AI demand has collided with limited memory manufacturing capacity, pushing server RAM and storage prices up fast and squeezing lead times. That doesn’t mean progress has to stall. There is no simple magic fix, but we can help you navigate the situation to deliver outcomes for your organization.
Why memory is tight (and why it affects almost everything)
Two years ago most enterprises weren’t running large‑scale generative AI. Today, AI workloads consume a significant share of the global memory supply, compounding already heavy demand from hyperscalers like Azure, AWS and Google.
To feed AI accelerators, vendors are shifting capacity toward HBM (High Bandwidth Memory). HBM consumes roughly 3× more wafer per gigabyte than conventional DRAM, so the same factory output yields fewer gigabytes for the broader market. In recent memory we have worked through supply chain issues caused by COVID or natural disasters, but this time it’s different. Supply of memory suddenly hasn’t suddenly disappeared; it’s being consumed by a highly demanding new segment.
With only three major memory manufacturers (Samsung, SK Hynix, Micron) and multi‑year timelines to bring new fabs online (not to mention capital investments in the tens of billions), supply looks to remain right well into 2027. With prices high, share prices are at record levels, producers are cautious about over‑expanding capacity and creating excess supply.
This isn’t confined to DIMMs. SSDs are impacted too, because NAND and DRAM share elements of the production ecosystem. Since every device includes these components we expect other hardware items such as networking kit, interactive displays and other devices to feel knock‑on effects. Even that smart toaster you've had your eye on (or, if you are like me, wondered why you'd ever want one) is going to be impacted.
What this means for pricing, quotes and lead times
Prices for memory and storage devices have surged. For example, list pricing on 64GB server DIMMs has increased by about 250% increase between November 2025 and February 2026. For customers operating dense virtualization estates and memory-hungry database workloads, we are already seeing overall hardware costs on projects increase by 50% or more compared to 2025.
Due to the pricing uncertainty we are seeing vendors reduce the validity of quotes (currently 14 days in most cases), extending lead times on many products, and in some extreme cases are reserve the right to amend pricing up until the point of shipment.
The net result: constantly re‑spec’ing configurations and playing the “what‑if” game can cost you time and money - by the time a decision is made, prices can move again.
A note on cloud pricing risk: Public cloud isn’t immune to the same supply‑and‑demand pressures driving on‑prem component volatility, and we have already seen Microsoft addressing challenges with capacity in certain key Azure regions. Providers can reprice pay‑as‑you‑go rates, adjust regional list prices based on energy costs and exchange rates, and pass through surcharges for premium storage, egress or inter‑datacentre traffic.
Do you have a burning platform?
Before committing budget, decide whether your current estate can safely ride out the turbulence or if constraints force action now. Signs of a burning platform include:
- End of support - no security patches or vendor maintenance, particularly if the organisation has compliance requirements such as Cyber Essentials or PCI.
- Loss of software support – software that is out of vendor support or no longer support the hardware you operate.
- Rising, unpredictable software costs that your current stack can’t mitigate.
- Unreliable, slow, or failure‑prone hardware that materially affects operations and opportunity cost.
If the current platform won’t carry you to 2027/2028, earlier replacement is often the least‑risk option. If it will - and you aren’t blocked by software or compliance - renewing support and deferring big memory purchases may be the smartest move.
Key lifecycle dates to keep in view:
- VMware vSphere 7 end of support: October 2025
- SQL Server 2016 end of support: July 2026
- Windows Server 2016 end of support: January 2027
How to pivot: four practical tracks
1) Partial refresh: fix the pain points first
Target the systems causing the most risk or friction – hosts at capacity, aging storage arrays, or time-intensive systems that drain your resources – rather than rip‑and‑replace across the estate. We can advise on how to address the most critical issues first and where we can buy time to address other issues later.
2) Software updates and rationalisation
Upgrading OS, hypervisor, and database layers can unlock performance, security and licensing efficiencies on existing hardware. Where appropriate, align with end of support milestones well in advance to avoid crunch‑time upgrades later. We recognize these upgrades can be extremely time consuming and can assist with engineering resources to help accelerate projects.
3) Platform alternatives to rebalance cost and risk
If existing licensing or platform costs are becoming unpredictable, consider if the estate is locked-in or if alternatives could be viable. Consider hybrid approaches such as using Public Cloud as a short-term home for some workloads. We can offer resources within our UK Sovereign Cloud for customers looking to re-platform onto a familiar server-based platform while avoiding hardware investments.
4) Cyber resilience: protect the business now
Invest in backup with immutability, disaster recovery, and security hardening. These initiatives raise resilience without large memory footprints and often deliver immediate risk reduction that benefits organisations in multiple ways. Managed services can accelerate time‑to‑value while internal teams focus on change events.
Buying tactics that work in a volatile market
- Decide fast on a desired specification. Avoid trying to compare multiple options to see which is cheapest; changing specifications to save a small amount results in re-quotes with higher pricing for a lower specification.
- Align internal approvals with 14‑day validity windows; pre‑brief budget holders on the market issues so that they understand the urgency of sign-off, and the cost of delay.
- Be open to mixed strategies; in a scenario where a traditional all‑or‑nothing refresh project might be a non-starter for you, instead look at a diverse set of projects to target pain points, maintain working systems, update software, and address security concerns .
Talk to us about your 2026–2027 plan. We’ll help you assess whether you’re on a burning platform, then help you prioritise partial refreshes, software upgrades, cyber resilience and platform alternatives that deliver business value without being hostage to volatile memory pricing.