A basic 16GB stick of DDR5 that cost around 45 dollars in late 2025 now runs well past 100 dollars in many markets. Laptop makers have quietly bumped prices twice this year. Apple raised the cost of MacBooks and iPads and pointed straight at memory and storage. None of this is a coincidence, and none of it is temporary noise in the supply chain. It is the visible surface of a structural fight over a resource that both your phone and a Nvidia data center need in order to function: DRAM.
The short version is that AI data centers are eating the world’s memory supply, and everyone else is left fighting over what’s left. The longer version involves an industry with only three real players, a decades-old boom-bust cycle, a class-action lawsuit accusing those three players of collusion, and a shortage that most analysts now expect to run into 2027 or 2028. This piece looks at how bad it actually is, why it happened, who’s angry about it, and where the memory market goes from here.
The Numbers Behind the Panic
DRAM prices have not crept up. They have spiked in a way that even veteran industry analysts describe as unprecedented. Counterpoint Research tracked an 80 to 90 percent jump in DRAM prices in a single quarter earlier this year. TrendForce, the Taiwan-based research firm that tracks memory pricing closely, recorded roughly a 90 percent quarter-over-quarter surge in Q1 2026 alone, followed by another round of increases in Q2. Some LPDDR5X modules used in phones and laptops jumped 89 percent in one quarter. DDR4, a memory type most people assumed was fading into irrelevance, saw prices climb more than 50 percent in the same window, and spot prices for it were up over 2,000 percent year over year at one point.
Put plainly: this is the sharpest memory price movement the industry has seen since the pandemic, and by several measures, it’s worse.
The pain isn’t limited to gamers building PCs. IDC estimates that memory now makes up 15 to 20 percent of the total bill of materials for a mid-range smartphone. When that input cost doubles or triples, phone makers either eat the margin hit or pass it straight to buyers. Most are choosing the second option. Cisco reportedly warned investors about margin erosion tied to memory costs. Tesla’s Elon Musk described the company as facing a “chip wall,” framing the choice bluntly as hit the wall or build your own fab. When a company with Tesla’s cash reserves says it’s struggling to secure memory allocation, that tells you something about how tight the market has gotten.
So Why Is This Happening
The core answer is AI, specifically a type of memory called high-bandwidth memory, or HBM. Unlike the standard DRAM in a laptop, HBM is built by stacking 12 to 16 layers of memory dies into a single “cube” that sits right next to the GPU, feeding it data fast enough to keep up with AI training and inference workloads. Every Nvidia, AMD, or custom AI accelerator chip going into a data center right now needs a pile of this stuff, and HBM has become one of the highest-margin products in the entire semiconductor industry.
Here’s the part that actually explains the shortage on your side of the counter: HBM and ordinary DDR5 come from the same wafer capacity and the same fabrication lines. When Samsung, SK Hynix, or Micron devote a production run to HBM, they are not adding capacity, they’re reallocating it. Industry estimates put the conversion ratio at roughly three bits of standard DRAM given up for every bit of HBM produced. Multiply that across an industry where the big three have reportedly shifted 80 to 93 percent of their combined production toward HBM and AI-linked server memory, and you get a very simple, very brutal math problem. There just isn’t enough capacity left over for the phones, laptops, and consoles everyone else needs.
This is not a normal cyclical dip. Thomas Coughlin, a longtime memory industry analyst, traces the roots of the current squeeze back to the COVID-19 era, when hyperscalers like Amazon, Google, and Microsoft stockpiled memory to avoid supply chain disruptions, a move that primed the market for exactly this kind of imbalance once AI demand hit. New DRAM fabs cost 15 to 20 billion dollars and take 18 months or more to come online, so manufacturers are historically reluctant to expand aggressively, especially given how violently this industry has crashed before. IDC now frames this less as a temporary shortage and more as a permanent, strategic reallocation of the world’s wafer capacity away from consumer devices and toward AI infrastructure. That’s a heavier claim than “shortage,” and if it holds, it changes how every device maker on the planet has to plan.
