The x86 server loses weight: AI drives Arm and non-x86 systems

The server market is changing faster than it seemed just a few years ago. For decades, talking about servers meant almost always referring to x86, with Intel first and AMD later competing for the bulk of CPUs installed in data centers. That picture remains important, but it no longer fully explains where the money is moving.

According to IDC data for the first quarter of 2026, x86 servers generated $63.9 billion, accounting for 52.1% of the global market revenue. Non-x86 servers reached $58.7 billion, or 47.9%, after growing 107.6% year-over-year. This figure doesn’t mean Arm has captured nearly half of the server CPU market on its own. It signifies a broader and perhaps more significant trend: systems based on non-x86 architectures and AI-accelerated platforms now approach half of the sector’s revenue.

The nuance matters. Much of the growth does not come from traditional volume servers but from very expensive AI infrastructure, with GPUs, accelerators, custom designs, Arm CPUs, or specialized components. AI is transforming how data center spending is structured. It’s no longer just about how many servers are purchased, but what types of servers are bought and how much each node costs.

The key data point: non-x86 already approaches half the market

The global server market earned $122.6 billion in the first quarter of 2026, a 30.4% increase from the previous year. Within this, x86 systems declined by 2.9% year-over-year, while non-x86 systems soared by 107.6%. This contrast indicates a shift in priorities: companies still buy x86 servers, but growth is increasingly centered on accelerated systems and alternative architectures.

The immediate takeaway is that x86 is no longer growing at the market’s pace. Incremental spending is shifting toward platforms that don’t fit the traditional general-purpose server category. AI is mainly driving this: each accelerator rack can cost much more than a traditional rack, and that difference directly impacts revenue.

This doesn’t mean x86 is dead. Most enterprise applications, databases, virtualization, private clouds, corporate backends, and legacy workloads still run on Intel Xeon and AMD EPYC. But the fastest-growing data center segments are no longer those workloads. Instead, they come from training, inference, GPUs, high-speed networks, HBM memory, massive storage systems, and AI-optimized platforms.

Arm gains ground, but the shift is broader than Arm

Arm is one of the main beneficiaries of this trend. Its architecture has shifted from being almost exclusively associated with mobile and tablets to becoming a serious option for servers. AWS Graviton, Ampere Altra, NVIDIA Grace, Microsoft Cobalt, Google Axion, and other designs have demonstrated that Arm can operate in data centers with high efficiency, density, and cost-effectiveness.

Arm’s advantages typically include energy efficiency, integration, and customization capability. For hyperscalers purchasing or designing millions of cores, each watt matters. If an architecture can run certain workloads with lower power or better cost per task, the decision moves from ideological to economic.

However, not all non-x86 growth can be attributed to Arm. ASICs, platforms with integrated GPUs, accelerated systems, and servers built around components where the CPU no longer dominates cost also play a role. In many AI infrastructures, the CPU matters, but the economic value of the node resides in the accelerator, memory, network, and the whole system.

That’s why it’s better to talk about shifting expenditure rather than architecture replacement alone. The relative weight of x86 diminishes because data centers are purchasing much more expensive machines where the general-purpose CPU is no longer the sole protagonist.

AMD gains within x86 while Intel loses ground

Within the x86 world, AMD is experiencing one of its best moments in servers. According to Mercury Research data quoted by specialized media, AMD reached a 46.2% share of x86 server revenue in Q1 2026, compared to Intel’s 53.8%. In units, AMD also gained ground, though its share in units is lower than in revenue, indicating EPYC’s strong growth in higher-value segments.

This data is significant because it highlights two simultaneous shifts: first, x86 is losing prominence relative to non-x86 and accelerated systems; second, within x86, AMD is closing the gap with Intel.

AMD has leveraged EPYC to grow presence in cloud, HPC, databases, and enterprise workloads. Intel maintains scale, longstanding relationships, a massive installed base, and a strong corporate presence, but it no longer enjoys the nearly complete dominance it had for years.

The pressure on Intel is twofold. It must defend its market share against AMD in x86 and, at the same time, compete in a market where Arm, AI accelerators, and custom designs are capturing increasing investments.

AI is changing the economics of servers

The core of it all is AI. Previously, the average server was assessed based on CPU, memory, storage, and virtualization capacity. Now, an increasing part of the market is measured by GPUs, HBM, network bandwidth, cooling, electrical density, and the ability to train or run models.

Revenue from GPU-accelerated servers continues to increase, and systems with other accelerators like FPGAs or ASICs are also growing strongly. This explains why the market can expand significantly in revenue without the number of conventional servers growing at the same rate.

In this new landscape, the “Intel vs. AMD” debate seems insufficient. While still relevant, it is no longer enough. The real question is which complete architecture delivers the best performance per watt, dollar, and rack.

Why hyperscalers are betting on Arm

Major cloud providers have incentives that traditional companies might not always have. They can adapt software, control platforms, design their own chips, negotiate manufacturing at scale, and amortize investments across millions of instances. This makes Arm especially attractive in their world.

