AINEWS
The server market is splitting in two, and a memory chip squeeze is the hinge
x86 servers, the Intel and AMD designed chips that have run datacenters for forty years, fell 2.9% in Q1 2026 even as Arm based AI systems grew 107%.
x86 servers, the Intel and AMD designed chips that have run datacenters for forty years, fell 2.9% in Q1 2026 even as Arm based AI systems grew 107%.
The server market stopped being one market last quarter. In Q1 2026, worldwide server revenue hit $122.6 billion, up 30.4% year on year, according to IDC's Worldwide Quarterly Server Tracker as reported by The Register. Inside that total, two server businesses are now running on different tracks.
Non-x86 servers, the bucket dominated by Arm-based machines carrying Nvidia's AI accelerators, generated $58.7 billion in revenue, up 107% from a year earlier and good for 47.9% of the worldwide market. x86 servers, the Intel- and AMD-designed machines that have anchored datacenters for four decades, brought in $63.9 billion. That total fell 2.9% year on year in a quarter when overall demand was anything but soft, per IDC's tracker figures as covered by The Register.
The split is sharper inside the AI tier. GPU-accelerated servers alone reached $68.9 billion, up roughly 25% year on year, while "other accelerated" servers, the systems carrying custom AI silicon from hyperscalers and other accelerator vendors, climbed 122% to $17.7 billion. Add those two categories to the Arm-based non-x86 bucket and the AI-built portion of the server market is now the majority of revenue, and it is growing at triple-digit or near-triple-digit rates while the conventional half shrinks.
The story behind the split is not that x86 buyers walked away. Order pipelines for traditional servers are strong. The story is that the memory chips those servers need are being routed somewhere else. DRAM and NAND makers are allocating their capacity to the higher-margin AI accelerator and GPU parts, which means conventional x86 boxes are getting built at the rate their memory allocation allows, not at the rate their demand would support. IDC attributes the divergence to that supply-side pressure, on top of the AI demand surge, according to The Register's writeup of the tracker.
That is what makes this look like a permanent bifurcation rather than a temporary reordering. The supply-allocation incentives are structural: a gigabyte of DRAM bolted to an AI accelerator earns a memory maker more than the same gigabyte bolted to a general-purpose server. As long as that margin gap holds, the memory throttle on traditional boxes will hold with it, and the gap between the AI tier and the conventional tier will widen rather than close. The 47.9% non-x86 share is not a milestone to celebrate; it is a doorway to a split market that is likely to deepen in coming quarters.
The order book tells the same story from the other side. Hyperscalers and large enterprises keep placing big x86 orders, and they are not all getting filled this year. Watch the IDC tracker's next two prints for whether the 47.9% non-x86 share keeps climbing, and whether x86 revenue stabilizes once DRAM and NAND capacity decisions reset for the second half. The split is the story. The open question is how wide it gets.