Genius Business Move or Coordinated Price Gouging
Not everyone accepts the “pure supply and demand” explanation, and that disagreement has now landed in federal court. On June 25, 2026, seventeen plaintiffs, a mix of individuals and small PC retailers, filed a class-action lawsuit against Samsung, SK Hynix, and Micron in the U.S. District Court for the Northern District of California. The suit invokes Section 1 of the Sherman Act and accuses the three companies, who together control somewhere between 90 and 95 percent of the global DRAM market, of coordinating a shift toward HBM as cover to deliberately curtail production of DDR3 and DDR4, driving consumer prices up by roughly 700 percent over four years.
That’s a serious accusation, and it isn’t coming out of nowhere. Samsung and SK Hynix have both pleaded guilty to DRAM price-fixing before. Back in the mid-2000s, following a Justice Department investigation, both companies paid fines, with SK Hynix alone handing over 185 million dollars in April 2005, and some executives served prison time. So when a law firm like Bathaee Dunne argues the industry is repeating old habits under a new AI-shaped excuse, regulators and courts aren’t likely to wave it off as conspiracy talk.
The counterargument from the memory makers, though they haven’t filed a detailed response yet, is the one they’ve been making publicly for months: this is a business responding rationally to unprecedented AI demand, not a cartel. They point to genuinely new fab investment as proof, and Jefferies, the investment bank, has said it doesn’t expect the lawsuit to move prices at all before the end of the year. Whether the case turns up internal communications showing actual coordination, or just circumstantial market data that happens to look damning, is going to matter a lot. If it’s the former, this becomes one of the bigger antitrust cases in tech in years. If it’s the latter, the “AI made us do it” defense probably holds.
It’s worth sitting with both sides here. Building a DRAM fab is a five-year bet on where AI demand will be, and getting it wrong has bankrupted companies before, so caution isn’t automatically evidence of a scheme. But a market where three companies control effectively the entire global supply of a good everyone needs, and prices move nearly 90 percent in a single quarter, is exactly the kind of environment antitrust law exists to scrutinize. Both things can be true at once.
Who’s Actually Making Money Here
Whatever the courts eventually decide, the current setup has been extraordinarily good for the memory makers themselves. Samsung’s chip division reportedly posted single-year profits that beat its combined total from the past 40 years. Samsung’s revenue per bit for traditional DRAM is projected to rise 116 percent year-on-year in 2026, hitting 79 cents, up from 36 cents the year before. SK Hynix is expected to see similar gains, and its stock surge has the company reportedly weighing a U.S. listing. Micron has said it secured demand for essentially its entire 2026 RAM production capacity months in advance, and it’s since exited the standalone consumer DRAM business altogether to focus on higher-margin products.
That last point matters more than it might look. When one of only three major suppliers walks away from the consumer segment entirely, the remaining two have even less reason to prioritize your laptop’s memory over a hyperscaler’s HBM order. This is basically what economists mean by a structural, not cyclical, shift. Nobody expects Samsung or SK Hynix to suddenly decide phones matter more than AI accelerators when AI accelerators pay three or four times as much per wafer.
Below the big three sits a graveyard of former DRAM makers worth remembering, honestly: Intel, NEC, Hitachi, Elpida, Toshiba, Texas Instruments, Qimonda, Infineon, and more, all of whom exited or got consolidated out of the memory business over past cycles. DRAM has always been a brutal, capital-intensive industry that eats weaker players alive during downturns. That history is exactly why the surviving three are being so cautious about overbuilding now, even with prices this high. They’ve all been burned before.
Is Anyone New Actually Coming to Compete
The obvious question is whether new capacity, from anyone, can break the current bottleneck. The honest answer right now is not soon. Micron has started DRAM manufacturing at its Manassas, Virginia plant and is building new fabs in Boise, Idaho, but initial output there isn’t expected until mid-2027, with meaningful volume more like 2028. SK Hynix’s chairman has talked about doubling wafer capacity, but only gradually over roughly five years, and he’s publicly floated the idea that the shortage could persist as far out as 2030. Micron has also struck a partnership with Taiwan’s Powerchip Semiconductor to squeeze more capacity out of existing infrastructure while its new fabs get built, which tells you how tight things are in the meantime.
