AI Agents Push Server CPUs Back Into the Spotlight
For two years the AI hardware conversation has been a GPU conversation. That is starting to change. Research firm TrendForce now expects server CPU shipments to grow 31% in 2026, up from steady growth that began in the second half of 2025, and to climb more than 40% again in 2027 as AI reasoning workloads expand and agentic services reach more users.
The driver is a shift in how AI services are built. Early large language models mostly answered one question at a time: a user typed, the model replied. AI agents work differently. They plan several steps, call tools, read files, query databases and coordinate with other software. Those jobs lean on general-purpose processors to schedule tasks, move data and handle the parts of a workload that do not fit neatly on a GPU. As more companies deploy agents, the machine room has to be balanced rather than stacked with accelerator cards.
That rebalancing is visible in purchasing. Data-center operators are moving from buying clusters of GPUs alone to buying whole compute racks that pair accelerators with a healthy share of CPUs, memory and networking. The logic of AI investment, in other words, is shifting from one component to a full system.
The market structure may also move. Server CPUs have long been dominated by the x86 architecture used by Intel and AMD. TrendForce suggests that dominance could face pressure as AI workloads diversify and alternative architectures gain attention, though it does not put a number on the shift.
The stakes are large because servers are the backbone of cloud computing. Faster CPU demand would ripple through memory, motherboards, power supplies and cooling, and it would give chipmakers beyond the GPU leaders a larger role in the AI build-out. It also matters for energy: a balanced rack can be more efficient than one packed with accelerators idling between jobs.
For China, the story connects to a broader push to build domestic computing supply chains. Chinese server makers and chip designers have been investing in processors for cloud and AI workloads, and a global swing toward CPU-heavy infrastructure could widen the addressable market for those products.
The caveat is that forecasts are not orders. The 31% and 40% figures are projections, and AI spending has shown itself capable of sharp swings. If agent adoption slows, or if buyers decide existing servers can handle the load, the CPU rebound could arrive later and smaller than expected. Still, the direction is clear: the AI data center is becoming less of a GPU showcase and more of an orchestra, and the conductor is the CPU.