The Global Memory Crunch: DRAM and NAND Price Spikes Reshape Computing Markets

The Memory Squeeze: Why DRAM and NAND Prices Are Soaring

 

Global prices for memory chips – both DRAM and NAND flash – have surged due to a perfect storm of supply and demand factors. The booming appetite for AI and cloud computing has semiconductor firms prioritising their limited production capacity for high-margin, advanced memory like HBM (High Bandwidth Memory) and cutting-edge DDR5 RAM, at the expense of mainstream products Simply put, tech giants are buying out entire years’ worth of memory supply for AI servers, which leaves fewer crumbs for consumer PCs and gadgets. At the same time, older standards like DDR4 are being wound down – Samsung even signalled an end-of-life for DDR4, causing supply to drop before many users were ready to move on And after the last boom-bust cycle, manufacturers like Samsung, SK hynix, and Micron remain cautious about overbuilding capacity, preferring profitability over flooding the market The result is a global memory squeeze years in the making, now reaching an inflection point.

The numbers tell the story. DRAM prices surged as much as 60% in 2025, and analysts forecast another 30–40% hike in 2026 Spot prices for certain chips are spiking even faster – server-grade DDR5 saw up to a 100% jump in some cases and some premium mobile DRAM parts more than doubled in price during 2025 NAND flash (used in SSDs) isn’t far behind; contract prices for flash memory jumped 20–60% in a single month late in 2025 In fact, industry reports warn that by mid-2026 there may be “no stock left to ship” for some flash products if nothing changes This isn’t a temporary blip – memory vendors and market watchers now talk of a multi-year “super-cycle” extending possibly to 2028, with chronic undersupply “baked in” until major new fabs come online.

 

Why is this happening? Several key drivers are behind the crunch:

 

• AI Demand Cannibalising Capacity:

  The AI boom has turned memory into the new oil. Hyperscale data centres building large AI training clusters require massive amounts of memory, both HBM stacks attached to GPUs and loads of DDR5 in host servers. An AI server with a set of high-end accelerators can need 3–4 terabytes of memory (HBM + DDR5) in total, whereas a regular server might use under 1 TB To meet this, Samsung, SK hynix, and Micron are reallocating more wafers to HBM and high-density DDR5 production. But HBM is laborious to make – it can take four times as many wafers per gigabyte as standard DDR memory So every wafer shifted to HBM for AI is a wafer not making commodity DRAM, worsening the shortage of regular PC memory AI giants are effectively outbidding everyone else: for instance, OpenAI struck deals with Samsung and SK hynix to output up to 900,000 high-bandwidth DRAM chips per month for its projects – a volume that would double the entire industry’s HBM capacity When one customer (AI) gulps down that much, little is left for others.

• End of an Era for Older Tech:

  Memory makers are accelerating the phase-out of legacy products like DDR4 and older NAND nodes. DDR4, which still powers countless PCs and servers, is quietly fading. Samsung shocked the market by announcing plans to EOL (end-of-life) DDR4 last year (later pausing formal EOL after pushback) but production has undeniably slowed. ChangXin (CXMT) in China also reportedly halted most DDR4 output As a result, DDR4 chips and modules are getting scarce and pricey despite being “old” tech – inventories for some DDR4 parts have dwindled from 13+ weeks down to just 2–4 weeks of stock Similarly, certain older flash memory products are in “historic shortage”, with some manufacturers receiving no allocations of legacy chips This puts PC builders and device makers in a bind: either redesign for newer memory (which might not be feasible or cheap), or scramble to hoard the last batches of older RAM. The transition to DDR5 and next-gen NAND is happening faster than the ecosystem might be ready for, adding instability.

 

• Profit-First Strategy by Suppliers:

  After years of volatile boom-bust pricing, the big three memory vendors are now unapologetically putting profit margins first. We’ve seen this playbook before – in the 2010s, memory firms shifted output to chase booming smartphone demand; in 2018 it was crypto-mining GPUs hogging memory; today it’s AI By concentrating on higher-margin products (HBM, enterprise SSDs, etc.) and holding back on capacity expansion, they have successfully driven up prices. In late 2025, SK hynix and Samsung’s memory divisions actually earned higher margins than TSMC’s cutting-edge chip foundry business – a testament to how tight supply and premium pricing have paid off for them. With smartphone and PC demand soft, these companies see little incentive to swing supply back toward consumer-grade chips Instead, they are content to enjoy improved profitability while the rest of the industry copes with the fallout. All major suppliers have shifted to shorter-term sales and even “allocation only” ordering, meaning customers are restricted to contracted volumes with no flexibility This protects the vendors’ profits against any oversupply, but leaves smaller buyers out in the cold when they need extra memory. In short, the big memory makers are intentionally keeping the market on a tight leash.

