This is the 11th instalment of AI Fluency Corner, a 16-part weekly series in Business Day building one connected mental model of AI in plain language.You ask an AI assistant to summarise a 50-page board pack, and 10 seconds later the summary appears. Nothing suggests the file went anywhere. Yet your document almost certainly travelled hundreds of kilometres, was processed by machines you will never see, and returned. This overlooked part of AI carries some of its largest risk. Most organisations think they are buying intelligence. Usually, they are buying access to someone else’s computers, inside someone else’s buildings, run under technical rules and contract terms few decision-makers have read. Before a model can answer, it needs somewhere physical to run and for your data to live. This week, we look at where that is.The cloud is not a place, it’s a business modelStrip away the marketing, and three plain words explain almost everything. A server is a powerful computer built to serve other computers. Servers are the digital economy’s workhorses: racks of machines packed with processors, memory, storage and increasingly the specialised graphics chips, or GPUs, that AI depends on. One AI training run might harness thousands in parallel. A data centre is the factory housing them: a secured building with backup power, industrial cooling, biometric access and round-the-clock monitoring. The cloud is simply the ability to use all of this remotely, over the internet, instead of running it yourself. Think of your laptop as a restaurant table and the server as the kitchen: you order where you sit, but the cooking happens elsewhere. This is not abstract for South Africa. The country hosts much of Africa’s large data centre capacity and is its biggest cloud market, with global providers investing billions of rand in local facilities so data can be processed closer to home, at lower latency and within the Protection of Personal Information Act’s (Popia’s) reach.Why AI and the cloud grew up togetherAI is often discussed as if it arrived alone. It did not. Modern AI became practical largely because the cloud made vast computing power rentable by the hour. Training a model can demand more processing in an afternoon than an office could supply in weeks. Large language models digest trillions of words; fraud engines weigh millions of transactions. Almost no organisation owns enough hardware, so they rent it. Much as factories stopped generating their own electricity and drew it from the grid, computing moved from something firms owned to something they consume. So in essence, when a company buys AI, it usually outsources computation. The model is visible; the real purchase is access to machines it could never afford to build. South Africa’s cloud computing sector was worth roughly $6.4bn in 2025 and is projected to reach $40bn by 2034, with AI driving growth and generative-AI adoption in large local enterprises near 67%.Control versus flexibility, and why most choose bothNone of this makes ownership obsolete. On-premises infrastructure — your own machines in your own building — offers direct control: you decide where data sits, who touches it and which laws apply. That matters for the most sensitive records and for operations where connectivity is unreliable. But it demands heavy upfront investment, skilled staff, cooling, backups and equipment that may sit idle.The cloud offers the opposite: speed, scale and resilience without buying everything first, backed by security investment few single firms could match. Its risks are different, not absent: bills that balloon as usage grows, dependence on a single provider, and outages that take thousands of customers down at once. There is a quieter cost too: providers make it cheap to move data in and expensive to take it out, so the price of leaving is rarely as visible as the price of joining. The honest answer for most organisations is, therefore, not one or the other but both: a hybrid approach. A bank may keep core payments in-house while training models in the cloud; a hospital may hold patient records locally while using cloud analytics for planning. The right question is never “Should everything move to the cloud?” but, “Which workloads benefit from cloud economics, and which demand local control?”The security question, and where your data really livesLearning your information sits elsewhere raises an obvious question: is it safe? Cloud security is shared security. The provider secures the building; you still decide who gets an access card and whether confidential files are locked away. A world-class platform can be badly configured, and most breaches begin not with someone smashing through the wall but with someone handed the wrong key.Geography compounds this. Under Popia, when customers’ data sits on servers abroad you have made a cross-border transfer, which is lawful only under specific conditions, while the responsible party stays liable. You cannot outsource accountability with storage.The most dangerous misconception in AI is that the technology is the asset. Usually it is not. The model can be replaced; customer histories, contracts and institutional knowledge cannot. A vendor demo reveals capability but little about ownership, residency, portability or exit. Your data does not simply live “online”. It lives on real machines, in real buildings, across networks and jurisdictions, under permissions and contracts. The cloud may vanish from sight. Your responsibility does not.Our task this weekChoose one cloud or AI tool your team relies on and, for 10 minutes, ignore what it does. Instead, trace one piece of information through it. Where is it stored? Where is it processed? Who can access it, including the provider’s own staff? Which country’s laws would reach it? And if you left the provider tomorrow, how would you get your data back, and in what format? Most organisations can explain what their systems do. Far fewer can explain where those systems actually live — and under Popia, that gap is your liability, not the vendor’s.• Mafinyani is senior partner in financial engineering & artificial intelligence at specialised finance, risk and applied technology firm Intellica Analytics.Next week: Enterprise AI — keeping data inside the walls. If the cloud means your information lives on someone else’s computer, the next question is how serious organisations use powerful AI without handing their most sensitive data to the open internet — and why the consumer tool and the enterprise-safe version are not the same thing.