AWS pioneered this with Graviton, demonstrating that a cloud provider could offer competitive Arm instances for many workloads. Later, others followed. Google launched Axion. Microsoft moved forward with Cobalt. NVIDIA invested in Grace to combine Arm CPUs with GPUs in AI systems. Ampere has built Arm servers geared toward native cloud and efficiency.

For enterprise clients, switching to Arm depends more on software. If the application is containerized, runs on Linux, uses modern languages, and has compatible dependencies, migration can be reasonable. However, if it relies on legacy proprietary software, specific drivers, or licenses tied to x86, the move is much more complicated.

x86 will remain huge but no longer dominates everything

x86 retains a structural advantage: compatibility. Decades of enterprise software, hypervisors, operating systems, management tools, and certification processes are built around Intel and AMD. That foundation doesn’t disappear quickly.

Additionally, AMD EPYC and Intel Xeon continue to be very competitive in many workloads. Databases, virtualization, ERP, Windows Server environments, enterprise appliances, storage platforms, and private clouds will continue purchasing x86 for years.

The transition will be gradual and uneven. Cloud-native and AI workloads may move to Arm or specialized architectures sooner, while traditional enterprise workloads will take longer. The result will be more heterogeneous data centers.

More architectures mean more complexity for IT teams

Diversity offers advantages but also adds complexity. Infrastructure teams might end up managing x86 for traditional systems, Arm for certain cloud-native workloads, GPUs for AI, ASICs for inference, specialized networks, high-performance storage, and different observability and security policies.

This requires thinking more in terms of platforms than individual servers. Companies will need orchestration, automation, monitoring, and security tools capable of operating in hybrid environments. Kubernetes, containers, Infrastructure as Code, and abstraction layers will become increasingly important to prevent architectures from becoming siloed.

The potential benefit is clear: deploying each architecture where it adds the most value. The risk, however, is creating an environment that is too complex, costly to operate, and difficult to audit.

What it means for Europe and Spain

For Europe, the rise of non-x86 architectures has a strategic implication. As the market diversifies, opportunities emerge for new vendors, proprietary designs, research centers, supercomputing initiatives, and sovereignty projects. At the same time, dependence on critical components increases: U.S. GPUs, Asian memory foundries, specialized networks.

Spain, aiming to position itself in AI with new compute infrastructure, must consider this reality. A gigafactory or a large AI data center is not just defined by size and electrical capacity but also by the architecture chosen: x86, Arm, GPUs, ASICs, networking, cooling, and software.

AI’s rapid evolution pushes toward more diverse infrastructure. This can be beneficial if well-governed but may also introduce new dependencies if companies replace one dominant supplier with another.

The CPU’s renewed importance, but in a different way

In the early days of generative AI, it seemed everything revolved around GPUs. CPUs played a secondary role. Now, the situation is more nuanced. GPUs remain critical, but AI servers increasingly rely on efficient CPUs to coordinate data, run services, support accelerators, and manage workloads.

Arm capitalizes on this window by offering efficiency and flexibility. AMD leverages pressure within x86 with EPYC maintaining strong performance and density. Intel aims to defend its installed base and regain momentum with new Xeon generations, foundries, and accelerators. Meanwhile, NVIDIA seeks to integrate CPU, GPU, networking, and software into complete platforms.

The conclusion isn’t that x86 will disappear or that Arm will dominate overnight. It’s that the server market can no longer be explained with a single architecture. AI has transformed data centers into a mixture of CPUs, accelerators, memory, networking, power, and software.

IDC’s data signals this transition. Non-x86 systems already approach half of market revenue. Practically, this means expenditures are shifting toward more specialized, costly, and AI-oriented infrastructures. For AMD and Intel, the challenge isn’t just competition but also defending the role of x86 in a data center where growth no longer depends solely on x86.

Frequently Asked Questions

Is Arm already capturing 47.9% of the server CPU market?

Not exactly. The 47.9% refers to non-x86 server revenue according to IDC, not to Arm CPUs in isolation. That category includes accelerated systems, alternative architectures, and AI platforms.

What’s the x86 market share?

In Q1 2026, x86 servers generated $63.9 billion, representing 52.1% of the global server market, according to IDC.

Why are non-x86 servers growing so rapidly?

Driven by AI infrastructure demand, which boosts systems with GPUs, accelerators, Arm CPUs, ASICs, specialized networks, and high-performance memory.

Is AMD gaining on Intel?

Yes. In the x86 server market, AMD achieved a 46.2% revenue share in Q1 2026, based on Mercury Research data cited by industry sources.

Will x86 disappear from data centers?

No. x86 will remain highly relevant for enterprise workloads, private cloud, virtualization, and legacy applications. What’s changing is that the fastest-growing segments are shifting toward AI infrastructures and more diverse architectures.