SuperMicro has muscled its way into second place in the server market, and Lenovo and HPE are the ones left in the dust.
IDC figures for the fourth quarter of 2025 put SuperMicro behind only the Grey Box Shifter Dell with $11.7bn revenue and roughly nine per cent of global server sales.
Dell led on $12.6bn and 10 per cent share, while IEIT Systems grabbed third with $5.2bn and four per cent.
Lenovo came next on $5.1bn, with HPE trailing at $3.9bn, which is not what you want when everyone is flogging “AI infrastructure” like it is bottled water.
IDC, research director of Worldwide Enterprise Infrastructure Trackers Juan Seminara said: “The race for AI [artificial intelligence] adoption is settling the market pace, and with companies starving for infrastructure looking not only at GPUs [graphics processing units], but consuming more CPUs [central processing units] among other components in order to feed their needs, we are going to see more price pressures, and that may impact on market dynamics with less units but higher average selling prices going forward.”
IDC said pricing is getting twitchy for parts such as GPUs, dynamic random-access memory, and solid-state drives, with buyers trying to lock in costs before the next surge.
It thinks the volatility will bite harder in 2026, with demand still outstripping capacity in the near term, so expect more kit to cost more and arrive later.
While the legacy server crowd scraps for scraps, AMD is trying to drag buyers away from cloud AI by selling the idea of on-premise “Agent Computers”.
In an AMD bog post, the firm lays out how to run OpenClaw locally on Ryzen AI Max+ processors and Radeon GPUs using Windows 11 and the Windows Subsystem for Linux.
AMD said a PC configured with 128GB unified memory can run “cloud-quality AI agent workloads efficiently” using OpenClaw, which is the sort of claim that dares you to check the receipts.
Using its benchmark data, AMD says the Qwen 3.5 35B A3B model pushes about 45 tokens per second and chews through 10,000 input tokens in about 19.5 seconds.
It claims a maximum context window of 260,000 tokens and up to six agents running concurrently, pitched as local experimentation without the usual cloud meter ticking.
AMD’s pitch sounds like a branch office server from the pre-cloud days, running autonomously and taking jobs via a browser on another Windows PC or through Slack or WhatsApp.
Vendors with “agent-ready” PCs include HP, Lenovo and Asus, which reads like an attempt to rebrand a powerful desktop as something you should rack and forget.
IDC said revenue for servers with an embedded GPU grew 59.1 per cent year on year in the fourth quarter of 2025, making up more than half of total server market revenue.
Ryzen AI Max+ packs an integrated GPU and shows up in Copilot+ certified machines, but AMD’s Agent Computer pitch leans towards a headless Windows box acting like a server.
AMD’s setup is tuned for LM Studio, using Ubuntu inside WSL to access large language models, then handing off to an OpenClaw server running locally on the same hardware.
