for Xeon, EPYC, or the cloud providers' own Arm processors may be more decisive for Intel's fate in the coming years than any presentation on the Computex stage.