 

Micron’s Consumer Exit: Crucial Brand Bows Out

Micron’s Crucial-brand DDR5 memory modules – once a staple for PC DIY upgrades – will be discontinued as Micron pivots to serving data centres and AI clients.

In a move that sent shockwaves through the PC building community, Micron Technology announced it is winding down its Crucial consumer memory and storage business by February 2026. The Crucial brand – known for nearly 30 years of affordable RAM sticks and SSDs for DIY enthusiasts – is being sunsetted so that Micron can redirect those resources toward enterprise and AI customers who are clamouring for more memory. Micron’s Executive VP Sumit Sadana put it bluntly: “Micron has made the difficult decision to exit the Crucial consumer business in order to improve supply and support for our larger, strategic customers in faster-growing segments.” Those “strategic” customers are the big fish in AI and cloud services driving today’s demand surge.

 

For consumers, this is a bittersweet milestone. Crucial has long been a trusted middle-ground option – not always the absolute cheapest, but known for solid quality and wide availability (after all, who hasn’t used a Crucial stick of RAM in a budget build at some point?). Now, that capacity is being reallocated to higher-margin enterprise products. Micron admitted it simply cannot produce enough memory to satisfy both the exploding AI market and the consumer/DIY market, so something had to give From a business standpoint, the pivot makes sense: memory for AI training rigs or data centre servers commands far better prices than a $40 consumer DIMM. One analyst noted that an AI accelerator can require memory modules by the thousands per deployment, meaning AI clients buy in volumes that dwarf the needs of any PC gamer So, Micron is choosing to serve the few customers who buy by the pallet, not the many who buy by the stick.

 

The immediate impact is one less major player supplying consumer memory, which tightens an already constrained market Distributors and PC OEMs that relied on Crucial-branded chips will have to find alternatives, likely turning to third-party module brands or Micron’s remaining OEM channels. Micron has promised to continue honouring warranties and will ship products through the Crucial channel until Feb 2026 to smooth the transition But after that, DIY buyers will be left mainly with brands like Kingston, Corsair, G. Skill, and Samsung’s retail offerings. Those companies may see surge in demand for their RAM and SSDs, but it’s unclear if they can fully fill the void – especially since they all ultimately source chips from the same few factories In fact, Micron’s exit could put extra pressure on “non-enterprise” DRAM and SSD supplies across the board

 

Micron’s decision is emblematic of a larger industry shift. It shows how quickly AI spending is reshaping priorities: memory manufacturers are willing to abandon decades-old consumer businesses – and the goodwill of PC hobbyists – in favour of lucrative AI contracts For PC builders, this translates to fewer choices and likely higher prices. As one commentator noted, “consumer memory is becoming a low-priority afterthought as chipmakers chase AI gold” The irony is that the very AI technologies promising to revolutionise computing are indirectly making basic components less accessible to everyday users Micron’s Crucial exit is more than just the end of a brand – it’s a stark signal that memory industry support is tilting away from the consumer realm.

 

No Quick Fix from Samsung or SK hynix

One might think that when a big player like Micron pulls back, competitors would rush in to grab the market share. But in the memory industry’s new paradigm, Samsung and SK hynix are not stepping up to flood the market with cheaper chips – quite the opposite. Both South Korean giants have indicated they will not significantly increase production of commodity DRAM or NAND to fill the supply gap. No major capacity expansions have been announced despite the severe shortage. Instead, these firms are also laser-focused on high-end products and maintaining healthy profits.

 

Samsung, for instance, has been deliberately limiting supply and even notifying customers of price hikes. Industry reports say Samsung is only fulfilling about 70% of incoming orders for certain memory products, essentially rationing output The company’s own mobile division has had to source mobile DRAM externally (even buying from Micron) because internal capacity was diverted elsewhere Likewise, SK hynix has been converting a huge chunk of its “commodity” DRAM manufacturing to its latest process node, which actually slowed its output in the short term – and even then, they acknowledged it “still won’t fully meet customer demand.” SK hynix temporarily halted shipments and renegotiated contracts at ~40% higher prices as part of this pivot In other words, the top suppliers are perfectly comfortable with a tight market. They are cautious of overbuilding (after learning painful lessons from past gluts) and prefer to enjoy the current seller’s market. As one analysis put it, this “conservative capacity expansion” approach means the industry will be in a state of perpetual undersupply for the foreseeable future.

