Here is a market that has functioned as a fairly predictable commodity trade for two decades, and it has collapsed almost entirely in the space of six months. The fundamental question is: who bears responsibility, and who pays the price?
The answer to the second part is simple. You do. If you are a consumer shopping for a desktop upgrade, a small business buying server hardware, or a mid-tier cloud provider trying to hold your pricing steady, the memory market has turned decisively against you. Samsung Electronics and SK Hynix have now notified customers of plans to raise DRAM prices again in the second quarter of 2026, with price negotiations already under way. This follows a run of increases that, by some accounts, has pushed DRAM prices to more than three times their October 2025 levels, driven by the hyperscaler buildout and a squeeze on available memory supply that began last year.
The underlying cause is not complicated. Manufacturers have reallocated production capacity toward high-bandwidth memory for AI data centres, which commands higher margins, prompting Samsung, SK Hynix, and Micron to shift away from traditional enterprise and consumer memory products. High-bandwidth memory production for AI accelerators consumes approximately three times the wafer capacity of standard DRAM per gigabyte, according to a Micron executive, forcing memory makers to pull back from consumer and enterprise products. The arithmetic is brutal and straightforward: the giants of artificial intelligence need more memory than ever, they can pay almost anything to get it, and the manufacturers are entirely rational to serve them first.
Given the memory market's shift to a seller-dominated structure, DRAM quotations are now revised quarterly for large-scale customers, while for smaller buyers prices can change within days, making procurement nearly impossible. Small-to-medium businesses dependent on memory supply face a particular problem because, given the volumes they acquire are far lower than those of companies like Nvidia and Apple, suppliers offer no flexibility on pricing or volume. For those buyers, acquiring DRAM without raising end-product prices has become functionally impossible.
What makes this crisis distinctly modern is the layer of automated opportunism sitting on top of already stressed supply. DataDome's Galileo threat research team uncovered a sophisticated operation where fraudsters were monitoring prices and targeting DDR5 inventory, with scalping bots making over 50,000 requests every hour and DataDome blocking more than 10 million scraping attempts in total. The bots have been hitting select sites every 6.5 seconds to query inventories, with the people behind them reportedly using AI tools to enhance their scraping effectiveness.
Jérôme Segura, VP of threat research at DataDome, described the evolving threat to The Register: "Fraudsters will combine various tooling and commodities to perform fraud at scale," he said, adding that threat actors have been observed using AI to reverse-engineer anti-bot protection and automate scripting tasks, noting that "AI is unique in that it gives leverage from script kiddies all the way to professional scrapers." The scrapers attempt to avoid detection by adding cache-busting parameters to every request and calibrating their speed to stay just below volumetric alarm thresholds.
As DataDome's own report puts it, by rapidly snapping up limited DDR5 inventory for profitable resale, these bots further deplete consumer supply, "effectively boxing out legitimate customers and driving market prices even higher." The scalping phenomenon is a direct consequence of scarcity; remove the scarcity and the profitability of the arbitrage disappears. The bot problem is a symptom, not the disease.
The counter-argument deserves serious consideration: proponents of unrestrained AI capital expenditure would argue that the infrastructure buildout will, over time, generate the productivity gains and cost reductions that justify present dislocation. There is something to this. Gartner senior principal analyst Kanishka Chauhan has noted that "demand for memory is strong, driven by ongoing AI investments", and the scale of that investment is genuinely unprecedented. Shares in Micron surged 240 per cent last year, while Samsung more than doubled and SK Hynix's market cap nearly quadrupled, suggesting investors believe the AI memory supercycle has real legs. If the productivity dividend from AI eventually materialises at scale, today's hardware pain may look, in retrospect, like the necessary friction of a structural transition.
But transitions have victims, and this one is already producing them in measurable ways. With so much supply spoken for by large cloud providers, mid-tier and smaller cloud vendors have been forced to raise their prices, and the starved supply chain is expected to limit entry-level PC and phone shipments. Less visibly, the shortage is also dismantling corners of digital culture. Rock Paper Shotgun reports that Myrient, a ROM distribution and game preservation site that has operated since 2022, will shut down on 31 March 2026. Its operator cited the AI-datacenter-driven hike in RAM, SSD, and hard drive prices as a contributing factor to hosting costs that had become unsustainable, compounding an existing monthly deficit of more than US$6,000 out of pocket. The site's closure is a small data point, but it illustrates how supply chain dislocations in semiconductor markets can reach into unexpected corners of the digital world.
Analyst IDC noted the "unprecedented" memory chip shortage before Christmas last year and warned this would have knock-on effects for both hardware makers and end users that may persist well into 2027. Some forecasters are less charitable about the timeline; Phison's CEO has reportedly predicted that the situation could persist through 2030, or even for a full decade.
Strip away the talking points and what remains is a market failure that government and industry have so far been content to observe rather than address. Competition regulators, including the Australian Competition and Consumer Commission, have the tools to examine whether concentrated market power among a handful of memory manufacturers is distorting outcomes for downstream consumers and businesses. Whether they exercise those tools is a separate question. In the meantime, the Australian Bureau of Statistics will eventually capture the price effects in CPI data, as higher hardware costs flow through to consumer electronics, cloud services, and business IT procurement across the country.
The honest assessment is that this is not a left-right issue; it is a competence issue, and a structural one. The concentration of memory manufacturing among three dominant global players, the insatiable capital appetite of the AI industry, and the absence of any meaningful mechanism to protect consumer access to commodity hardware have all converged at once. Reasonable people can disagree about how aggressively governments should intervene in semiconductor supply chains. What is harder to dispute is that ordinary buyers, small businesses, and niche digital communities are paying a real price for a transition they had no part in designing. Acknowledging that cost honestly is the least any serious policy conversation should manage.