Storage buyers in South Africa are running the numbers this year and not liking the answer. The capacity they need costs far more than the budget they set 12 months ago, because the memory that every server and array depends on has become expensive almost overnight, driven by demand from AI datacentres. Refreshes are being pushed out, hardware is being kept running longer than planned and analysts are reporting that companies are extending the life of what they already own rather than replacing it.
This means businesses are delaying their purchasing decisions, says Daniel Teixeira, engineering manager at Everpure, the vendor formerly known as Pure Storage. “They’re sweating what they have and getting very creative with software features and cloud to ride out the storm.”
And yet, the software that manages that storage keeps growing. IDC forecasts South Africa’s software- defined storage (SDS) market to expand at roughly 17% a year through 2030, while the global rate is at about 18%. As capacity grows more expensive, it’s worth asking why organisations are pouring money into the software layer, and whether taking storage out of the appliance earns back what it costs.
The idea behind SDS is not new, but the memory crisis has given it fresh impetus. It splits the software that manages storage from the hardware it runs on, which breaks the link between adding capacity and committing to one vendor’s box. “SDS was a turning point. It lets you be independent from the hardware, not locked into any specific vendor,” says Warren Olivier, Cloudera’s regional vice president for Africa. SDS lifts the management of capacity, resilience and performance off the underlying hardware, so a business can pool drives from several vendors and run them as one estate.
“With legacy storage, you were bound to a specific device or silo,” says Oliver. “To grow, you needed downtime and a forklift upgrade. With SDS, you just add new commodity nodes to the cluster and expand linearly.” There are three reasons for adoption: cost, AI-driven data growth and the attempt to avoid lock-in. Commodity hardware costs a fraction of a proprietary appliance, existing kit can be repurposed instead of scrapped, and the savings are realised as financial, operational and capacity efficiencies.
There is also the matter of independence. “You don’t want to be locked into a specific hardware technology, a specific platform or a specific hyperscaler,” says Oliver. “SDS is about interoperability and optionality – being able to move when you need to.”
SDS is about interoperability and optionality – being able to move when you need to.
Warren Olivier, Cloudera
In a market where three manufacturers – Samsung, SK Hynix, and Micron Technology – control most of the world’s memory, that freedom to move has become a strategy in itself. It also comes with a bill, because SDS cannot escape the economics of the hardware beneath it.
“SDS sounds like it frees you from hardware, but it still has to run on servers with DRAM and NAND,” says Teixeira. “You’re not escaping the hardware problem; you’re just moving it.” Those servers need the same DRAM, flash and networking.
There is a another cost too, because platforms optimised to run on multiple servers are not as efficient. If you need a petabyte of object storage, the underlying servers may need to carry closer to two, once redundancy and erasure coding have taken their share. Running that cluster will also be more expensive for the customer, from firmware and drivers to the high-speed network that ties the nodes together. “When you go SDS, you become the integrator. With appliances, all of that is the vendor’s problem,” says Teixeira.
Of course, this doesn’t mean that SDS is the wrong call, but it transfers the burden of running everything onto the buyer. “It’s a balancing act. Do you want one throat to choke and a highly efficient platform, or do you want to build and run your own cluster and take on that complexity yourself?” asks Teixeira. “Both models will exist. It’s not a religious war.”
AI is what lifts this from a procurement headache to a question of strategy. Most of the storage that feeds it never touches a public cloud. “Cloud is not a price refuge because hyperscalers pass memory and storage costs through in subscription fees,” says Jon Tullett, associate research director at IDC Sub-Saharan Africa. With IDC expecting about a third of South African organisations to move workloads from public to hybrid cloud this year, with a handful pulling them back on-premises entirely, that churn is what SDS is built to absorb, spreading data across on-premises, private and public environments while managing all of it as one.
The advice from analysts and vendors alike is to plan for scarcity rather than abundance, and to move deliberately rather than all at once. “The opportunity with SDS is to start with discovery and cleanup, then move in small, phased pilots. Prove the value on one use case, then scale,” says Cloudera’s Olivier. “It’s a methodical, almost agile way of modernising storage.”
And the work doesn’t end the moment a system goes live. Where the capital budget has run dry, the way out is often to rent rather than own. Teixeira says that an organisation should ideally look at Storage-as-a-Service for the next two or three years. “Pay for what you actually use, get through the worst of this cost spike and then reassess when the market has settled,” he suggests.
Appliances and software-defined storage will keep coexisting rather than one displacing the other, and which one fits comes down to the workload in front of you, the budget you have to work with and how long the hardware you already own still has to last. “There’s no one silver bullet,” says Teixeira. “That’s not how the world works.”
THE MEMORY CRISIS
For years, storage capacity followed one dependable rule. It was never cheap, but it grew cheaper every year, and buyers could plan around that. The convention broke in 2026. TrendForce put the rise in conventional DRAM contract prices at roughly 98% quarter-on-quarter in the first three months of the year, with NAND flash close behind, and a component that had fallen in price for a generation began climbing by the quarter. The reason sits upstream of any datacentre. A small number of producers make almost all the world’s memory, and they swung production towards the high-margin parts that AI infrastructure devours. “A handful of memory producers control 70% to 80% of the market. They’ve pivoted hard to high margin hyperscalers, and the rest of us have to get to the back of the line, says Daniel Teixeira, engineering manager at Everpure. “That’s scary if you work in infrastructure.”
In Africa, the higher prices are already changing what companies can build. Rising memory costs have sent Zoho back to the drawing board on its South African infrastructure projections. “The datacentre is going to be the next big thing. It’s a big, big cost for us,” says Andrew Bourne, Zoho’s regional head for Southern Africa. “Since the memory prices have gone through the roof, we’re busy redoing all our forecasting.” With few analysts expecting relief before 2027, and some predicting the pricing the strain will last into 2028, the practical answer across the continent is the same, to stretch the hardware you already own rather than buy your way out of a market that has stopped playing by the old rules.
* Article first published on www.itweb.co.za
Share