 

It’s not that no new factories are coming – Samsung and others have fab projects in the pipeline – but these are long-term solutions. Building new memory fabs takes years and billions of dollars. Meaningful new supply from recently announced projects isn’t expected until late 2026 at the earliest, more likely 2027 or 2028 Micron, for one, has accelerated plans for a new mega-fab in New York and an expansion in Idaho (with government incentives), but even its “most aggressive” roadmap doesn’t bring significant output online before 2027 Samsung and SK hynix similarly have no immediate stopgap. In fact, Samsung’s memory division has explicitly avoided committing to any big near-term capacity jump for NAND flash, opting to ride out the shortage with premium pricing to favoured customers instead. This stance carries a risk – they could be missing out on even more sales now – but it also avoids creating a future oversupply crash. From their perspective, it’s better to have demand exceed supply (and thus support high prices) than to overshoot production.

 

In short, don’t expect relief from the supply side. The top memory makers are essentially saying: “We will allocate what we have to our most important (read: highest paying) clients, and everyone else will have to deal with it.” This means the memory crunch isn’t likely to ease until existing producers ramp up in a big way or new players enter the market – neither of which is happening soon. Even China’s emerging memory firms (like CXMT or YMTC) are not significantly alleviating the gap; one Chinese DRAM maker has largely exited DDR4, and their moves into DDR5 will take time All signs point to a continued discipline in supply from the major vendors. They are effectively acting to keep the market tight. For them, it’s a logical strategy: the limited supply has driven record profits, so why open the floodgates now? The unfortunate flip side is continued high prices and shortages for consumers and many businesses around the world.

 

Impact on Consumers and PC Makers

For everyday consumers and PC manufacturers, the memory shortage is already hitting home. Prices for PCs, laptops, and DIY components are on the rise globally as memory costs skyrocket. Research by IDC and others suggests average computer prices could climb by up to 8% in 2026 solely due to memory chip shortages Major PC vendors – including Lenovo, Dell, HP, Acer, and ASUS – have all warned customers to brace for higher costs in the second half of 2026, with some planning 15–20% price hikes on systems to offset the surging RAM and SSD prices Dell’s COO even remarked he’s “never seen memory-chip costs rise this fast” in his career Such steep increases are forcing PC makers to make tough choices. In some cases, manufacturers have responded by downgrading the specifications of the products they ship – for example, a laptop model that used to come with a 512GB SSD might now be sold with only 256GB, or an 16GB RAM configuration might be cut to 8GB in base models, just to keep prices palatable Reports indicate standard 512GB SSDs being reduced to 256GB, and 1TB drives down to 512GB, across many mainstream PC lines Even then, many PC brands have signalled they will need to raise prices as the component cost pressure becomes untenable.

 

DIY enthusiasts and smaller “white box” PC builders are arguably feeling the squeeze the most. They don’t have the long-term contracts or buying power that big OEMs do, so they rely on the spot market or retail channels, which have dried up or become very expensive. Some smaller PC integrators say this is the toughest sourcing environment in a decade In fact, industry insiders note that lower-tier and local PC assemblers will bear the greatest burden of the shortage, as they struggle to secure memory at reasonable prices This presents an opportunity for large OEMs (Dell, HP, etc.) to gain share, since they can still get memory (albeit at higher cost) whereas mom-and-pop shops or DIY builders might simply be priced out or face long waits As an example, popular DIY-friendly company Framework – known for its modular, repairable laptops – had to stop selling standalone RAM upgrade modules on its site, because it anticipates prices going even higher and supply getting tighter in coming months Gaming PC boutique vendor CyberPowerPC similarly announced surcharges on memory upgrades due to soaring RAM costs These examples are not outliers, but early indicators of how tightening memory supply is affecting different parts of the PC ecosystem at different speeds.

