Arm is nearing half of the server market share by revenue

The data center is changing architecture faster than recent server history suggested. For decades, x86 was almost synonymous with enterprise infrastructure: Intel Xeon first, AMD EPYC later, and around them a whole market of manufacturers, integrators, hypervisors, operating systems, and management tools. This reality hasn’t disappeared, but artificial intelligence is shifting the market’s economic weight elsewhere.

According to IDC data compiled by Tom’s Hardware, ARM-based platforms surpassed 45% of global server market revenue in the first quarter of 2026. This figure does not mean that ARM sells more processors than x86 in units. The nuance is important: x86 still dominates in volume, but large accelerated AI systems—especially racks with GPUs and integrated ARM CPUs—are elevating ARM’s revenue share to levels that just a few years ago seemed unlikely.

AI has changed how the server market is measured

The key data point isn’t just about ARM. It’s about the market composition. The global server business reached $122.6 billion in revenue in the first quarter of 2026, a 30.4% increase year over year, according to IDC. Within that figure, systems accelerated with GPUs, ASICs, or FPGAs generated more than 70% of the revenue for the quarter.

This shifts the traditional interpretation. A general-purpose two-socket x86 server can still be essential for virtualization, databases, ERP, private cloud, storage, enterprise services, and thousands of critical workloads. But an AI rack with state-of-the-art accelerators costs significantly more per unit. When the market is measured by revenue, these platforms tilt the balance.

That’s where ARM enters. Many of the most expensive AI architectures incorporate ARM CPUs as part of the complete system. NVIDIA Grace Blackwell, for example, combines Blackwell accelerators with Grace CPUs based on ARM. Major hyperscalers have also developed their own ARM processors for cloud and AI workloads, such as AWS Graviton, Google Axion, and Microsoft Cobalt. The result is a less visible transition than that of the PC, but more significant for infrastructure spending.

x86 is not collapsing, but its economic prominence is waning

The temptation is to interpret this movement as an immediate defeat for x86. That would be exaggerated. x86 servers still form the backbone of a significant portion of the installed base, and according to the same data, they still account for around 52% of market revenue. Units-wise, their position is even stronger. AMD and Intel continue shipping millions of EPYC and Xeon processors annually to data centers.

What’s changing is the center of gravity. The infrastructure that is growing most and generating the highest revenue is no longer the general-purpose server, but the high-value accelerated system. In that category, ARM appears integrated into designs where the CPU is no longer the sole protagonist but works as a coordinating component with GPUs, memory, networking, and storage at very high performance levels.

There’s also a situational factor: supply restrictions. IDC attributes part of x86’s weakness to component availability issues, such as CPUs, DRAM, NAND, and hard drives. It’s not just demand. The supply chain is prioritizing AI platforms and higher-margin components, which influence what can be manufactured and delivered. Dell’Oro Group also noted that semiconductor and data center component revenue grew 116% YoY in Q1 2026, driven by AI infrastructure expansion and increasing memory prices.

For Intel and AMD, this message is uncomfortable but not conclusive. They remain strong in general-purpose workloads, enterprise virtualization, and traditional servers. AMD, in particular, has increased its share in x86 with EPYC. But the most dynamic part of spending is shifting towards accelerated designs, custom CPUs, and architectures where power consumption, rack density, and integration with accelerators outweigh traditional compatibility.

ARM turns licensing into a strategic position

ARM does not sell chips like Intel or AMD in their traditional model. Its business has been based on architecture licenses, intellectual property, and royalties per chip. This approach allowed it to dominate mobile and subsequently enter automotive, embedded devices, edge, and data centers. Now, AI is reinforcing that position.

In Q4 of its fiscal year 2026, ARM reported record revenues of $1.49 billion, a 20% increase year-over-year. License revenues grew 29%, reaching $819 million, and royalties increased 11%, to $671 million. The company highlighted that its data center royalties more than doubled compared to the previous year, supported by the adoption of Armv9, Arm CSS, and ARM chips in cloud and AI workloads.

The new development is that ARM no longer confines itself to licensing designs. In 2026, it introduced the ARM AGI CPU, its first proprietary chip for data centers, aimed at AI infrastructure. The company claims this CPU can deliver more than twice the performance per rack compared to x86 platforms and reduce up to $10 billion in CapEx per gigawatt of data center capacity, although this is an ARM estimate primarily for commercial argument.

This move is delicate. ARM has grown as a neutral provider for a wide ecosystem of manufacturers. Developing its own silicon could create tensions with partners also designing ARM-based chips. However, in the AI market—where hyperscalers seek density, efficiency, and custom designs—owning a CPU can give ARM more control over the platform and greater ability to capture value.

The clear implication for the cloud sector is that data center architecture is becoming less homogeneous. x86 will remain essential, but will coexist with ARM CPUs, proprietary accelerators, GPUs, DPUs, SmartNICs, ultra-low latency networks, and storage designed to support training and inference workloads. The server is no longer a standard box but a component within much larger, denser, and more expensive systems.

For companies and cloud providers, this shift requires a reassessment of assumptions. Infrastructure decisions can no longer rely solely on historical compatibility. Considerations will include cost per watt, performance per rack, software availability, ecosystem maturity, vendor dependence, virtualization support, observability, security, and the ability to operate hybrid architectures.

ARM has not yet gained market share in servers by units, but it has demonstrated its capacity to capture significant value when spending concentrates on AI. That’s the real breakthrough. The race is no longer just about how many CPUs are sold but which architectures dominate within the systems that account for most of the data center budget.

Frequently Asked Questions

Does ARM already sell more servers than x86?
Not necessarily. The over 45% figure refers to revenue, not units. x86 still leads in volume, but AI-related ARM systems tend to be high-priced and thus increase their economic share.

Why does AI favor ARM in data centers?
Because many AI platforms combine GPUs or accelerators with efficient, highly integrated ARM CPUs. Additionally, hyperscalers are developing custom ARM-based chips to improve cost, power efficiency, and density.

Are Intel and AMD out of the server market?
No. Intel Xeon and AMD EPYC remain essential for general-purpose servers, enterprise virtualization, databases, and many workloads. What’s changing is that revenue growth is increasingly concentrated in accelerated systems.

What is the significance of the ARM AGI CPU?
It’s ARM’s first custom chip for data centers. It consolidates its shift from a provider of intellectual property toward a more comprehensive AI infrastructure platform, though it still needs to prove real adoption against x86 alternatives and custom designs.