The Memory Squeeze
If you’ve ever had a server refresh “ready to go” except for one missing piece of RAM, you’ve seen the new reality of IT procurement: memory is no longer a boring commodity. It’s become a pacing item. The kind that turns a neat project plan into a string of reschedules, quote extensions, and awkward conversations with finance.
And the uncomfortable truth is this: the memory supply situation isn’t a short-term glitch. The forces driving it are structural, which means the way we plan infrastructure needs to change for the next few years.
Why does this feel different from previous shortages
In past cycles, memory price spikes were usually a case of demand outpacing supply for a quarter or two, then the market corrected. What’s happening now is more like a permanent lane change.
A big chunk of the world’s “best” memory output is being pulled into AI infrastructure, especially high-bandwidth memory (HBM) used in modern AI accelerators. That matters because the same suppliers and factories that used to feed mainstream DRAM (your server and laptop RAM) are increasingly prioritising what hyperscalers want first. When high-margin AI demand gets first call on capacity, everyone else is left negotiating what’s available and when.
That’s why you’re seeing pricing and availability behave less like a quick spike and more like a sustained squeeze. TrendForce, for example, recently forecast very sharp quarter-on-quarter contract price increases into Q1 2026 for both conventional DRAM and NAND, with server DRAM singled out as especially tight.
It’s not only about RAM. SSD supply is part of the story, too.
On the storage side, NAND is facing its own tug-of-war between demand and disciplined supply. Some manufacturers are actively managing output, and there are also reports that 2026 NAND capacity is effectively presold in certain segments. The result is familiar: price pressure, constrained availability, and fewer “quick wins” for last-minute upgrades.
Put all of that together, and you get the situation many IT teams are living right now:
- quotes that expire before approvals land,
- lead times that shift mid-project,
- and builds that need to be redesigned around what you can actually source.
The part most IT teams miss: memory is now a risk item, not a line item
Most organisations plan infrastructure around the big-ticket items: servers, storage arrays, firewalls, and licences. Memory tends to get treated as “we’ll add that at the end”. That habit made sense when RAM and SSDs were plentiful, and pricing was predictable.
Now, it’s the opposite. Memory is the constraint that can delay the entire project—so it needs to be managed like a dependency, the same way you’d manage a critical contractor or a key compliance gate.
A useful mental model is this: if you can’t guarantee memory, you can’t guarantee delivery.
So what do you do? You change the planning approach from “perfect design first” to “deliverable design under uncertainty”.
A more practical way to plan for 2026–2027
1) Start with a “minimum viable build”, not an ideal build
Instead of designing a single perfect spec and hoping procurement can match it, define two builds:
- Minimum viable configuration: the smallest RAM/SSD spec that still meets performance and supportability
- Target configuration: the ideal spec if supply and pricing allow
This sounds simple, but it removes the biggest project killer: waiting indefinitely for the perfect module.
It also keeps your internal stakeholders calm. You’re not “downgrading”; you’re executing a planned option.
2) Standardise your memory choices like you standardise your security stack
In a constrained market, variety is expensive. The more unique RAM and SSD SKUs you approve, the more you’re at the mercy of whatever is out of stock this week.
Standardisation gives you leverage:
- Procurement has fewer variants to chase,
- You can swap between suppliers more easily,
- And you can hold spares that actually fit more than one system.
IDC has pointed out that this isn’t just a demand spike but a strategic shift in how capacity is allocated, and that forces device and IT buyers into a more “fluid” supply environment.
In plain English: the teams that simplify their options will cope better.
3) Buy constraints first, then everything else
When memory is tight, the order you procure in matters.
In many environments, you can secure chassis/CPU platforms fairly reliably, but you’ll get caught on the memory configuration (and then everything sits in boxes while you wait).
A better flow is:
- Confirm validated compatible RAM/SSD options
- Secure the memory allocation (or physical stock)
- Then commit to the rest of the order
That one shift reduces the number of “we have everything except…” scenarios significantly.
4) Extend lifecycle intentionally (and don’t let it become “hope”)
Most organisations can stretch hardware longer than they think if they do it deliberately.
The mistake is stretching the lifecycle without changing operations. That’s when you get surprise failures, performance complaints, and emergency purchases (the worst way to buy memory in a tight market).
If you’re going to run equipment longer, pair it with:
- tighter monitoring (especially for memory errors on critical hosts),
- planned replacement of the riskiest nodes first (not just “the oldest”),
- and a small pool of known-good spares for business-critical platforms.
This turns lifecycle extension from a gamble into a managed decision.
5) Use cloud as a bridge, not a religion
Not every workload belongs in the cloud, and not every organisation wants to shift operating costs. But the cloud is extremely useful as a temporary gap filler when on-prem projects slip due to parts.
A short-term cloud bridge can keep deadlines intact for:
- dev/test environments,
- burst compute needs,
- analytics jobs,
- or temporary capacity while you wait for memory modules to land.
You don’t need a full migration to benefit. You need a pragmatic “plan B” that buys you time.
6) Fix the finance problem: quote validity and approval speed
This shortage exposes a common internal bottleneck: even when the technical plan is solid, procurement and approvals move more slowly than the market.
With memory pricing moving quickly, that delay becomes real money. Reuters recently quoted a major PC manufacturer warning that surging memory prices could affect the industry through 2027, with memory potentially taking a much larger share of PC bill-of-materials costs.
So it helps to align finance early around a few realities:
- Pricing validity windows will shorten,
- Some quotes will need contingencies,
- And approvals need to be faster for time-sensitive BoMs.
The goal isn’t to create panic—it’s to reduce rework.
The “foreseeable future” view: what to expect next
Nobody can give a perfect timeline, but the direction of travel is fairly clear:
- AI infrastructure keeps pulling memory capacity and investment
- Pricing remains volatile
- and allocation behaviour becomes normal, not exceptional
That means the winners won’t be the organisations that try to time the market. They’ll be the ones who plan for uncertainty:
- fewer unique configurations,
- earlier procurement for constraints,
- realistic fallback builds,
- and a managed approach to lifecycle extension.
In South Africa specifically, the impact can feel sharper because imported stock, distributor allocation, and FX movement can amplify the pain. Which is exactly why planning matters: when supply is tight, good preparation is a competitive advantage.
A closing thought
Memory used to be something you “added” to a project. Now it’s something you plan around. The teams that adapt their refresh and procurement playbooks won’t just survive the squeeze; they’ll deliver more consistently while everyone else is still waiting on a backorder.