 

Even retailers are struggling: reports surfaced of unscrupulous sellers resorting to counterfeits, such as disguising older DDR4 sticks as DDR5, to exploit unwary buyers in a shortage – a scam that highlights just how desperate the situation has become in the consumer channel

 

Globally, consumers are thus facing a double whammy: either pay significantly more for the same PC hardware, or get a downgraded spec for the same price as before. Market analysts now predict the PC market (by unit shipments) will actually shrink in 2026 as a result of these shortages and price increases – one estimate forecasts a decline of around 2.4% to 8.9% in PC shipments, whereas prior forecasts had expected modest growth In other words, some consumers and businesses will delay or cancel PC purchases because the memory shortage has made them too expensive or hard to get. Alarmingly, Hewlett Packard and Dell have even indicated they might drop certain laptop models entirely in the next year, if those models would require memory they can’t afford or secure enough of This kind of drastic measure underscores how severe the crunch is – imagine a PC maker saying “we won’t sell Model X anymore because we can’t get enough RAM for it at a viable cost.” It’s almost surreal, but it’s the new reality in this constrained environment.

 

What this means in practice is that not all PC builders are affected equally. Manufacturers that plan component purchasing well in advance, maintain diversified supply relationships, and avoid short-term spot buying are better positioned to absorb volatility in memory pricing. While no builder is immune to sustained price inflation, those with disciplined procurement strategies can often soften the impact for customers by maintaining more stable pricing across upgrades and full system builds.

 

Impact on Business and Enterprise Computing

While consumer PCs are directly feeling the pain, the enterprise and cloud computing sector is both a cause of the crunch and a victim of it in some ways. Big data centre operators – the cloud service providers, hyperscalers, and AI firms – are the ones driving the demand shock, and they generally get first dibs on supply. Memory manufacturers are prioritising these large customers, as evidenced by deals like the OpenAI-Samsung partnership for huge monthly DRAM wafer allocations Hyperscalers like Google, Amazon, Microsoft, as well as AI firms building training clusters, have been stockpiling memory chips years in advance to fuel their expansion plans As a result, enterprise and server-grade memory (e.g. high-capacity DDR5 RDIMMs, HBM stacks) is effectively sold out long before it’s even made – one report noted that HBM (the specialised memory for AI accelerators) is completely booked up through 2026, with capacity “sold out years ahead” due to multi-year procurement by big players Cloud companies are willing to pay top dollar; memory makers have publicly said they can only meet about “half to two-thirds” of the demand from their key AI customers, even running flat-out

Enterprise buyers, unlike consumers, can’t simply postpone critical deployments – they need memory to expand their data centres or roll out new services. So, they are paying whatever it takes, which in turn reinforces the upward price spiral. For instance, enterprise SSD prices jumped over 25% quarter-on-quarter in late 2025 and server DRAM contract prices have been resetting higher frequently (sometimes mid-quarter) as suppliers renegotiate to account for the tight market Many large enterprises have long-term agreements, but even those are shifting towards shorter-term or “floating” pricing because no one wants to be locked into a “too low” price when the market is moving up so fast Some data centre customers are reportedly on allocation – meaning they get a fixed amount and can’t buy more on the spot market even if they want to. This has led to scenarios where AI projects are delayed, not due to lack of chips like CPUs or GPUs, but due to lack of memory modules to plug into servers.

 

Business IT departments outside the cloud giants are also impacted. A medium-sized company looking to upgrade its servers might find that memory upgrades are easily double the cost they budgeted a year ago NetworkWorld reported that server memory prices could potentially double by 2026 for enterprise buyers, given the combination of 50% rises already seen and more on the horizon Companies that planned hardware refreshes to coincide with events like Windows 10 end-of-life (which typically drives PC upgrades) are encountering this “perfect storm” of cost inflation Some may choose to delay upgrades, run existing hardware longer, or accept lower memory configurations – which could have knock-on effects on performance or capacity for their workloads.

 

On the flip side, enterprise demand being so strong means these sectors are somewhat shielded – the largest enterprises will generally get the memory they need (since suppliers favour them), but at much higher expense. This ultimately could translate into higher costs for cloud services and enterprise products down the line. If a cloud provider pays 2-3x more for memory, that cost either cuts into their margins or gets passed to customers eventually. In essence, enterprise computing is gobbling up the world’s memory supply and paying a premium to do so, while inadvertently making memory pricier for everyone else. Data centres now account for roughly 50% of total DRAM bit demand worldwide (even before counting this latest AI surge), and AI workloads alone make up about 30% of that demand Those figures illustrate why enterprise trends dictate market pricing: when half of all memory produced goes into servers and AI boxes their willingness to pay more sets the price floor for everyone.

