THE CRUNCH
Microsoft is pressing distributors, resellers and enterprise buyers to treat PC refreshes as part of their AI budgets rather than a separate line item. Speaking as first-time headline sponsor of the Canalys Forum in Barcelona, Mark Linton, corporate vice president for Windows and Device Sales, argued that running smaller AI models locally on Windows PCs avoids the cloud token costs that make frontier-model agents expensive. The pitch lands in a difficult market: IDC figures put calendar Q3 PC shipments at 62.7 million units, down 20.1 percent year on year, and Microsoft's Windows OEM and Devices revenue fell 22 percent to $3.55 billion in the quarter ended September 30, with the company warning Q2 revenue may drop a further 30 percent.
The obstacle is cost. Memory chipmakers have prioritised higher-margin HBM for AI datacenters over conventional DRAM, pushing up prices just as buyers were already hesitant about AI PCs' higher costs and thin killer apps. Omdia, which owns Canalys, said upstream component cost increases have now fully worked through to end-market pricing, with consumers and IT decision-makers responding by delaying purchases.
Linton's answer is a hybrid model: powerful PCs with ample memory run suitable AI workloads on device, keeping data local, cutting token costs and improving latency. He framed Copilot as only one component, positioning Windows as a platform for all models and agents, with management tools such as Agent 365, Intune and Entra, plus the June-announced Microsoft Execution Containers for virtualising and policing agents at operating system level.
The channel is being courted with money as well as messaging. Microsoft is funding AI-readiness headcount at top partners and encouraging resellers to use AI internally so their sales teams can sell it. Dell, meanwhile, expects commercial PC demand to stay compressed this year, though its CFO David Kennedy pointed to more than 400 million PCs over four years old as pent-up demand once budgets settle.