There is one wildcard worth naming: CXMT, a Chinese memory maker that some ex-Samsung executives have suggested could eventually help ease the shortage. China has obvious strategic reasons to build out domestic DRAM capacity, and CXMT has been expanding fast. But it still has a real technology and process gap compared to the big three, and geopolitics around chip exports and equipment access aren’t exactly making that gap easier to close quickly. Betting on CXMT to rescue global consumer memory supply in the next two years feels premature.
So for now, the market stays split into two tiers. On the enterprise and server side, TrendForce expects demand to stay strong through 2027, partly cushioned by long-term supply agreements that cap some of the wilder price swings. On the consumer side, the picture is messier. TrendForce’s most recent quarterly survey actually shows the rate of increase cooling a bit, projecting DRAM contract prices to rise “only” 13 to 18 percent quarter over quarter in Q3 2026, down from the roughly 60 percent jumps seen earlier in the year. That’s not because supply improved. It’s because regular consumers and PC makers have simply hit the ceiling of what they’re willing or able to pay, and manufacturers can’t keep raising prices past that point without killing demand outright.
The Ripple Effect Nobody Budgeted For
Walk through what this actually looks like outside of earnings calls and lawsuits. A 512GB NVMe SSD that cost around 82 dollars in early 2026 was going for over 126 dollars by the middle of the year, a 54 percent jump in a single quarter. UFS storage used in phones more than doubled in price over the same stretch. PC builders who were used to budgeting maybe 60 to 80 dollars for a 32GB DDR5 kit are now seeing quotes north of 250 dollars for the same kit, and that’s before shipping. Framework, the laptop maker known for repairable, modular hardware, delayed a product launch this year, though it said the delay wasn’t directly tied to the DRAM crunch. Whether or not that specific delay was memory-related, the fact that a company felt the need to publicly clarify it tells you how jumpy the entire hardware industry has gotten.
Game consoles are catching the same wave. Console margins have always been thin, sometimes deliberately sold near cost with the expectation that game and subscription sales make up the difference later. When a core input like memory triples in price mid-cycle, that math stops working cleanly, and console makers are left choosing between eating losses, raising prices on hardware that was never supposed to be a profit center, or quietly trimming specs on future revisions. None of those options are great for anyone who buys a console at launch price expecting years of stable value.
Enterprise buyers aren’t spared either, even with deeper pockets. Networking gear, servers, industrial equipment, medical devices, and automotive electronics all lean on the same DDR4 and DDR5 pools that phones and laptops draw from, and none of those industries can just switch to a cheaper alternative the way a consumer might swap a wishlist item. Sourcing guides aimed at OEM and EMS buyers are now openly recommending 12 to 24 month demand forecasting, last-time-buys on legacy DDR4 parts before they disappear entirely, and long-term supply contracts instead of spot purchases. That’s the kind of advice you give when a market has stopped behaving like a normal commodity market and started behaving like an allocation queue.
What This Means Going Forward
Lenovo has already told the market flatly that high memory prices are the “new normal,” not a blip to wait out. AMD said during Computex 2026 that it doesn’t expect DDR5 pricing to normalize until 2028. Jefferies expects another 40 to 50 percent DRAM price increase in Q3 and 30 to 40 percent more in Q4, with no real relief before 2028 either. If those forecasts hold, the phone or laptop you buy over the next two years is going to cost more, or come with less memory for the same price, than it would have in 2025.
There’s a genuine tension sitting underneath all of this that’s easy to miss if you only look at the price tags. The AI boom that’s making memory expensive is also the thing generating the capital that’s funding new fabs, which are the only real long-term fix for the shortage. McKinsey projects roughly 7 trillion dollars in data center spending through 2030, with more than 5 trillion of that AI-focused. That’s the demand side that broke the old equilibrium, and it’s also, eventually, the thing paying for new supply. It’s just that the fabs take years to build and AI demand doesn’t wait around.
Nobody buying a laptop or a phone right now particularly cares whether this is a rational market response to genuine scarcity or something closer to coordinated gouging dressed up in AI language. Either way, the bill lands on the same people. What the lawsuit against Samsung, SK Hynix, and Micron will actually settle is a narrower question: whether the shortage that’s reshaping device pricing worldwide was simply the price of the AI era, or whether three companies controlling 90 percent of a market decided to make that price higher than it needed to be. Given how the 2005 case played out, that’s not a question regulators are likely to ignore for long.