On October 22, 2026, TrendForce released a forecast that global server CPU shipments will grow more than 40% year over year in 2027. That follows 31% growth in 2026, and the firm sees expanding CPU demand driven by the spread of AI agents as the main factor. At the same time, the market share of Arm-based CPUs is expected to rise to about 30% in 2027. However, because the overall market is expanding so quickly, shipments of x86 processors, used by Intel and AMD, could still increase even if x86's share declines. As the way AI workloads are processed changes, the role expected of CPUs is widening, creating conditions in which Arm's growth and rising x86 shipments can coexist.
AI agents expand the role of the CPU
TrendForce is focusing on the changing ratio between the CPUs installed in servers and the GPUs or AI-specific ASICs alongside them. According to the forecast the company released on October 22, the ratio of CPUs to GPUs/ASICs is expected to shift from the conventional 1:8 to 1:2, and eventually to 1:1. The forecast is based on the server rack configurations and growth outlooks planned by major operators, and does not describe the average configuration of all data centers currently in operation.
Behind this shift is the spread of AI agents. With conventional generative AI, the core task was a model generating a response to user input. AI agents, by contrast, do more than generate responses: they carry out a series of tasks in sequence, such as executing code, calling external tools, and taking the next step based on the results. Even when GPUs or dedicated accelerators handle the AI model's inference itself, CPUs are needed to run the surrounding software and execution environments. As AI agents take on more complex and longer-running work, the processing power of the CPUs that support them becomes as important as adding more accelerators.
A reference system configuration with FuriosaAI that Arm published on October 15 illustrates this division of labor concretely. In this configuration, the Arm AGI CPU handles execution management for AI agents, tool execution, and auxiliary services, while dedicated accelerators take care of processing inputs to the AI model and generating outputs. Although this is a configuration introduced by Arm itself, it shows that CPU demand is growing for reasons beyond simply replacing GPU processing with CPUs.
NVIDIA, too, cites Python execution environments and code execution in isolated environments as key uses for the Vera CPU it announced on May 31. Vera is designed to be used not only in systems paired with GPUs but also in standalone, CPU-centric servers. As the execution environments and auxiliary services used by AI agents multiply, demand arises beyond the host CPUs connected to GPUs.
The impact of the changing CPU-to-accelerator ratio is not small. Assuming the number of GPUs and ASICs stays constant, moving from 1:8 to 1:2 would quadruple the number of CPUs required, and moving to 1:1 would make it eightfold. In practice, however, use cases, system configurations, and deployment timing differ, so these multiples cannot be applied directly to the growth rate of overall market shipments. The 40%-plus growth rate TrendForce forecasts is an outlook for the entire server CPU market, including these configuration changes.
Arm's share heads toward 30%, while x86 shipments may still rise
In the 2025 server CPU market, the x86 architecture held about 90% of the share, while Arm-based CPUs had just passed 10%. TrendForce forecasts that Arm's share will exceed 20% in 2026 and reach about 30% in 2027. Arm-based CPUs here include products developed and supplied by companies such as AWS and NVIDIA, not only CPUs sold by Arm itself.
A decline in share, however, does not mean the same thing as a decline in shipment volume. Market share is the proportion of total shipments, so if the market itself is expanding rapidly, sales volume can grow even as share falls. Given the forecast that the overall server CPU market will grow more than 40% in 2027, Arm's share gains will not necessarily translate into lower x86 shipments.
The x86 market also includes both Intel and AMD. If Arm's share grows while AMD's presence within the x86 market increases, Intel's and AMD's shipment volumes could move differently. For that reason, Intel's sales trends on its own cannot be judged from share by architecture alone. Evaluating individual companies' growth also requires looking at shipment volumes and supply conditions by company.
Cloud giants' in-house CPUs support the Arm camp's growth
Behind the rapid spread of Arm-based server CPUs are the development of proprietary CPUs by major cloud providers and NVIDIA's entry into the market. TrendForce cites, in addition to AWS's Graviton series, large-volume shipments of Google's Axion and mass production of Microsoft's Cobalt 100 as growth drivers. Arm's AGI CPU is also joining the market, and TrendForce expects expanded shipments of NVIDIA Vera to significantly lift Arm's overall share in 2027.
