The AI Build-Out Is Quietly Reshaping Your IT Budget — Here's What SMEs Need to Know

Apr 2
9 min read

The Server You Ordered Three Months Ago Still Hasn't Arrived. Here's Why.

You've been planning that server refresh for months. The budget is approved, the vendor quote is in hand — and then the lead time comes back: 20 to 40 weeks. The price has increased 15 to 20 per cent since the original quotation. And the standalone memory modules you needed to upgrade your existing kit? The distributor won't even quote them separately anymore.

Welcome to the new IT procurement reality, and it has almost nothing to do with your plans or your vendor's efficiency. It has everything to do with a global AI arms race playing out thousands of kilometres from your office — one that is quietly reshaping the cost and availability of every piece of hardware your business depends on.

The AI Infrastructure Boom Nobody Warned SMEs About

The scale of what is happening in the global technology industry right now is extraordinary. In 2024 and 2025 alone, the world's biggest technology companies poured more than $300 billion into AI infrastructure. Microsoft, Google, Amazon, and Meta are individually spending tens of billions of dollars each year on data centres and the compute hardware that fills them. Gartner estimates that total global data centre spending reached $475 billion in 2025, up 42 per cent on the prior year. McKinsey projects that $5.2 trillion in data centre investment will be required by 2030.

The central problem for every SME in South Africa is that the global semiconductor and hardware supply chain is not a separate pipe for hyperscalers and a separate pipe for businesses like yours. It is one shared infrastructure. When Microsoft and Google lock in multi-year purchase agreements for memory chips, server CPUs, and enterprise SSDs, they are competing for the same manufactured capacity that your IT refresh budget depends on.

Memory: Where the Shortage Cuts Deepest

Memory is where the crisis is most acute. The reason is technical but important: the type of memory that powers AI accelerators — High Bandwidth Memory, or HBM — requires approximately three times the wafer manufacturing capacity needed to produce the same amount of standard DDR5. For every wafer of silicon dedicated to the AI boom, three wafers' worth of capacity are effectively removed from the general enterprise and SME markets. As AI chip producers like NVIDIA and AMD launch new accelerator generations annually, their demand for HBM has become the dominant force in global memory manufacturing.

The three largest manufacturers — Samsung, SK Hynix, and Micron, who together control roughly 95 per cent of the DRAM market — are rationally redirecting their most profitable production capacity toward HBM for AI clients. The consequence for conventional computing is severe. Server memory prices are up more than 300 per cent year-on-year. DDR5 prices have increased 70 per cent year-on-year, with specific modules spiking as high as 170 per cent. Samsung has raised the price of a standard 32GB DDR5 module from $149 to $239 in a matter of months. SK Hynix has predicted the memory shortage will persist through late 2027.

In practical terms, server DRAM lead times that historically averaged 8 to 12 weeks are now stretching to 20 to 40 weeks or more. Memory pricing has become so volatile that suppliers in South Africa are now holding quotes for as little as 24 hours — meaning the price at which you approved a project may not be the price when you place the order.

Storage: Enterprise SSDs Hit Hardest

The solid-state drive market has experienced one of the most severe price dislocations in the industry's history. Between Q2 2025 and Q1 2026, pricing for 30TB enterprise-grade SSDs increased by 257 per cent. A drive that cost approximately $3,062 in April 2025 now costs close to $11,000. Enterprise SSDs now cost 16 times more than equivalent-capacity hard disk drives — a ratio that has overturned the economics of business storage planning.

NAND flash, the raw material of every SSD, is being pulled almost entirely toward AI server infrastructure. Phison's CEO told investors that NAND prices had risen 50 to 75 per cent, and that "the AI needs storage for inference" — meaning AI demand extends not just to training workloads but to ongoing operational deployment. Major manufacturers like Kioxia have reported that their entire 2026 production capacity is already sold out, with nearly half dedicated to AI applications.

For SMEs, this does not just mean paying more for new storage. It means that planned ERP upgrades, backup systems, virtual machine expansions, and server builds all carry a significantly higher storage cost than any budget prepared 12 months ago would have anticipated.

CPUs and Servers: Longer Waits, Fewer Options

Server prices from Dell, Lenovo, and HP increased by 15 to 20 per cent entering Q1 2026. Both Intel and AMD have raised processor prices by 10 to 15 per cent across client and server product lines. Orders that previously shipped in one to two weeks now take eight to twelve weeks, with extreme cases reaching six months. Both companies' entire server CPU capacity for 2026 is effectively sold out.

The knock-on effect for standard business deployments is lead time pressure and component bundling restrictions. Many vendors are now refusing to quote standalone RAM, SSD, or CPU upgrades unless they are attached to a full new server configuration purchase. For IT managers trying to extend the life of existing infrastructure, this is a serious operational constraint.

The endpoint market is affected too. ASUS has warned of laptop price rises of up to 30 per cent in 2026, and the emerging category of "AI PCs" with built-in neural processing units is draining constrained LPDDR5X memory supply further. A standard fleet refresh that would have cost a certain amount in 2024 could now exceed that by a third or more, with lead times stretching to eight weeks.

The South African Double Exposure

For SMEs in South Africa, the global hardware shortage does not arrive as a single problem. It arrives as two simultaneous pressures that compound each other.

The first pressure is global: component prices are rising in US dollar terms due to the AI-driven supply squeeze described above. The second pressure is local: virtually all technology hardware sold in South Africa is imported and priced in US dollars. When the rand weakens against the dollar — as it has under current macroeconomic and geopolitical conditions — the rand cost of dollar-denominated hardware increases automatically, entirely independently of any global supply shift.

