Artificial intelligence can change the world and still be a bad investment
What if artificial intelligence (AI) changes almost everything, yet still proves to be one of the worst investments of all time? It sounds contradictory, but it is not. Technology can transform society while the companies that build it spend too much and leave investors to foot the bill. The internet did precisely that: The world gained, even though many shareholders lost money.
This possibility hangs over AI. The biggest technology companies in the United States (US) could spend $900 billion on chips, data centres, and power in 2026, rising to $1.4 trillion in 2027. Yet, AI revenue is estimated at $150 billion to $220 billion. The industry may need $2.5 trillion a year to justify spending. This does not make AI a fraud or a bubble. Railways, electricity, and the internet all required substantial investment before their productivity gains became apparent. But, transformative infrastructure often rewards society more than its financiers. Technology can change everything while overbuilding, competition, and falling prices ruin the economics of those funding it.
China makes this tension impossible to ignore: Its firms are spending less than a tenth as much on data centres as their US rivals, yet their best models are approaching the frontier. Lower land and labour costs help. So do distillation and engineering. More surprisingly, US chip restrictions have forced Chinese developers to squeeze more performance from less computing power.
Scarcity has imposed discipline, but it has also created waiting lists, rationed services, and long processing delays. China proves neither that frugality always wins nor that spending is irrelevant. It shows that the meaningful measure of AI is not how clever a model appears, but how much useful output each dollar produces.
The US may be failing this test. Businesses are buying AI before rebuilding themselves around it. Nine out of ten executives report no productivity effect over the past three years. The missing investment is not another chatbot; it is clean data, redesigned workflows, staff training, and managers willing to discard obsolete processes. Otherwise, AI merely helps employees to perform yesterday’s work faster.
India offers a picture of what this means for workers. Its technology workforce has not collapsed, but routine outsourced work is weakening while in-house capability centres expand. Demand is shifting from generic information technology skills towards people who combine AI, technical knowledge, and commercial judgement. The danger is not that every job disappears. It is that entry-level tasks vanish before young workers gain the experience required for more valuable tasks.
Then, there is South Korea, where AI demand has generated chip profits, bonuses, and tax receipts. Its stock market doubled, then suffered a violent correction as concentration and borrowed money magnified the reversal. Once again, an excellent industrial story became a dangerous trade. The sellers of picks and shovels may earn fortunes during a gold rush, but their shareholders can still overpay for those fortunes.
This matters to South Africans, whose global portfolios are increasingly exposed to US technology shares. Investors must separate three questions: Is AI transformative? Which companies will capture the value? Is that value already reflected in the price? A “yes” to the first says little about the other two.
South Africa cannot compete by building more data centres than the US, directing more capital than China, or manufacturing more chips than Korea. Oddly, that may be helpful as capital scarcity should force us to concentrate on adoption: Applying AI to financial services, mining, agriculture, logistics, and healthcare; lowering the cost of serving overlooked customers; and raising the output per worker.
In a country with mass unemployment, success cannot mean automating the largest possible number of jobs. It must mean producing more, creating new markets, and enabling workers to become more valuable. Every AI investment should face three practical questions: What costly problem disappears? What additional revenue becomes possible? How soon does the return exceed the full cost of changing the organisation? The AI race will not ultimately be won by whoever spends the most. It will be won by whoever wastes the least.
