COLUMN - Imagine you run a bakery that can produce 100 loaves a day. Customers have signed contracts for several hundred loaves a day, but you simply cannot bake them. You don't have the ovens.

So you spend heavily. New ovens, a bigger building, an upgraded electricity supply, more delivery vans. 

Your bank balance looks terrible for a while. But you are not building in the hope that customers might arrive one day. 

The orders are already sitting on your desk.

That, in a nutshell, is what the world's biggest technology companies say they are doing right now. Whether you believe them is the question behind all the "AI bubble" talk in markets this month.

The factories behind AI

Over the past two weeks, Amazon, Microsoft and Google all released their quarterly results. One theme dominated: the extraordinary amounts being spent on AI infrastructure. These three companies plan to spend roughly $595 billion this year. 

That's larger than South Africa's annual economic output.

AI may feel like software, but it runs on an enormous amount of physical hardware. When you ask an AI system a question, the work is done by specialised computer chips sitting inside huge data centres.

Think of the chips as the ovens and the data centres as the bakeries. The electricity, cooling and network connections are everything else needed to keep the bread coming out.

Why investors are nervous

Not because these companies can't afford it. They are among the most profitable businesses in history. Amazon made $27.5 billion in operating profit last quarter. Google made $40.8 billion. Microsoft $40.6 billion. In a single quarter.

The worry is about the return. Building AI capacity is expensive, and the chips can become outdated quickly. For the spending to make sense, customers must eventually pay enough for AI services to cover the cost of all this infrastructure and still leave an attractive profit.

That's not guaranteed. AI usage could grow slower than hoped. Competition could force prices down. Customers may find AI useful, but not useful enough to justify the amounts being discussed.

When Google raised its spending plans two weeks ago, its share price fell 6% in a day, despite excellent results. That was the market saying: we see the spending, now show us the return.

So where are the bread orders?

This is where the latest results get interesting. Each company reports a backlog: services customers have committed to buy but which have not yet been delivered. Contractual commitments, not management forecasts.

Combined, that is $1.69 trillion of contracted backlog against $595 billion of planned spending. In bakery terms, the order book is nearly three times the cost of this year's new ovens.

And all three said the same thing on their earnings calls: they cannot build fast enough.

 Amazon's CEO said that even at $220 billion of spending, they won't have enough capacity in 2026, and probably not in 2027 either. Google said demand keeps exceeding supply. Microsoft's backlog grew by $51 billion in a single quarter, and it grew even when you exclude its biggest AI customer, OpenAI.

One qualification

The backlog is reassuring, but it doesn't prove the spending will earn a good return. A full order book doesn't guarantee your new ovens will pay for themselves.

Backlog is revenue, not profit. From that revenue the companies must still pay for electricity, staff, maintenance and the constant replacement of ageing chips.

The contracts also run over several years. Microsoft says only about 30% of its backlog becomes revenue in the next 12 months. A $500 billion backlog does not mean $500 billion arriving this year.

So the numbers are strong evidence of real demand. They are not a mathematical guarantee of profit.

Is there an AI bubble?

There may well be pockets of speculation in the AI market. Some companies will invest too much. Some data centres will earn disappointing returns. 

Some AI businesses will fail, and certain share prices may already reflect expectations that are impossible to meet.

But the latest results do not suggest that the largest technology companies are blindly building infrastructure nobody wants. Their cloud businesses are accelerating. Customers have signed enormous long-term contracts. 

New capacity is used the moment it comes online, and demand still exceeds supply.

The unanswered question is not whether demand exists. It is whether the profit eventually earned from that demand will justify the extraordinary capital going in.

As investors, we should be careful of both extremes. We should not assume every AI investment will succeed simply because the technology is exciting. But we should also not assume that heavy spending automatically means a bubble.

Sometimes a company builds a new factory because management has become carried away.

Sometimes it builds one because customers are already waiting outside.

Matthew Matthee has a wealth management business that specialises in retirement planning and investments. He writes about financial markets, investments, and investor psychology. He holds a Masters Degree in Economics from Stellenbosch University and a Post Graduate Diploma in Financial Planning from UFS. [email protected]      

‘We bring you the latest Garden Route, Hessequa, Karoo news’