The world spent $122.62 billion on servers in the first quarter of 2026, up 30.4 percent year on year but down only 2.1 percent sequentially, according to IDC. That near-flat sequential figure is the story. Component shortages cut shipments; price increases on CPUs, GPUs, DRAM and flash storage papered over the gap.

Some of that pricing looks deliberate. Big OEMs have hinted that supply exceeds demand by perhaps 25 to 30 percent, and where contracts have not locked in machine costs, opportunistic increases are entirely plausible. Server makers serve the most penny-pinching customers on earth. A little margin, for once, seems fair.

Why the small sequential decline matters

Historically, the fourth quarter always beat the first. IT shops spent their budgets before year-end or lost them, producing a sawtooth pattern in server revenue that held for decades before the GenAI boom. A 2.1 percent sequential drop is almost nothing against that historical norm.

The quarter also sits at an awkward junction. A global refresh cycle for non-AI systems is colliding with a component market where demand runs far ahead of supply. Prices for memory, storage and accelerators are climbing accordingly. The revenue line, in short, reflects price, not units.

What did GPU and XPU systems earn?

IDC put GPU-accelerated server revenue at $68.9 billion for the quarter, up 24.8 percent year on year and down 2.5 percent sequentially. GPU systems took 56.2 percent of all server revenue. A datacenter GPU with its HBM memory costs roughly $50,000 today, and next-generation multi-chip devices will approach or exceed $100,000.

The more striking disclosure: IDC broke out other XPU-accelerated systems for the first time in its public data. That segment reached $17.1 billion, or 13.9 percent of server revenue, up from 8.2 percent a year earlier. The mix is mostly Google TPUs and AWS Trainium, with smaller contributions from Microsoft, Cerebras Systems and Groq.

• GPU servers: $68.9 billion, 56.2 percent of revenue
• Other XPU servers: $17.1 billion, 13.9 percent of revenue
• Combined accelerators: $86 billion, 70.2 percent of revenue, up from 62.2 percent a year ago

Is x86 dominance ending?

Counted by money rather than units, the answer is nearly yes. X86 hosts drove $63.9 billion in sales, down 2.9 percent year on year and down 8.5 percent sequentially. Absent expensive, scarce components, IDC's figures suggest the X86 business would likely have grown strongly on the refresh cycle.

Non-X86 machines took in $58.7 billion, up 2.1 times year on year and up 5.8 percent sequentially. Hyperscaler-built Arm processors dominate that segment, alongside Nvidia systems based on its Grace Arm server CPU. IBM's Power Systems RISC machines and System z mainframes hold a slice too. RISC-V machines, the source notes, will not be far behind.

Revenue share now stands at 52.1 percent for X86 against 47.1 percent for everything else. That gap will close quickly if component supply loosens and hyperscalers keep pushing their own silicon.

Who is winning on the vendor side?

Dell, anointed as Nvidia's chosen OEM, benefits beyond its enterprise base through sovereign, neocloud and model-builder deals. Its AI server business is an order of magnitude larger than HPE's, which makes its overall server business 5.5 times bigger.

The ODMs serving hyperscalers and cloud builders are running at full tilt, constrained only by component supply. Their business would be considerably larger if the chips were easier to obtain. Nobody, IDC's numbers imply, believes demand is the problem; the coming quarters hinge on whether supply catches up before prices finally throttle the refresh cycle.