There was a brief, glorious window in 2023 when PC builders could do no wrong. SSDs had cratered to their lowest prices in a decade. DDR5 kits that had launched at eye-watering premiums were finally sensible. Budget builders were picking up 2TB NVMe drives for less than $80. It felt like the pandemic chip crisis had finally exhaled.

That window is now firmly, possibly permanently, closed.

What's happening to RAM, SSD, and semiconductor supply right now isn't a cyclical hiccup. It isn't the kind of market correction that fixes itself in six months when manufacturers spin up a few more fab lines. IDC called it explicitly in February 2026: this is a "potentially permanent, strategic reallocation of the world's silicon wafer capacity." The trigger was AI, the mechanism is raw purchasing power, and the collateral damage lands squarely on everyone who isn't a hyperscaler.

Understanding how we got here — and where it goes — requires looking at a supply chain that was already fragile and then asking what happens when three or four of the world's wealthiest companies decide they need to own most of it.

The Economics of a Zero-Sum Game

DRAM and NAND flash are manufactured in cleanrooms with extraordinarily high capital costs and multi-year construction timelines. Three companies — Samsung, SK Hynix, and Micron — control roughly 95% of global DRAM production. For decades, their primary customers were PC makers, smartphone manufacturers, and enterprise server vendors. Demand was large but distributed, and those three suppliers played a kind of pricing referee role for the broader tech ecosystem.

AI broke that model completely.

The GPU clusters that train and run large language models require a specialized form of memory called High Bandwidth Memory, or HBM. Producing HBM uses the same underlying DRAM dies as consumer RAM — it just stacks them differently and sells them at margins that make standard consumer modules look like commodity gravel. When OpenAI and Microsoft and Meta come to Samsung with contracts spanning hundreds of thousands of wafers per month, Samsung's finance team doesn't see a difficult decision. It sees an obvious one. Every wafer allocated to an HBM stack destined for a data center GPU is a wafer that doesn't become a DDR5 kit for a gaming PC. That's not hyperbole — that's the literal, physical tradeoff happening inside these fabs right now.

OpenAI's Stargate project reportedly contracted for up to 900,000 DRAM wafers per month from Samsung and SK Hynix, a figure that would represent close to 40% of total global output. Whether the full allocation materializes is somewhat beside the point. The fact that a contract at that scale was even negotiated tells you everything about where the industry's center of gravity has shifted. TrendForce reports that demand for RAM chips currently exceeds supply by 10% — and it's growing faster than manufacturers can respond. Their data showed a 50% price increase in Q4 2025 alone, with another 40-70% projected through 2026.

The consequences cascaded fast. SK Hynix announced in October 2025 that it had already secured buyers for its entire 2026 RAM production. Micron's CEO said supply tightness would continue into 2027. And then Micron did something that would have seemed unthinkable a year prior: it announced it was shutting down Crucial, its 30-year-old consumer memory brand, to redirect all capacity toward "larger, strategic customers." The company that made budget RAM and SSDs accessible to millions of builders and IT departments decided the consumer market wasn't worth the margin trade-off anymore. We went from three major DRAM suppliers serving the consumer market to two, essentially overnight.

This is not just a pricing cycle. It is a structural reorganization of who memory is made for.

What It's Doing to PC Builders and Consumers

If you built a PC in the last year, you've probably already felt this. If you haven't yet, the numbers will reframe what you're looking at when you open Newegg.

DDR5 spot prices have quadrupled since September 2025. TrendForce forecasts conventional DRAM contract prices rising 55-60% quarter-over-quarter in Q1 2026. CyberPowerPC, one of the largest prebuilt PC vendors in the United States, issued a public notice in December 2025 that RAM costs had "ballooned" by 500%, forcing price hikes across their entire system lineup. Taiwanese distributors started requiring OEMs to buy memory only when bundled with motherboards at a 1:1 ratio — you want 100 memory kits, you buy 100 motherboards too. Retailers in Japan's Akihabara district instituted per-customer purchase limits on RAM and SSDs.

The DIY community is absorbing the worst of it. IDC noted explicitly that white-box builders and smaller regional assemblers — the very market segment that gaming culture runs on — will bear the greatest burden. Large OEMs with pre-existing component contracts and supply chain leverage can still move product at relatively controlled prices, at least through H1 2026. But the person trying to build a midrange gaming rig from scratch is buying components on a spot market that changes prices daily. Some retailers have abandoned fixed pricing entirely, listing RAM kits the way a seafood market prices lobster: whatever the catch demands this morning.

The spec degradation angle is perhaps more insidious than the outright price increases. TrendForce's Avril Wu warned that manufacturers will likely pursue what amounts to shrinkflation — holding a laptop's $599 price point while quietly dropping it from 16GB to 8GB of RAM, or swapping a faster panel for a cheaper one. The $600 laptop of 2026 may look identical to the 2025 version and perform meaningfully worse. Counterpoint Research projected that average smartphone selling prices will rise roughly 7% in 2026, while simultaneously some lower-end devices are being cut from lineups entirely because the memory costs to build them make the margin math impossible. Budget and midrange Android phones — the category that serves the widest population of buyers globally — face either disappearing or downgrading components like cameras and displays just to hold a price point.

