Lenovo Alerts: Cheap Memory Won’t Return Soon, and AI Has Already Changed the Price of PCs

Memory is no longer a cheap component purchased at the end of system configuration. The demand from artificial intelligence, data centers, and the concentration of capacity in higher-margin products are transforming the DRAM and NAND markets with an intensity that already impacts servers, laptops, mobile devices, consoles, and workstations. Lenovo summarized this at ISC 2026 with a provocative phrase: memory prices might never be the same as they were a year ago.

The expression carries a bit of humor and a warning. It doesn’t mean that RAM prices will rise forever or that the market has lost all cyclicality. The memory industry has always navigated between shortages, oversupply, and sharp corrections. But it points to something more structural: even when new factories come online and capacity increases, the “new normal” could be well above the levels seen in 2024 and early 2025.

The reason isn’t just a shortage of DDR5 modules for building PCs. The issue is higher up the supply chain. Memory manufacturers are prioritizing HBM, server DRAM, LPDDR for AI platforms, enterprise SSDs, and long-term contracts with major clients. When hyperscalers and accelerator manufacturers reserve supply years in advance, the consumer market loses priority.

RAMageddon is no longer just forum exaggeration

The term RAMageddon started as a meme, but each month it seems less exaggerated. TrendForce had already revised its forecasts upward for 2026: conventional DRAM was projected to increase by 90-95% in the first quarter, and NAND Flash by 55-60%. For the second quarter, the consultancy expected further strong increases, with DRAM between 58-63% and NAND between 70-75%.

This isn’t limited to the supply chain. Apple has already raised prices on MacBooks, iPads, and other products due to rising memory and storage costs. In PCs, the impact will be particularly harsh on entry-level ranges, where RAM and SSDs constitute a significant portion of the total cost. In servers, the blow is via another route: new platforms demand more memory channels and modules to meet performance promises.

Lenovo proposed a five-step survival guide: review requirements, adjust operations, choose the right CPU, adapt applications, and consider GPU when appropriate. Behind the light tone lies a serious idea. Buying systems based solely on maximum supported RAM is no longer realistic. Memory has become a variable in architecture and budgeting.

Servers with maximum RAM are starting to make less sense

Until recently, many manufacturers boasted about the maximum memory capacity their servers could support. The more terabytes a platform supported, the better its technical specs sounded. That message has become more complex. New CPUs with more memory channels promise high performance, but also require populating enough banks to avoid wasting bandwidth.

In dual-socket systems with 16 channels per CPU, talking about 32 memory banks involves very high minimum configurations if you want to fully leverage the platform. Below 1TB of RAM, many configurations may be unbalanced or not justify moving to a more expensive architecture. The problem is that filling those banks in 2026 costs much more than a year ago.

A paradox arises: GPUs are extremely expensive, but for certain workloads, they can be more efficient if they reduce the need for vast amounts of CPU memory. Lenovo pointed out that for some applications, moving towards GPU-accelerated solutions might be better than trying to solve everything with CPU systems filled with DRAM. It won’t always be true, but the comparison is becoming unavoidable.

AI is favoring the most cost-effective memory

Memory manufacturers are following margins. HBM for AI accelerators, server DRAM, and enterprise SSDs offer more attractive contracts than many consumer products. SK hynix, Samsung, and Micron have clear incentives to direct capacity where clients pay more, sign earlier, and commit to purchases for years.

Micron has signaled this clearly with its strategic supply agreements. The company has signed 16 long-term deals covering a significant part of its future DRAM and NAND volume, typically from 2026 to 2030. These contracts provide visibility for the manufacturer and security for the customer, but they also reduce the amount of memory available on the open market.

Meanwhile, SK hynix has accelerated capacity plans. The company aims to triple wafer production by 2034, advancing a roadmap that originally looked much further ahead. The clear message is: if manufacturers are investing billions in capacity, they’re not expecting prices to crash in just two years.

The old memory cycle hasn’t disappeared but has evolved. For decades, the market suffered from capacity surpluses when everyone invested simultaneously. Now, manufacturers aim to reduce that volatility through contracts, strategic clients, and high-margin products. Memory is becoming less of a cheap commodity and more a critical infrastructure component.

Consumers will share part of the cost

The most visible consequence will be device prices. A low-cost laptop doesn’t have much room to absorb a sharp increase in RAM and SSD costs. If costs rise, manufacturers have few options: raise prices, cut memory, reduce storage, trim other components, or eliminate less profitable models.

This might negatively affect the segments that matter most for education, small businesses, and home users. For years, the assumption was that each generation would bring more RAM and SSD at a lower cost. That logic has been broken. An entry-level laptop with 16GB and 512GB may cease to be standard if costs do not support it. Similar trends could occur in smartphones and tablets: less base storage, higher prices, or more aggressive differences between versions.

The refurbished and secondhand markets might gain significance. Selective upgrades, repair, extended refresh cycles, and leaner configurations will also become more common. For many users, buying “just in case” will lose appeal if each memory upgrade significantly increases expenses.

Companies: it’s time to review architecture, not just budget

In enterprise, the issue can’t be solved by simply negotiating better with suppliers. If memory remains expensive for years, revisiting architecture becomes necessary. Many enterprise applications have been designed assuming abundant and cheap RAM. That premise is now starting to fail.

In-memory databases, analytics, dense virtualization, VDI, over-provisioned caches, and memory-heavy clusters will need better justification. It’s not about making cuts blindly but about measuring actual usage, adjusting reservations, reviewing overcommitment, shifting workloads, leveraging fast storage when viable, and distinguishing between memory essential for operation versus memory used solely for convenience.

Procurement departments will also need to adapt. Instead of waiting for last-minute discounts, large companies will tend to lock in supply agreements earlier. Smaller firms will have less negotiating power and suffer more from volatility. This could widen the gap between hyperscalers, large corporations, and smaller players.

The new normal won’t be cheap

Lenovo’s warning shouldn’t be taken as an exact prophecy. No one knows for sure what will happen by 2030. Memory has often surprised the industry with sudden rises or collapses. But the current trend is clear: AI has shifted supply chain priorities, and new capacity won’t materialize overnight.

If data centers continue to grow, HBM margins stay high, and long-term contracts absorb capacity, consumer electronics will have to learn to compete for memory under less favorable conditions. The outcome will be a less comfortable market for manufacturers and buyers alike.

Cheap RAM was one of the silent foundations of modern computing. It enabled more capable PCs, phones with larger storage, dense servers, and increasingly heavy software. If that phase ends, prices will change, but so will how systems are designed.

For years, the advice was to buy more memory because it was cheap. In this new era, the real question will be: how much memory do you truly need, where should it be, and which parts of your application can function without it?

Frequently Asked Questions

Why are RAM and NAND prices rising so much?
Due to demand from artificial intelligence and data centers, which are drawing capacity towards HBM, server DRAM, and enterprise SSDs, while new supply takes years to develop.

Will prices return to early 2025 levels?
Lenovo believes not in the coming years, although its statement was intentionally exaggerated. The underlying idea is that a new higher price level might form around 2030.

Which products will be affected?
Laptops, desktops, mobile devices, tablets, consoles, servers, workstations, SSDs, and any device relying on DRAM or NAND.

What can companies and users do?
Reassess actual needs, avoid over-provisioning, extend lifecycle where possible, plan purchases strategically, consider refurbishments, and redesign applications that consume memory without a clear necessity.