During its second-quarter earnings call on July 30, Samsung Electronics dropped a sobering outlook for anyone planning a PC upgrade or enterprise hardware refresh. The company’s memory sales and marketing lead, Kim Jaejune, told investors that the global supply-demand imbalance for DRAM and NAND will become “more severe in 2027 than 2026,” with little chance of a meaningful production increase before 2028. The warning lands as both businesses and consumers are already grappling with elevated RAM and storage costs driven by an AI-fueled datacenter build-out.

What Samsung Actually Said—and Why It Matters

Samsung’s warning wasn’t a vague hand-wave. Kim laid out a concrete timeline: the semiconductor industry’s long production cycles mean a three-year lag between deciding to build new capacity and seeing chips come off the line. Even with aggressive investment, “it will be unlikely to see any significant increase in incremental supply through 2028.” Unmet customer demand from this year, he noted, will roll into 2027, keeping the market tight.

The prediction aligns with SK Hynix, which earlier characterized 2027 as potentially “the worst year in the industry’s history from the supply perspective.” Together, the world’s two largest memory makers are signaling that the shortage isn’t a temporary spike but a structural squeeze driven by insatiable appetite from cloud and AI hyperscalers. These buyers are already negotiating multi-year supply agreements, locking up capacity for high-bandwidth memory (HBM) and conventional server DRAM alike.

That’s a crucial detail: HBM, the specialized stacked memory that sits alongside AI accelerators, is not a separate manufacturing universe. It competes for the same fab space, equipment, and engineering resources as the DDR5 and LPDDR5X modules inside your laptop or desktop. When a hyperscaler orders millions of HBM stacks, it pulls production capacity away from standard components, tightening the whole market.

What the Shortage Means for You

The impact on Windows users and IT buyers breaks down by role:

For home users and PC enthusiasts
If you’re holding out for a price drop on a 32GB DDR5 kit or that 2TB NVMe SSD, don’t hold your breath. Samsung’s forecast suggests that by the time Black Friday 2027 rolls around, memory prices may not look much cheaper than they do today. Expect OEMs to keep base configurations at 8GB or 16GB for another product cycle, while premium upcharges for higher capacities remain stubbornly high. Budget builders may find fewer bargains; people who need RAM for content creation or local AI workloads should price it into their next PC purchase now.

For IT administrators and procurement teams
A 2027 endpoint refresh that assumes commodity DRAM will follow its usual downward cost curve could blow a hole in your budget. Enterprise laptops, workstations, and thin clients often come with soldered memory, leaving zero upgrade path. Locking specifications into requests for proposal (RFPs) early—including floor RAM capacities of 16GB or 24GB, with pricing commitments—becomes essential. Organizations with fleets still running 8GB Windows 11 machines should audit and remediate sooner rather than later; the longer you wait, the more expensive those stopgap upgrades become.

For developers, data scientists, and AI tinkerers
Local compute is eating RAM like never before. Running quantized LLMs, containerized dev environments, or virtual machines on a notebook with less than 32GB is increasingly painful. Treat high-memory workstation purchases as capacity-constrained equipment, not routine PC orders. Consider that even external GPU enclosures or eGPU cards won’t save you if your system’s system memory is the choke point.

Samsung’s own device business illustrates the pain: the mobile division posted its first-ever operating loss this quarter, partly because memory and component costs are rising faster than the company can pass along to consumers. If Samsung struggles to insulate its own phones and tablets from the shortage, smaller OEMs and aftermarket buyers have even fewer levers.

How We Got Here

The current memory bottleneck traces back to two overlapping forces. First, the AI boom created a new class of memory-hungry hardware. Training and inferring large models requires not just GPUs but massive pools of fast memory. HBM, in particular, has seen demand explode; it’s a must-have component in Nvidia’s H100 and B200 platforms, and both Samsung and SK Hynix are pouring capital into expanding HBM production.

Second, memory manufacturing is inherently lumpy. Building a new DRAM or NAND fab is a multi-year, multi-billion-dollar commitment. During the pandemic-era shortages, chipmakers ramped up, but the subsequent PC market slowdown in 2024 and 2025 caused them to dial back expansion plans. Now, as AI demand soars, they’re caught in a lag—new fabs ordered today won’t contribute significant volumes until 2029 or later. Samsung’s Kim hinted at this: “Beyond 2029, it is hard to say because of more limited visibility,” but until then, the supply pipeline is essentially fixed.

Meanwhile, the conversion of existing production lines from conventional DRAM to HBM is not frictionless. It requires retooling, and even hybrid lines face yield trade-offs. That means every wafer dedicated to HBM is a wafer not producing DDR5 or LPDDR5X. For the PC ecosystem, which competes with servers and smartphones for the same silicon, the result is a sustained cost floor.

What to Do Now

Samsung’s forecast hands buyers a rare gift: a 12- to 18-month warning. Here’s how to act on it:

  • If you manage corporate device fleets: Freeze your 2027–2028 hardware specifications this quarter. Negotiate RAM and storage configurations as fixed-cost line items in vendor contracts, not variable add-ons. Where possible, demand socketed rather than soldered memory to preserve future flexibility. Evaluate leasing or Device-as-a-Service models that shift the upgrade risk to the vendor.
  • If you’re a professional or power user: Order your next workstation or laptop with the maximum RAM you’ll need over its 3–4-year lifespan. Don’t bet on a cheaper aftermarket upgrade in two years. If you’re in the market for a PC now, consider buying barebones and sourcing memory before prices inch higher still.
  • If you’re a home user with a modest budget: Lower your expectations for future price drops. If your current system can squeak by with a RAM or SSD upgrade, do it while today’s pricing holds. For new builds, choose an AM5 or LGA 1851 motherboard with four DIMM slots so you can start with 16GB and add a matched kit later if prices loosen—but don’t count on that happening soon.
  • Watch for OEM design shifts: Manufacturers may respond by pushing thinner, soldered-down designs that ship with “just enough” memory to hit a price point. If you see a laptop you like, check whether the RAM is user-upgradable; it’s a health indicator for the supply environment. A wave of sealed-down designs often signals that OEMs are trying to protect margins from soaring component costs.

Outlook: No Quick Fix

Samsung’s forecast doesn’t end in 2028. Kim said that even beyond that horizon, visibility is poor, but the company is already negotiating multi-year supply deals with AI customers—suggesting that demand will remain elevated for the foreseeable future. The next generation of AI inference that runs locally on device, a cornerstone of Microsoft’s Copilot+ PC push, only adds more pressure: those machines lean heavily on fast, unified memory.

For Windows users, the memory shortage is no longer a distant supply-chain abstraction. It’s a line item in your next shopping cart and a budget variable in your next fiscal year. Act before the numbers get worse.