Category: Technology | Published: 2026-06-09
There is a number worth sitting with for a moment. The five largest technology companies, Amazon, Alphabet, Meta, Microsoft, and Oracle, are projected to spend somewhere between 660 and 690 billion dollars on capital infrastructure in 2026 alone. Roughly three quarters of that, around 450 billion dollars, is directly tied to AI: the servers, the chips, the data centres, and the power systems needed to train and run increasingly large AI models.
For context, that is more than the annual GDP of many developed economies, spent in a single year by five companies on a single technology bet. Goldman Sachs estimates the four largest hyperscalers will collectively spend 5.3 trillion dollars between 2025 and 2030. Morgan Stanley puts global data centre spending over that same period at 3 trillion dollars. JPMorgan and McKinsey have estimated total global AI and data centre spending through 2030 at between 5.5 and 7 trillion dollars.
Those are the kinds of numbers that attract economists as well as investors. And some of the most respected voices in global finance are now asking a straightforward question: can the returns possibly match the spending?
What History Suggests
The Bank for International Settlements, often described as the central bank for central banks and one of the most influential economic institutions in the world, has published a warning that sits uncomfortably alongside those investment figures.
The BIS has drawn explicit parallels between the current AI investment boom and three previous periods of technological excitement: Britain's railway mania of the nineteenth century, the electrification boom of the 1920s, and the dotcom bubble of the late 1990s. Each of those periods was driven by a genuine technological breakthrough that really did change the world. And each of them attracted far more capital than the eventual commercial returns could justify.
The BIS puts it directly: previous investment booms all shared one common trait, a genuine technological breakthrough that attracted capital in excess of what commercial returns could ultimately justify.
The critical point in that formulation is that the BIS is not saying AI is worthless. It is saying that even when the technology is real and the transformation is real, the investment can still be excessive, and the consequences when it corrects can be severe.
The Dotcom Comparison Is More Nuanced Than It Sounds
The dotcom comparison is often dismissed because the internet obviously did succeed. Amazon, Google, and countless other companies that emerged from that era went on to reshape entire industries. The internet transformed how we work, shop, communicate, and consume information in ways that arguably exceeded even the most optimistic projections of the late 1990s.
But that is precisely the point the BIS is making. The technology was real. The transformation was real. And the investment bubble still burst. Enormous sums were lost. Companies that should have survived did not, because funding dried up. Investment that would have accelerated genuine innovation was destroyed. The wider economy felt the effects through tighter credit, falling business confidence, and a prolonged period of reduced technology investment.
AI could follow exactly the same path. The technology succeeds and reshapes everything we do, and a correction in AI investment still causes significant economic damage along the way.
The Scale of What Is at Stake
One of the underappreciated risks in the AI investment boom is how many industries are now connected to it, far beyond the technology companies making the bets.
Building data centres at this scale requires construction companies, engineering contractors, specialist equipment manufacturers, and enormous amounts of electricity. Power grids in multiple countries are being expanded partly in anticipation of AI demand. Property developers are acquiring land and planning facilities. Chip manufacturers have ramped production capacity. Financial institutions have lent heavily into the ecosystem.
The BIS has warned that a sudden pullback in AI investment spending could turn what it calls the capex boom into a protracted investment bust, with knock-on effects spreading through financial conditions well beyond the technology sector itself.
Morgan Stanley has noted that roughly half of the 3 trillion dollars expected to be spent on data centres through 2028 will be covered by private credit rather than traditional public market financing. If a correction hits and those private lenders become cautious, the effect on credit availability could spread well beyond AI-related businesses.
The AI Agent Risk in Financial Markets
Separately from the investment bubble question, there is a second AI-related financial risk that regulators are watching closely. Sarah Breeden, Deputy Governor for Financial Stability at the Bank of England, has warned that autonomous AI agents operating in financial markets could amplify instability rather than smooth it.
The concern is one of herding. AI systems trained in similar ways on similar data and pursuing similar objectives may respond to market stress in broadly similar ways. In a scenario where large numbers of autonomous systems sell, buy, or withdraw from positions simultaneously, the speed and scale of the response could be significantly greater than anything human traders could produce.
Breeden has warned that agentic AI could amplify volatility in stress events. Regulators are actively considering whether mechanisms such as circuit breakers or kill switches may eventually be needed to interrupt AI-driven market behaviour before it accelerates a crisis rather than containing one.
This is no longer a theoretical technology debate. Financial regulators are treating AI market risk as a serious stability concern.
The Productivity Question That Determines Everything
Underneath both of these risks is a simpler question: is AI actually generating enough economic value to justify what is being spent on it?
The evidence is mixed. Research from the National Bureau of Economic Research found meaningful productivity improvements from generative AI in customer support work, with particularly strong results among less experienced workers who benefited most from AI-assisted guidance. That is a genuine, measurable gain.
The challenge is extrapolating from individual task-level efficiency to sustained, organisation-wide financial returns. A business may save a team member thirty minutes on a task. That does not automatically reduce headcount, lower costs, or increase revenue. The gap between impressive demonstrations and measurable improvements to the bottom line is one that most organisations adopting AI are still working to bridge.
Hyperscalers are spending 45 to 57 per cent of their revenue on capital expenditure, ratios that historically characterise utility or industrial companies rather than technology businesses. For those ratios to make financial sense at the scale being committed, AI needs to unlock productivity and revenue gains that are not yet visible in the data at anywhere near the required magnitude.
What a Correction Would Actually Look Like
If AI investment does pull back sharply, the effect on most businesses would not come from AI failing directly. It would come through the channels that connect AI infrastructure spending to the wider economy.
Tighter credit conditions would make borrowing more expensive or harder to access. Construction and engineering companies that expanded to meet data centre demand would pull back. Energy companies that committed to new capacity would face different demand projections. Technology companies that recruited and invested heavily would reduce spending. All of that would ripple into employment, business confidence, and investment decisions well beyond the technology sector.
None of this is inevitable. The BIS and the Bank of England are expressing concern, not issuing predictions. The AI investment cycle may generate the returns that justify the spending, in which case the boom continues and the productivity transformation eventually shows up in economic data.
What This Means for Businesses Adopting AI
For most businesses, the right response to the bubble debate is neither to accelerate recklessly nor to wait for certainty that will not arrive on any predictable schedule.
The sensible approach is to focus AI investment on areas where the return is measurable and near-term rather than speculative and long-term. What tasks take up significant time that AI can genuinely reduce? What processes have clear inputs and outputs where AI assistance can be evaluated objectively? Where can a business run a genuine pilot with real metrics before committing to broader deployment?
That discipline matters regardless of whether the broader AI investment cycle corrects or continues. Businesses that adopt AI tools because they have identified specific, measurable value will be better positioned than those that adopt them because the technology is exciting or because competitors appear to be moving.
If you want practical guidance on where AI tools can deliver genuine, measurable returns for your business rather than a theoretical conversation about future potential, our AI Consultancy page is a good starting point.