 

It’s worth noting that not all enterprise segments are equal in this crisis. Cloud and AI projects are the priority – for example, automotive and industrial clients (who also need memory) are important but can be charged more or made to wait longer, since they are smaller portion of the pie Some enterprise IT buyers are trying to sign longer contracts or buy ahead to secure inventory, which further fuels the cycle of scarcity as everyone orders more than they immediately need, just in case In summary, the enterprise computing sector is heavily driving the memory crunch, and while it has the means to obtain supply (thanks to deep pockets), it is also grappling with unprecedented lead times and costs.

 

Outlook and Advice: Act Sooner Rather Than Later

Looking forward, the consensus is that this memory shortage and pricing boom won’t abate quickly. With demand from AI and data centres still red-hot and no major new supply coming online for at least another year or two, industry experts anticipate elevated prices and tight availability through 2026 and into 2027 In fact, some analysts go as far as saying the industry may operate in a “perpetual undersupply” mode for the foreseeable future – a complete flip from the oversupply gluts of past years. This means consumers and businesses need to adapt their strategies. If you’re a consumer or small business in need of a memory upgrade or a new PC, waiting for prices to drop is not a safe bet right now. The old advice of “RAM and SSD prices always fall over time” does not hold during this rare super-cycle. Cameron Crandall, an executive at Kingston (a major memory module manufacturer), advised PC upgraders plainly: “do it now and not wait because prices are going to continue to go up… it will be more expensive 30 days from now, and more likely more expensive 30 days after that.” In other words, buy sooner rather than later if you need to buy at all.

From a system builder’s perspective, the current environment places a premium on forward planning. Memory procurement is no longer a just-in-time exercise; it increasingly requires long-term forecasting, volume commitments, and a willingness to carry inventory during periods of uncertainty. Builders that adopt this approach are better able to maintain continuity in system specifications and reduce the frequency of reactive price adjustments.

 

For consumers, another tip is to manage expectations and consider alternatives. You might opt for a slightly lower-spec machine that’s within budget – e.g., 16GB of RAM instead of 32GB – with the plan to upgrade later (though upgrade memory might be costly for a while). Keep an eye on reputable retailers and consider setting price alerts; occasional temporary inventory might appear and you’ll want to snag it quickly. Also, be wary of deals that look too good to be true (like high-end DDR5 at suspiciously low prices) – as mentioned earlier, there have been cases of counterfeit or misleading listings popping up amidst the chaos It’s safer to stick to known brands and authorised sellers during shortages.

 

For businesses and larger IT purchasers, planning and communication are key. Companies should forecast their memory and storage needs further ahead than usual and secure supply contracts if possible. It may be wise to lock in a portion of future needs through agreements with suppliers or distributors, even if it means paying a bit of a premium, to avoid total shortfalls. IT departments might also evaluate whether all projects needing new hardware are truly urgent – if a non-critical upgrade can be deferred to late 2026 or 2027, it might save money (assuming by then some new fabs relieve the pressure). However, given current projections, prices in 2027 could still be high, just perhaps not climbing as crazily as now. Some PC OEMs are already coordinating closely with their component suppliers and even redesigning products to tolerate a mix-and-match of memory (to use whatever is available) Flexibility and having contingency options (like qualifying multiple memory vendors or modules) can help mitigate risk in this environment.

 

In conclusion, the surge in DRAM and NAND prices – driven by a confluence of AI-driven demand, strategic supply decisions, and transitional growing pains – is transforming the landscape of the computing market worldwide. Consumers are facing higher costs and fewer choices, businesses are adjusting their plans and budgets, and the industry as a whole is refocusing on where the money is (AI and enterprise). Until new capacity comes online or demand stabilises, memory will remain a precious commodity. If you have critical computing needs that require more memory or storage, it’s prudent not to delay those purchases too long. For buyers considering an upgrade or a new system, this also highlights why pricing differences between builders may widen over the coming months. In a constrained market, stability is often the result of preparation rather than luck. Builders that have secured memory supply ahead of sustained price rises are typically better placed to offer consistent configurations and more predictable pricing, even as broader market costs continue to climb. The era of dirt-cheap RAM and abundant SSDs will return eventually, but not before this cycle runs its course. In the meantime, staying informed and proactive is the best way to navigate the memory shortage. As the saying goes in tech procurement circles: the best time to buy needed memory was yesterday – and the second best time is today