The AI ​​bubble will burst unless businesses can grow 25 times faster than Amazon

Tuesday 21 October 2025 12:26 WIB
| Updated:

Tuesday 21 October 2025 12:36 WIB

Share

• 
  Share on Facebook
• 
  Share on Twitter
• 
  Share on LinkedIn
• 
  Share on WhatsApp
• 
  Share on Email

When the numbers get big enough, it no longer makes sense, and the scale of investment in AI has made the dotcom bubble smaller, says Chris Clothier.

Hardly a day goes by without a new announcement regarding huge amounts of money that will be spent on AI. In early October, Open AI, the company behind ChatGPT, announced a partnership with AMD worth “tens of billions of dollars.” Just before that they announced a deal with Nvidia for $100 billion and last September a deal with Oracle to spend $300 billion. This is just an OpenAI deal. The so-called Hyperscalers – Google, Amazon, Microsoft, Meta, and the like – are expected to collectively spend about $500 billion per year over the next few years to build AI infrastructure.

We find this very worrying.

Our first concern is a matter of history. Every technological boom in history has been accompanied by overinvestment resulting in massive misallocation of capital. The results were similar on each occasion. Society benefits two-fold, firstly from the adoption of new technology (lines, trains, bicycles, cars, telecommunications) and secondly from excess supply which means that the cost of the new technology to the final consumer will be lower than it would otherwise be. Those gains came at the expense of investors who – for the most part – experienced far worse impacts.

It seems this error is recurring. The scale of investment really goes far beyond the Dotcom bubble. At its peak during the internet/telecommunications boom of the late 1990s, annual capital spending in those sectors was $150 billion per year, roughly equivalent to “only” $300 billion today.

Basic mathematics

We can do some basic calculations to find out what income is needed to support current investments. Assume collective spending of $500 billion per year through 2030. That’s a cumulative spending of $3 trillion. Assume, on a charitable basis, that these assets will have an average lifespan of 10 years. By 2030, this means annual depreciation costs will reach $300 billion per year. Companies like Google Cloud and Amazon Web Services (AWS) already provide AI computing. It seems reasonable to assume that their business model offers a reasonable template. So this shows an operating margin of 20-30 percent of sales, depreciation of 10-15 percent, and the remainder as operating costs (55-70 percent). This means revenues of $2-3 trillion per year are needed by 2030.

When the numbers get big enough, I have trouble understanding them. What does $2-3 trillion really mean? Well, AWS is the market leader in providing cloud computing. It is also one of the most successful “start-up” companies in history. It was launched in 2006 and currently has revenues of around $125 billion. So, to justify that investment would require new business growth of 16 to 24 times AWS in less than 5 years.

Another way to think about this is through the prism of US GDP. Today it stands at around $30 trillion, $2-3 trillion representing between six and 10 percent of GDP. Where does this spending come from? There are three possible sources. First, it displaces spending from other parts of the economy. If this happens then we can expect that businesses outside of technology in the US will experience low growth or even no growth at all as consumer and business spending is diverted.

Second, AI may be used to cut costs by replacing labor. Fears of job losses due to innovation are ancient in the discipline of economics. But historically, these fears have been overblown: the farm and factory workers of the past are now baristas, personal trainers and software engineers. But such a large displacement in such a short time was unprecedented and therefore very painful. The newly unemployed will definitely find work eventuallybut it will take time.

A third possibility is that the rise of AI produces gradual changes in economic growth rates that, in turn, finance the large expenditures required by industry. That’s what AI advocates hope. We are more skeptical. This reminds me of Robert Solow’s quip that “you can see the age of computers anywhere but in productivity statistics.”

Most likely this will not happen and AI revenues will be very disappointing. Public companies that spend on capital expenditures generate large amounts of free cash flow. They will probably continue to generate a lot of free cash flow from their core business in the future. So the market may ignore any mistakes just as they ignored the $45 billion that Meta spent on a “meta-verse” that, right now, is basically worthless. So this doesn’t need to be an economic disaster. As Jeff Bezos observes, this is a “virtuous bubble” by which he means that – unlike financial maniacs – the chips bought, and the data centers built, will ultimately be put to good use, even if they produce poor financial returns.

But there are two reasons to be careful. First, by some estimates, AI-related data center spending is the only thing keeping US GDP positive. If earnings disappoint and spending is constrained then US growth could slow quickly. Second, the disappointment will likely impact share prices. With savings rates in the US historically low and private allocations to equities high, even a small pullback in equity markets could result in a large decline in final demand.

Chris Clothier is co-CIO and co-manager at CG Asset Management