We entered 2026 expecting a familiar device refresh strategy wave: Windows 10 end-of-support, the Artificial Intelligence (AI) Personal Computer (PC) push, and a predictable lifecycle rhythm. Instead, that device refresh cycle is colliding with a memory market that is not behaving like a cycle at all. Supply and supplier attention are being reallocated up the stack, and endpoints are absorbing the memory-scarcity tax.
Early signals are explicit
- Micron has stated that Dynamic Random Access Memory (DRAM) shortages could persist for quite some time
- Intel has confirmed there are about 9-12 months of laptop inventory before higher memory costs show up more directly in pricing and configurations
- Dell has pointed to the need for targeted pricing actions, while Lenovo’s device leadership has called out unprecedented cost increases, particularly across memory and Solid-State Drives (SSDs)
Enterprises are already reacting. Many are pulling forward Windows 10 EOS-critical cohorts to lock in supply and pricing, while deferring non-critical cohorts to protect budgets. Device refresh strategies are also shifting from age-based replacement to experience-based replacement, using Digital Employment Experience (DEX) telemetry to identify and retire the worst-performing devices first. This approach can stretch parts of the fleet from three or four years toward five, and in some cases six, but it is not a universal answer.
Reach out to discuss this topic in depth.
Why this flips the endpoint playbook
For workplace leaders, refresh strategy is no longer an IT calendar decision. It is a portfolio design decision. As DRAM and SSD economics turn volatile, the question shifts from which device do we buy next to where should compute live.
The instinctive answer is often a single word: cloud. Jeff Bezos framed this direction in 2024 using a pre-grid analogy. Before electricity grids, hotels and factories ran their own generators. He argued that computing is at a similar stage today, where everyone believes they need their own data centers, and that this model will not last. Compute, he suggested, will increasingly be bought off the grid.
Translated to endpoints, the model is straightforward. The PC becomes peripherals plus access: a screen, keyboard, mouse, identity, and a Windows session running in the cloud.
We can already see this ecosystem forming:
- Microsoft introduced Windows 365 Link as a purpose-built Cloud PC device
- Microsoft is consolidating access patterns by retiring the Remote Desktop app and directing customers to the Windows App for Windows 365 and Azure Virtual Desktop (AVD)
- AWS launched WorkSpaces Thin Client as a low-cost, managed endpoint for virtual desktop environments
Momentum, however, does not equal universality.
Why cloud PC fails as a blanket strategy
The workplace is not a single fleet. It is a collection of personas with different tolerances for latency, offline work, peripherals, and performance spikes. This is why cloud PC strategies repeatedly break in execution when applied universally.
Three boundary conditions that matter:
- Experience physics: latency sensitivity and offline requirements are non-negotiable for many roles
- Unit economics: cloud PC shifts spend to Operational Expenditure (OpEx); without discipline, costs sprawl quickly
- Constraints: application dependencies, data residency, and regulated workflows force carve-outs
The strategic question, therefore, is not whether to move compute to the grid, but where compute should live by persona and workload, local, cloud, or hybrid, and what guardrails are required to prevent cost or experience drift. Exhibit 1 illustrates how compute placement decisions vary across common enterprise personas.
Exhibit 1: Persona-led compute placement map
