In September 2025, a 32GB DDR5 memory module cost about US$95. By December, that same module was selling for around US$400.
That’s not a typo. It’s not a temporary supply disruption. It’s a fourfold price increase in three months, and it’s unlikely to be the peak.
We’ve been through memory shortages before. Demand spikes, supply lags, prices rise, then the market corrects. Earthquakes in Taiwan. Floods in Thailand. COVID-19. Forecasting misses. Each time, the industry eventually found balance again.
This time feels different.
While this can seem daunting, our experts are working hard to assist customers to navigate this shift. Here’s what’s happening, why it matters, and some advice on next steps so you’re not left behind.
In late 2024, OpenAI signed long-term agreements with the two biggest memory makers Samsung and SK Hynix, covering up to 900,000 DRAM wafers per month. For context, Samsung and SK Hynix together control around 70 per cent of global DRAM and 80 per cent of high‑bandwidth memory (HBM) market1.
Put simply, one organisation secured nearly half of the world’s memory supply.
The ripple effects were immediate. Cloud providers found themselves unable to secure incremental allocations for AI accelerators. Consumer electronics manufacturers experienced the sharpest memory cost increases on record. Some retailers quietly stopped taking desktop PC orders altogether, not because of demand, but because they couldn’t price or guarantee supply with confidence.
Industry reports show memory and storage chips, CPUs and complete systems all experiencing significant upward cost movement. DRAM contract prices have risen by roughly 50 per cent year to date in 2025, and analysts expect further increases through 20262.
Memory suppliers are also signalling that this trend is far from finished. Several have already announced contract price rises of between 30 and 60 per cent for server‑class DRAM3. In parallel, enterprise SSDs have become both the largest and the tightest NAND segment, driven by the accelerating build‑out of AI infrastructure and the associated demand for high‑performance storage. From the manufacturers’ perspective, the decision made sense. High Bandwidth Memory (HBM) carries materially higher margins than commodity DRAM. Multi-year contracts remove volatility and guarantee utilisation. Compared to the consumer market, it’s predictable, profitable, and low risk.
For everyone else, it marked a structural reset.
Research by Macquarie Equity Research4 explains why this isn’t a short-term issue.
A single 1GW-scale AI data centre, configured with roughly 400,000 GPUs, consumes more than 18,000 DRAM wafers per month when you account for both HBM and system memory. At the same time, net new global DRAM supply is only around 250,000 wafers per month.
At that rate, the industry can support roughly 15GW of new AI data centre capacity over the next two years without displacing existing markets.
Industry forecasts, however, point to 40 to 50GW of new AI data centre construction over the next three years, alongside compound annual growth of around 40 per cent in AI chip shipments.
Something has to give. For now, what’s giving is availability and price stability for anyone operating outside the AI supply chain.
The practical impacts are already showing up.
Quote validity is shrinking from thirty days to as little as seven. Anything that contains meaningful amounts of memory or storage, laptops, servers, storage platforms, even some network devices, is exposed. Buyers without forward commitments are being pushed into spot-market pricing, or worse, being told allocation simply isn’t available.
NAND flash (storage) is following the same trajectory. One-terabit TLC NAND rose from about US$4.80 in July to more than US$10.70 by November5, and several major manufacturers have already sold-out portions of their 2026 production.
The old procurement playbook assumes time, optionality, and stable pricing. None of those assumptions currently hold.
There are still levers available, but they require intent and speed:
We’re working closely with our vendor partners and distributors to help customers as the market evolves. That includes buffer-stock strategies, price-lock mechanisms, and specification flexibility to minimise disruption. We’re also structuring hybrid refresh approaches, selectively extending maintenance, and using tailored finance options where forward purchasing makes sense.
As Australia’s largest partner for Cisco, Microsoft, Dell Technologies and HP, we have both the relationships and the balance sheet to engage early and secure outcomes that are unrivalled by others.
This situation is evolving quickly. If you’re planning a refresh or expansion in 2026, now is the time to start the conversation. Talk to your Data#3 account manager or contact our expert team here.
Information provided within this form will be handled in accordance with our privacy statement.