South African technology retailers including Evetech, Dreamware Technology, and Tech.co.za have told TechCentral that the price pressure is already severe and likely to worsen. Their analysis is that 70 to 80 per cent of current pricing pressure is driven by global component scarcity, with the remaining 20 to 30 per cent attributable to exchange rate volatility. Major brands including Lenovo, Dell, HP, Acer, and ASUS have issued formal warnings of steep price increases.

On top of this sits what local commentators have termed the "Africa Tax" — a structural premium that South African businesses already pay compared to markets in Europe, the US, or Asia, arising from import duties, logistics costs, and limited channel competition. Rising oil prices are compounding the problem further by pushing up shipping costs for a country that sits at the end of long global supply chains.

ITWeb South Africa has specifically reported on how these global memory shortages are already pushing out server lead times in the local market, with resellers now placing orders far earlier in the project cycle because memory availability — not the server chassis itself — has become the determining factor in delivery timelines.

What This Means for Your Business

Projects are stalling. Organisations planning infrastructure refreshes, ERP deployments, cybersecurity expansions, or private cloud build-outs are finding it increasingly difficult to procure the necessary hardware on schedule. For SMEs on tight timelines — a new branch opening, a compliance deadline, a system migration — hardware unavailability is no longer a theoretical risk.

Budget assumptions are outdated. Any IT capital expenditure plan built on pricing from 12 months ago is likely to be materially wrong. Server configurations that were within budget when initially scoped may now exceed approval thresholds. IDC research confirms that global memory shortages will intensify through 2026 as production is prioritised for AI and hyperscale infrastructure. The shortage is structural, not temporary.

The procurement window has narrowed. The period between budget approval and hardware availability has compressed dramatically. Decisions that could previously wait for a formal approval cycle now need to move faster than most SME procurement processes are designed to handle. Industry guidance for 2026 recommends placing server orders six to twelve months ahead of requirement, not the six to eight weeks that used to suffice.

How Long Will This Last?

SK Hynix has predicted the memory shortage will persist through late 2027. SSD supply constraints are projected to continue well into that same year. AI demand for inference compute — the operational phase of AI deployment — is growing even faster than training demand, meaning hyperscaler appetite for storage and memory will not diminish as AI models move from development to production.

New fabrication facilities are under construction, some supported by government incentives, but semiconductor fabs take at least two years to build and commission. Memory manufacturers are cautious about expanding too quickly, mindful of previous cycles where oversupply cratered margins. The structural reallocation of semiconductor manufacturing capacity toward AI-optimised components is a multi-year transition, not a quarter's disruption.

Planning on the assumption that hardware availability and pricing will return to 2023 norms within 12 months is not a prudent basis for budgeting.

What SMEs Should Do Now

The new procurement environment requires a deliberate strategy, not reactive purchasing.

Move your procurement horizon out significantly. If your business has a hardware refresh, new branch opening, or infrastructure project planned for the next 12 to 18 months, start the procurement process now. The gap between order placement and delivery has widened enormously, and acting early is the only reliable way to protect your project timelines.

Lock in pricing when you receive a quote. With memory pricing held for as little as 24 hours in the current market, a quote that is not acted upon is likely to come back higher. Treat approved quotes as time-critical commitments, not standing offers. Align your IT, finance, and leadership teams to ensure quotes can be approved and funded within 48 hours.

Audit your existing assets before buying new. In a market where refreshing hardware costs significantly more than it did 18 months ago, extending the usable life of certified, well-maintained equipment can be a legitimate financial strategy. Assess which existing servers, workstations, and networking equipment can be maintained for an additional 12 to 24 months while the market stabilises.

Consider certified refurbished hardware. Unlike new equipment subject to fragile global production cycles and extended lead times, refurbished assets already exist within the local market. For the majority of business workloads — productivity platforms, collaboration tools, virtual desktops — the performance gap between new and refurbished hardware is operational rather than theoretical, and certified refurbished can ship from in-stock inventory within days.

Evaluate cloud-first alternatives for new workloads. Where new hardware procurement is being driven by capacity expansion rather than replacement, cloud infrastructure from providers like Microsoft Azure can deliver the required capacity on a more predictable cost model. This shifts capital expenditure to operational expenditure, insulating your budget from component price volatility. It does not suit every workload — particularly where POPIA data sovereignty requirements mandate that data remains within South African borders — but for many SMEs it reduces exposure.

Work with a multi-vendor procurement partner. Single-vendor dependence is a significant risk in the current environment. A hardware partner with access to multiple OEM channels, as well as certified refurbished inventory, is better positioned to find available stock and work around lead time constraints.

The Bigger Picture

The AI infrastructure build-out is one of the largest capital deployments in technology history. Its knock-on effects on the broader hardware market are real, measurable, and likely to persist until late 2027 at the earliest. For South African SMEs — absorbing global hardware inflation, local currency depreciation, and a structural market premium simultaneously — this warrants strategic attention and early action.

The businesses that come through this in the best shape will be those that treat hardware procurement as a strategic function, not an administrative one. That means understanding your hardware estate now, knowing your options, and making timing decisions with the market in mind — not waiting for a laptop to fail or a server refresh to fall due at peak pricing.

If you want a clear picture of where your environment stands and what your options are over the next 12 to 18 months, get in touch for a complimentary IT procurement assessment.

Big enough to deliver enterprise-grade solutions. Small enough to care about getting it right for your business.

📞 +27 11 663 0000 ✉️ helpdesk@firstconsulting.co.za