South African organisations appear to be moving AI workloads out of public cloud at a higher rate than companies anywhere else in a new global survey. The tempting conclusion is that enterprise cloud computing has hit some kind of local reversal. The numbers support something less dramatic and probably more consequential: companies are becoming much less willing to assume that public cloud is automatically the right place for every workload.
Research commissioned by Cloudera and conducted by Wakefield Research found that 83% of the South African organisations surveyed had moved at least some AI workloads from public cloud into private cloud or on-premises environments. That was considerably higher than the global average of 66%, as well as the Americas at 73%, APAC at 63% and EMEA at 55%.
There are reasons to be careful with that 83%. The South African sample consisted of 100 respondents, drawn from enterprise architects, cloud infrastructure leads and data architects. Companies here also only needed 250 employees to qualify for the study, compared with a 1,000-employee minimum in most of the other markets surveyed. South Africa may well be moving faster, but “highest in the world” is doing more work than a survey of 100 local technology decision-makers can comfortably carry.
There’s another qualification buried inside the headline number. Respondents weren’t saying that they had abandoned public cloud. They were saying that they had moved at least some AI workloads elsewhere, which leaves open the possibility that substantial parts of their infrastructure remain exactly where they were. The distinction changes this from a story about companies reversing their cloud strategies into one about companies becoming more selective about them.
The rest of the research supports that reading. Asked how they expect to build AI infrastructure over the next two years, 31% of the South African respondents chose a hybrid-first approach. Another 28% said they would increase cloud spending, while 26% expected greater investment in on-premises infrastructure and 15% pointed towards edge computing. Public cloud plainly isn’t disappearing from that picture. What looks less secure is its old position as the assumed destination for enterprise computing, particularly once AI enters the infrastructure budget.
Cloudera’s global version of the research gives some indication of why. Across the full survey of 1,500 respondents, the company says infrastructure costs have risen as businesses deploy AI, while existing data architectures, governance and compliance requirements are complicating deployment.
Those are Cloudera’s numbers, and Cloudera sells hybrid data and AI infrastructure, so its enthusiasm for a move towards hybrid computing isn’t exactly disinterested. But there’s a credible problem underneath the sales opportunity. AI asks companies to do more with larger quantities of data, move that information between systems, give models access to material that may previously have remained inside tightly controlled environments and pay for the computing needed to process it. Where those workloads run can therefore affect cost, risk and governance in ways that make an inherited “cloud-first” policy less useful as an answer.
South African organisations are making those decisions inside a market that has spent years investing heavily in public cloud. AWS, Microsoft and Google all operate local cloud infrastructure, and enterprises have shifted systems and applications away from their own data centres. That makes the Cloudera finding more interesting than a simple story about a country that never embraced public cloud particularly enthusiastically. The infrastructure is here, companies use it and AI is still encouraging some of them to reconsider where particular workloads should sit.
Control over data has also become a more explicit policy question. South Africa’s National Data and Cloud Policy places data sovereignty, security and cloud infrastructure within the same national policy conversation. That doesn’t tell private companies where every AI workload should run, but it reflects a wider shift towards treating the location and control of data as questions with economic, regulatory and strategic consequences rather than as purely technical choices.
The old binary between cloud and on-premises becomes particularly awkward in this environment. A bank could run one model against sensitive customer data within tightly controlled infrastructure while using public cloud services for other applications. A company might move a computationally expensive AI workload because the economics have changed while leaving dozens of conventional systems exactly where they were. Certain datasets may need tighter governance even when the applications using them span several environments.
The technology industry is already building for that messiness. AWS’s AI Factories put AWS-managed AI infrastructure inside a customer’s own data centre, while Cisco’s Sovereign Critical Infrastructure portfolio is designed for organisations that want greater control over where infrastructure operates and who can access it. The boundaries become difficult to describe cleanly when a hyperscaler can bring something resembling its cloud environment into a private data centre and traditional infrastructure vendors are simultaneously trying to give on-premises systems more cloud-like characteristics.
There’s a commercial irony in that shift. The technology industry spent years persuading enterprises that running their own infrastructure was an expensive burden and that public cloud offered a more flexible alternative. Now cloud providers have products for customers that want infrastructure closer to home, while infrastructure vendors have an obvious reason to welcome renewed spending on private environments. The customer in the middle is left with more options, but also more vendors able to argue that their particular version of hybrid is the sensible response to AI.
Warren Olivier, Cloudera’s regional vice president for Africa, says cost, data sovereignty and regulatory exposure are pulling South African AI workloads away from public cloud and that organisations increasingly want infrastructure they can “see and control”. His preferred answer is understandably hybrid, with workloads sitting on-premises, in private cloud or in public cloud depending on cost, performance and compliance.
That fits Cloudera’s business, but the useful part of the argument sits one level above the product pitch. AI is forcing companies to revisit infrastructure decisions they made when their computing requirements looked different. Earlier this year, Huawei’s AI infrastructure pitch in South Africa similarly put storage and data architecture back into a conversation that has spent the past few years obsessing over models and GPUs. Different vendors will prescribe different infrastructure, but they’re responding to the same uncomfortable fact: an AI strategy quickly becomes a data and infrastructure strategy once a company tries to deploy it at scale.
That also explains why “moving off public cloud” can be misleading language. A company doesn’t have to reject cloud computing to decide that a particular model, dataset or processing job belongs somewhere else. It can simultaneously increase its overall public-cloud spending, buy private infrastructure and run workloads at the edge. The Cloudera survey’s own South African responses suggest precisely that kind of overlap.
The 83% figure should still be treated cautiously. A sample of 100 respondents isn’t enough to establish that South Africa has permanently become some global outlier in AI infrastructure, particularly when the South African company-size threshold differed from most of the other surveyed markets. What the result does give us is a useful glimpse of companies becoming less ideological about where computing happens.
For much of the cloud era, migration itself could be presented as progress. The destination was the cloud and the remaining question was how much of the old environment still needed to get there. AI is making that journey look considerably less linear. South African organisations aren’t necessarily turning around and heading back to their own data centres. They’re starting to behave as though there was never a single destination in the first place.