NVIDIA's Vera can be used not only as a host CPU controlling GPU servers but also in standalone CPU servers that handle AI agents' code execution and data processing. This could allow the company to capture CPU-centric compute demand in addition to GPU-centered AI systems. However, at the time of the May announcement, Anthropic was still evaluating the adoption of Vera. Companies being named as potential adopters should be distinguished from order volumes actually being finalized.
On the x86 side, the challenge is whether vendors can supply enough products to meet the growth in demand. According to TrendForce, Intel's shipments have been delayed by issues with the manufacturing process and packaging of new products, and it continues to rely on the existing Emerald Rapids and Granite Rapids. Constraints on its internal production capacity also mean it cannot fully meet customer orders, which TrendForce says is contributing to its declining market share.
AMD is also constrained by supply capacity. Its main products are Genoa and Turin, and while the next-generation Venice began small-scale mass production in 2026, TrendForce analyzes that tight TSMC production capacity limits its ability to fill additional orders. Even so, AMD's share of the server CPU market is expected to rise to about 27% in 2026. This 27% is its share of the entire server CPU market, not just within the x86 market.
Given these circumstances, competition in the server CPU market is no longer decided by instruction set or standalone processor performance alone. Whether cloud providers can design CPUs suited to their own services, and whether chipmakers can reliably supply the necessary volumes, are also important. As AI agents expand CPU demand, which companies capture that growth will depend on both product competitiveness and production capacity.
CPU shipments up 40%, AI server units up 36%, revenue up 90%: what each figure measures
In a separate market forecast released on October 14, TrendForce projected that AI server shipments in 2027 will grow about 36% year over year, with revenue rising more than 90% to about $920 billion. These may look similar to the latest forecast of more than 40% growth in server CPU shipments, but each measures a different thing.
The server CPU shipment forecast counts individual chips and includes CPUs for general-purpose servers. AI server shipments, on the other hand, are counted in whole systems, and revenue represents sales value. Simply comparing these figures or calculating the gaps between their growth rates cannot tell you whether CPUs are in short supply or capital investment is excessive.
| 2027 market forecast | Year over year | Scope / unit | Announced |
|---|---|---|---|
| Server CPU shipments | Over 40% | Number of CPUs, including those for general-purpose servers | October 22, 2026 |
| AI server shipments | About 36% | Number of AI server systems | October 14, 2026 |
| AI server revenue | Over 90% | Sales value of AI servers | October 14, 2026 |
This table organizes TrendForce's published market forecasts for server CPUs and AI servers by scope and unit. All target 2027, but the announcement dates and what is being counted differ. Also, because the number of CPUs installed in a server varies by configuration, CPU shipments and server unit numbers will not necessarily grow at the same rate.
The large gap between the growth rates of AI server units and revenue also reflects rising system selling prices. TrendForce explains that shortages and price increases for components such as memory will push up the average selling price of next-generation AI server racks. Therefore, even if AI server revenue grows more than 90%, the compute equipment actually deployed will not necessarily increase at the same rate. For cloud providers and data center operators, what matters is not only the quantity of equipment needed but also how much it costs to secure that configuration.
Similar caution applies to capital expenditure. TrendForce forecasts that capex by the world's nine major cloud providers will exceed $1.5 trillion in 2027, raising its year-over-year growth outlook from 50% to 60%. Behind this are not only higher prices for servers and components but also an increase in prepayments to secure supply. In other words, the compute equipment actually put into operation will not necessarily grow in proportion to the increase in capex.
In the 2027 server market, CPUs are expected to become even more important in supporting not only AI model inference but also code execution and tool operation by AI agents. The focus going forward will be whether the CPU and accelerator configurations companies are planning translate into actual product shipments, and whether they can secure production capacity to match growing demand. To efficiently support AI agents through a complete series of tasks, an appropriate combination of CPUs and accelerators is essential, as is a setup that can reliably procure the necessary hardware.