How hardware manufacturers can overcome the storage supply squeeze

For years, dynamic random-access memory (DRAM) and NAND flash were treated as reliable, almost invisible components in the global technology supply chain. They were abundant, predictable and, crucially, affordable. But in the space of just a few months, that stability has collapsed. The global memory market has turned on its head and the consequences for organisations that depend on high-performance infrastructure are now feeling the full impact.

 Contract prices for DRAM alone are forecast to rise by as much as 90-95%, while NAND flash, the technology underpinning SSDs and NVMe drives, is up 144%, with some deals even hitting price surges of up to 400%.

 Yet the most alarming development is not the cost. It’s the fact that, in many cases, memory simply cannot be obtained at any price. This is no longer a procurement problem; it’s a structural shift that is reshaping the entire hardware ecosystem, from hyperscalers with billion pound budgets to the small system builders serving enterprise clients.

The market bottleneck in storage manufacturing

Although countless brands appear to sell memory, the underlying silicon comes from only a handful of manufacturers, everyone else is merely an assembler. These organisations are now operating at full capacity, with no slack to absorb the sudden spike driven by AI and cloud workloads. Building new fabrication plants takes around seven years, making rapid relief impossible.

The result is a world where organisations planning next generation compute environments are discovering that the limiting factor isn’t GPUs or CPUs, it’s the memory needed to feed them. What the spike in demand looks like today is a server configuration that previously cost around £46,000 has jumped to over £98,000 and is still rising, purely due to DRAM and NAND price inflation. This is not normal market behaviour. It’s what happens when global demand exceeds physical manufacturing capacity.

How hyperscalers are exacerbating the issue

The big cloud and AI players have responded to the shortage in the only way they can: by buying vast quantities of memory and GPUs as far in advance as possible.

Some are even purchasing components for data centres that haven’t been built and AI projects that don’t yet exist, driven by fear that supply will dry up just when they need it.

From our own sources, we know that one of the leading car manufacturers ordered 7,000–9,000 high‑end GPUs, each valued at around £25,000, without the immediate infrastructure ready to use them. This tells you the priority is securing supply, not deploying it. This behaviour triggers an industry stockpile loop where demand increases simply because others are stockpiling too. For smaller players and enterprise IT teams, the effect is devastating, even if they can afford the new memory prices, the product itself is unavailable.

The view from the frontline

As a leading UK server manufacturer that designs, builds and ships high performance servers all over the world, we’ve had a front row seat to the unfolding crisis from day one.

As a business, to keep hardware flowing to our customers, we’re now buying ten times more memory upfront than we ever have before, driven by aggressive forecasting and the need to secure physical stock before it disappears into hyperscaler pipelines.

One large customer alone is signalling demand for 350–500 servers, on top of a projected run‑rate that could reach around 1,000 units this year, forcing us to increase planned memory purchases by 50% at short notice. Alongside going above and beyond to meet the needs of our loyal customer base, we’re also facing inbound requests from leading distributors desperate for inventory, offering us five times the original price for certain SSD and NVMe stock we already hold. At the time of writing, drives valued internally at around £170,000 are now receiving offers upwards of £870,000.

In other words, stockpiling has become so extreme that simply possessing memory is now seen as a strategic asset. And even then, availability is not guaranteed. Over the pond, a DRAM SKU that cost $700 at the end of 2025 is today being quoted at $2,400, with no assurance that supply will materialise. This is what a supply chain breaking under strain looks like.

The demand isn’t stopping here, the ongoing challenges for smaller builders

While the crisis affects everyone, not all organisations can respond equally.

Smaller system builders, especially those who buy components only when an order is placed, are at serious risk. With thin cash reserves and little forecasting infrastructure, not only can they not commit to long‑term memory orders and pay upfront for 12–18 months of inventory, they also can’t absorb the cash flow impact of stockpiling components at inflated prices.

This isn’t just about paying more, many small system builders simply won’t be able to get it. Larger manufacturers, by contrast, are positioned to weather the storm. They have the forecasting models, supplier relationships and capital needed to secure allocation in advance. Some will even grow as customers seek reliability over price.

The crisis is accelerating consolidation across the hardware ecosystem, not because of competition, but because of scarcity.

What CIOs and CTOs must do now

The organisations that navigate this crisis with success will treat memory and storage not simply as a commodity, but as a strategic enabler for growth. It’s essential to share demand signals early and work closely with server builders and distributors. This level of visibility enables them to bid for limited allocation on your behalf and secure stock months before deployment.

Given the uncertainty and fluctuation facing the sector, building realistic, long‑range forecasts and shifting from quarterly procurement cycles to 12–18-month horizon planning is recommended. If AI or high-performance workloads are expanding, that expansion must be reflected in memory forecasting. CIOs and CTOs must also rethink specification strategies. Over‑provisioning made sense in a world of cheap memory but this no longer the reality. Organisations must design systems around what can be reliably sourced.

A new era defined by memory

 The memory shortage is no longer a temporary fluctuation in the market, it’s here to stay. DRAM and NAND have moved from background elements to critical components that influence everything from server design to investment planning. We are entering a period where the limiting factor in AI, cloud and high‑performance computing is not innovation, it’s allocation. With no short‑term increase in global memory manufacturing capacity, this squeeze will continue to shape pricing, delivery times and architecture decisions well into this year alone. The organisations that recognise the strategic implications early and act decisively, long before the next wave of shortages hits, will be ahead of the curve.