If you've priced up a memory upgrade recently and wondered why RAM is so expensive right now, you're not imagining things. DDR5 modules, LPDDR5 laptop memory, and server-grade ECC RAM have all seen meaningful price increases over the past 12 months, driven by a combination of supply constraints, shifting demand from AI infrastructure buildouts, and deliberate production cuts by the three companies that manufacture the overwhelming majority of DRAM on the planet: Samsung, SK Hynix, and Micron.
The DRAM oligopoly and why it matters for prices
Memory manufacturing is one of the most concentrated industries in technology. Samsung, SK Hynix, and Micron together control roughly 95% of global DRAM supply. That means pricing power sits almost entirely on the supply side. When any of these three companies adjusts production volumes, the effect moves through the entire market within a quarter. In 2023, all three cut wafer starts significantly to burn off oversupply. Those cuts created a supply trough that's now making itself felt in 2026, as demand has recovered faster than capacity can be added.
New fab capacity takes years to build and qualify. Samsung's Taylor, Texas facility and SK Hynix's new Indiana plant are both ramping, but qualification cycles for leading-edge DRAM nodes are long. Supply won't catch up with demand quickly, and the manufacturers have little incentive to rush: tighter supply supports better margins.
AI is consuming memory at a scale the industry didn't fully anticipate
High-bandwidth memory (HBM), the stacked DRAM that sits on AI accelerators like NVIDIA's H100 and B200 series, is consuming a disproportionate share of Samsung's and SK Hynix's leading-edge output. HBM production is significantly more complex than standard DRAM, requiring wafer-on-wafer stacking and through-silicon vias. The same fab capacity that might produce a large volume of DDR5 modules produces far fewer HBM stacks. As hyperscalers have accelerated GPU cluster buildouts through 2025 and 2026, HBM allocation has crowded out conventional memory production.
The knock-on effect is real. Standard DDR5 and LPDDR5 output is lower than it would otherwise be, because the manufacturers are prioritising the higher-margin HBM contracts. For anyone buying desktop RAM, laptop upgrades, or server memory kits, that trade-off shows up directly in the price they pay.
What's happening specifically to laptop and enterprise memory
LPDDR5 and LPDDR5X, the low-power variants soldered into most modern business laptops, have been particularly tight. These modules share process nodes with HBM and high-performance mobile chips, so the same capacity crunch applies. If you're procuring a fleet of business laptops right now, you'll notice that upgradeability has also become a secondary issue: most devices ship with memory soldered directly to the motherboard, meaning the price at purchase is locked in. There's no retrofitting your way out of an expensive configuration.
For teams evaluating business laptop memory types like LPDDR vs DDR, the distinction matters for more than just performance. LPDDR is almost universally soldered, so the per-unit procurement cost is the only lever available. DDR SO-DIMMs, found in a shrinking number of business laptops, still allow post-purchase upgrades, which can partially offset sticker price by allowing a base configuration purchase now and an upgrade later when prices ease.
Server and workstation ECC DDR5 is similarly elevated. Data centre build programmes have continued at pace globally, and registered DDR5 (RDIMM) commands a substantial premium over consumer modules. Australian IT teams running on-premises infrastructure have felt this acutely, particularly those refreshing server hardware in 2026.
How much RAM do you actually need before you spend?
Before committing to a higher-spec configuration at today's prices, it's worth auditing what you actually need. The question of whether 16GB of RAM is still enough depends heavily on the workload. For general knowledge-worker tasks, 16GB remains functional, though tighter than it was two years ago given browser memory consumption and background telemetry from enterprise security agents. For developers running containers, engineers using large datasets locally, or teams running AI inference on-device, 32GB is now the practical baseline.
The discipline here is to match spec to role rather than uniformly upgrading a whole fleet. Buying 32GB configurations for staff whose workload peaks at 12GB of active usage is a real cost at today's prices.
What to do if you can't avoid buying now
A few practical options exist for buyers who can't wait out the cycle.
- Buy at base spec where LPDDR is soldered, and defer refresh for roles where 16GB still covers actual usage.
- For DDR5 desktop or workstation builds, consider 32GB kits at the entry-level frequency rather than high-speed kits; real-world performance differences between DDR5-4800 and DDR5-6000 are small for most workloads.
- Check refurbished enterprise memory from reputable resellers for server refreshes. Grade A pulls from decommissioned servers are tested to spec and can carry meaningful discounts over new.
- Time large procurement runs to avoid Q4 demand peaks, when consumer electronics seasonality adds a second layer of pressure on consumer-grade modules.
Spot pricing for DDR5 consumer kits has fluctuated between AU$150 and AU$280 for a 32GB (2x16GB) kit across the first three quarters of 2026, depending on brand and speed grade. That's a significant range. Watching spot pricing across a 6-to-8-week window and committing when it dips can save a meaningful amount on larger fleet orders.
When will RAM prices come down?
The honest answer is: not quickly. New fab capacity from Samsung and SK Hynix's overseas expansions will add supply, but it won't be qualified and productive at scale until 2027 at the earliest. HBM demand from AI accelerator deployments shows no sign of slowing. The manufacturers have also demonstrated in past cycles that they're willing to hold production discipline when margins are strong, rather than flooding the market and collapsing prices.
A correction is likely, as it always eventually arrives in DRAM cycles. But buyers planning infrastructure refreshes in the next 6 months shouldn't bank on a significant price drop within that window. The more practical approach is to right-size procurement now, build in a hardware refresh review for late 2027, and use the interim period to audit actual memory utilisation across the fleet so the next procurement cycle is based on measured need rather than assumed headroom.
For teams also weighing storage decisions alongside memory upgrades, the process of formatting and installing an SSD is one of the more cost-effective ways to improve system responsiveness without touching RAM at all. A fast NVMe SSD reduces the performance penalty of memory pressure by improving swap performance, which won't fix a genuinely undersized configuration but does extend the useful life of a system running at the edge of its memory ceiling.
The RAM market in 2026 is tight, expensive, and structurally constrained. Knowing why prices are elevated is useful, but the more valuable question is whether you can optimise what you buy and when you buy it.