Dell has signaled price hikes of up to 30% on PCs. Lenovo, HP, Acer, and ASUS have warned clients of 15-20% increases. Gartner projects DRAM prices will rise 47% overall in 2026. The era of cheap, abundant memory and storage that defined the first half of this decade is over, at least for the medium term.

The Enterprise Reality

From where I sit managing over 12,000 endpoints, the shortage has a different texture than it does for the consumer market — but it's no less painful. The consumer feels it at the point of purchase. Enterprise IT feels it across procurement timelines, refresh cycles, and budget models that were built when memory was predictable.

Lead times for standard server DIMMs extended from roughly 25 weeks to 45+ weeks by December 2025. Organizations that once planned hardware refreshes on rolling quarters are now being told that the components they need might not be available until next year, if then. SHI's procurement guidance as of February 2026 is telling IT buyers to plan for a 30-60% price uplift over the January baseline in H1 alone, and to phase deployments carefully — defer anything non-critical to Q3 or Q4 when there's at least a chance of pricing stabilizing.

The smarter enterprise procurement teams saw this coming. Lenovo reportedly built up extra DRAM and NAND inventory by mid-2025, increasing its stock-to-sales ratio by roughly 10% year-over-year in anticipation of what was developing. That move reportedly contributed to a 25% boost in Q3 2025 profits, because while competitors were scrambling, Lenovo was shipping. But that inventory position has a shelf life, and when Lenovo needs to replenish at current market prices, those margins compress again.

The challenge for most enterprise buyers is that they're not Lenovo. They don't have dedicated supply chain intelligence teams watching NAND spot pricing or direct relationships with Samsung's allocation teams. They're relying on their resellers, who are themselves navigating a market where pricing validity windows have shrunk from quarterly to sometimes weekly. A quote that's good for 30 days was the norm eighteen months ago. Now some suppliers are quoting with 48-hour windows.

This creates a pressure dynamic that's particularly brutal for organizations planning Windows 11 refreshes — and there are a lot of them, given the October 2025 end of Windows 10 support. Companies that were already reluctant to replace hardware are now facing a choice between paying the elevated hardware prices and paying for extended security support. Either way, the CFO is unhappy. The AI investment that IT leadership promised would improve efficiency is now competing with hardware refresh spend that just became 20-30% more expensive than it was in last year's budget model.

IDC projects the PC market could contract by nearly 9% in a pessimistic 2026 scenario, with average selling prices rising 6-8%. Even the moderate scenario projects a 4.9% contraction. For IT departments managing multi-year refresh cycles, this isn't just a line item problem — it changes what's feasible.

The Long View, and Why It Doesn't Get Better Soon

The optimistic case for resolution points to new fab capacity. Samsung, SK Hynix, and Micron are all investing in expanded manufacturing. Micron's facility in Boise, Samsung's projects in Texas, and SK Hynix's Ohio investment are real, and they will eventually produce more memory. The timeline is the problem. TrendForce's Avril Wu put it plainly: meaningful new production won't make a noticeable difference in global supply until 2028. The major producers are already selling allocation capacity for 2027 and 2028 to hyperscalers who are themselves operating on multi-year AI buildout roadmaps. New capacity isn't being built to relieve the consumer market — it's being built to serve the customers who are already signing the largest contracts.

The secondary pressures compounding this are substantial. Tariffs and export controls on semiconductors have added friction to global supply chains that were already strained. The concentration of DRAM production in three companies representing 95% of global output is a systemic vulnerability that doesn't go away when prices eventually stabilize — it just waits for the next disruption. And the AI buildout shows no structural signs of slowing. AI capex in 2025 was estimated at more than $400 billion, and projections for 2026 run higher. The hyperscalers are not building toward a ceiling; they're building toward an architecture where model capability scales with compute, and compute scales with memory.

Counterpoint Research's MS Hwang said something worth sitting with: that "there is no limit" to how much manufacturers will pay for memory. When the economic logic of an AI cluster means that securing a DRAM module at twice the normal price is still the right business decision — because the compute cluster it enables generates orders of magnitude more value — the normal market mechanisms that would moderate pricing simply don't function. The AI buyers are not price-sensitive in any way that helps the rest of the market.

There's a bleaker long-term theory circulating in some corners of the industry: that sustained hardware scarcity, combined with rising device prices, accelerates the shift toward thin clients, cloud streaming, and subscription-based computing. If a capable local machine becomes genuinely expensive to build or buy, and cloud-based alternatives keep improving, the economics of personal computing change. Microsoft's investments in cloud infrastructure and streaming, the growth of Chromebook-style devices, and the broader SaaS shift in enterprise already point in this direction. The memory crisis doesn't cause this shift — but it greases the slope considerably.

What's happening in the RAM and SSD market right now isn't a temporary dislocation. It's the visible edge of a deeper reallocation: an entire industry quietly deciding that AI infrastructure is its highest-value customer, and pricing everything else accordingly. For anyone managing a hardware refresh, building a system, or trying to plan an IT budget through 2027, that reallocation is the new baseline. The sooner procurement strategies adapt to it — and stop assuming the cheap memory era is coming back — the better positioned the response will be.

The cheap storage and abundant memory that defined the last five years are gone. The question now isn't when they return. It's what computing looks like when they don't.