Schneider Electric’s warning is not really about artificial intelligence. It is about the socket in the wall.
The company’s East Africa president, Ifeanyi Odoh, told TechCabal that African governments are asking the wrong question when they focus mainly on generating more electricity. His argument is sharper than the usual “Africa needs more power” line. The grid itself, he says, is becoming the constraint. It was built for another period, before AI data centres, cloud regions, electric vehicles and distributed renewable energy started competing for serious, reliable electricity.
That is the part business leaders should pay attention to. The AI conversation has become crowded with model names, chatbots, copilots and automation pilots. Those matter. But none of them matter very much if the underlying digital infrastructure cannot be powered consistently, cooled properly and scaled without unpleasant surprises.
In practical terms, Africa’s AI race may be decided less by who announces the boldest national AI policy and more by who can run dependable compute at industrial scale.
The grid is becoming part of your technology stack
Most executives still treat electricity as a facilities issue. IT teams worry about cloud architecture, cybersecurity, system uptime and integrations. Operations teams worry about generators, inverters, diesel, solar, batteries and landlord promises. Finance sees the bills. Everybody suffers when the assumptions fail.
AI makes that separation harder to defend.
A basic business automation project may not change your energy profile much. A customer service chatbot, a document processing workflow or a forecasting dashboard can run comfortably on cloud services. But as organisations move towards heavier workloads, private data environments, local hosting requirements, real-time analytics, computer vision, call centre intelligence, fraud detection or sector-specific AI systems, the power question becomes a design question.
Where is the workload hosted? What level of uptime is acceptable? What happens when connectivity is fine but power to a local node is not? Which systems must keep running during outages? How much latency can the business tolerate? If your AI system supports credit decisions, dispatch routing, hospital administration, payments, manufacturing or security monitoring, the answer cannot be “we will see.”
This is why Schneider Electric’s point lands. More generation is useful, but a grid designed around old demand patterns will struggle with new digital loads. AI data centres do not consume electricity casually. Cloud infrastructure does not forgive unstable supply. EV charging and distributed renewables add further complexity because power is no longer flowing in the simple, predictable way older grids assumed.
For businesses, the message is not to panic or postpone AI plans. It is to stop treating infrastructure as an afterthought. AI readiness is not only about data quality and staff training. It is also about power, cooling, hosting choices, network resilience and disaster recovery.
The overhyped part is the idea that every company needs to build around AI infrastructure as if it were a hyperscaler. Most do not. A retailer, school, hospital, logistics company, insurer or professional services firm does not need to think like a data centre operator. But it does need to know which parts of its operation can tolerate interruption and which cannot.
That line is often unclear until a rollout fails.
For African businesses, “cloud-first” still needs local discipline
One tempting answer is to push everything to the cloud and let large providers handle the power problem. For many companies, that is still the right default. Cloud services reduce the burden of maintaining servers, improve access to modern tools and make it easier to test AI use cases without buying expensive hardware.
But cloud does not remove local operating reality. Your staff still need access. Your branches still depend on connectivity. Your devices still need power. Your payment points, scanners, routers, customer kiosks, CCTV systems, biometric devices and mobile apps still sit in the real world.
In Ghana and across many African markets, serious technology planning has to assume mixed conditions. Some customers will be mobile-first. Some branches will have inconsistent connectivity. Some teams will work around outages with manual processes if the system design allows them to. Some integrations will touch banks, mobile money providers, logistics partners, government platforms or legacy databases that do not always behave like a clean diagram.
This is where a lot of AI ambition becomes expensive. Not because the algorithm is impossible, but because the process around it was never mapped properly.
If your team still reconciles orders in spreadsheets after an outage, an AI dashboard will not fix the control weakness. If your customer records are duplicated across three systems, a chatbot may simply respond faster with the wrong information. If your field staff depend on a mobile app that assumes perfect connectivity, the model behind it can be excellent and still fail in daily use.
The grid conversation should push companies to ask better questions before procurement. Not “Which AI tool should we buy?” but “Which business process deserves AI, and what must be true for it to work every day?”
That includes electricity, but it also includes data governance, integration quality, user training, vendor accountability and fallback procedures. Most rollouts stall at the handover point, not the demo point. The demo has clean data, stable internet and motivated users. The branch does not always have those luxuries.
The companies that win will design for constraint
There is a useful discipline in building for environments where infrastructure cannot be taken for granted. It forces clarity.
A business that designs its systems around intermittent connectivity often ends up with better offline workflows. A company that classifies which services are mission-critical usually makes better cloud and backup decisions. A team that measures process failure points before adding AI tends to automate the right work rather than the most fashionable work.
Schneider Electric’s framing of the electricity grid as the next competitive battleground should not be read only as a government or utility sector issue. It will affect private investment decisions as well. Data centre location, cloud adoption, edge computing, renewable energy planning and business continuity will become part of the same conversation.
For decision-makers, the sensible move is to separate AI experiments from AI dependency.
Experiments can be light. Run a pilot. Test a workflow. Use cloud tools. Measure time saved. Learn where the data is messy. Keep the risk low.
Dependency is different. Once an AI-supported system becomes part of daily operations, the infrastructure beneath it deserves board-level attention. That does not mean every board member must understand model architecture. It means somebody must be accountable for uptime assumptions, data controls, vendor risk, security exposure, energy resilience and the cost of failure.
The quiet opportunity is that many African companies are not locked into decades of rigid technology architecture. They can make cleaner choices now: mobile-first interfaces, cloud services where sensible, local resilience where required, APIs instead of manual file exchange, role-based access controls, and automation that respects how staff actually work.
But clean choices do not happen by accident. They happen when management refuses to buy technology as theatre.
AI will keep attracting attention because it is visible. Power infrastructure is less glamorous until it fails. Schneider Electric’s argument is a reminder that the next phase of digital competition will be physical as well as software-driven. The businesses that understand that early will plan differently: not slower, but with fewer blind spots.
At Exquode, this is the kind of conversation we prefer to have before a system is built, not after it disappoints. If your organisation is exploring AI, automation, cloud migration or a more reliable digital operating model, the useful starting point is a hard look at the process, data, infrastructure and failure points that will decide whether the technology works in practice.