AI Is Not the Bubble. The Price of Believing in It Might Be

For much of the past three years, questioning artificial intelligence valuations has risked sounding like questioning artificial intelligence itself. The two are not the same.

AI can be a profound technological shift and still be accompanied by an investment bubble. The internet was real. Railways were real. Electrification was real. Each transformed the economy, but each also produced periods when investors paid prices that the eventual economics struggled to justify.

That distinction is becoming increasingly important.

Valuations Are Running Ahead

The European Central Bank has now warned that current technology valuations are once again approaching levels associated with the dotcom era. Its analysis points to a familiar historical pattern in which major technological revolutions attract extraordinary optimism, capital and speculation before the economic benefits are fully understood.

The important point is that a correction in valuations would not mean AI had failed. It could simply mean investors had become too optimistic about how quickly the financial returns would arrive.

The internet continued to transform the world after the technology crash of 2000. Many of the companies built around it did not survive. A technology can change the world while still producing poor investments along the way.

The Scale of Investment Is Extraordinary

The amount of capital now flowing into AI infrastructure is difficult to overstate. US capital expenditure has moved beyond one trillion dollars annually, while the largest technology companies are increasingly turning to debt markets to finance data centres, chips, power and computing capacity.

This marks an important change in the nature of the AI boom. The first phase was largely financed from the huge cash flows of the world’s largest technology companies. The next phase is increasingly being supported by borrowing, leases and infrastructure finance.

That does not make the investment irrational. AI requires enormous physical infrastructure. But the use of more debt means the financial returns matter much more.

The question is no longer simply whether AI will work. Investors now need to know whether AI will produce enough economic value, quickly enough, to justify the capital being committed.

The Missing Number Is Return

Corporate enthusiasm for AI is now almost universal, but detailed evidence of financial returns remains surprisingly limited. Many companies talk extensively about AI adoption, productivity and transformation while providing relatively little information about what that investment is contributing to profits.

That gap will become increasingly difficult to ignore.

Investors will want to know how much cost has been removed, how much additional revenue has been created and whether employees are genuinely becoming more productive. They will also want evidence that AI investment is improving margins rather than simply adding another large layer of technology spending.

The conversation is likely to move rapidly from how much companies are investing in AI to what they are getting back.

The Opportunity Is Moving Towards Infrastructure

None of this removes the investment opportunity. It may simply change where the strongest opportunities are found.

Power generation and grid infrastructure matter because AI requires enormous amounts of electricity. Semiconductors, networking, cooling and data centres matter because computing capacity is becoming industrial infrastructure. Software that reduces the cost of running AI systems may become increasingly valuable as companies attempt to control their spending.

Enterprise applications may also prove particularly attractive where they can demonstrate measurable reductions in cost or increases in revenue. The strongest businesses may not be those making the biggest claims about AI, but those able to show precisely what economic benefit their products create.

The Risk Is Becoming Financial

Until recently, much of the debate around AI focused on technology risk. Would the systems become capable enough to justify the excitement surrounding them?

That question is becoming less important. AI is already sufficiently capable to influence software development, marketing, research, customer service, professional services and many other industries.

The emerging risk is financial. How much infrastructure needs to be built, how much debt will finance it and what return will investors receive on that capital?

A highly successful technology can still disappoint investors if the profits arrive more slowly than expected. Markets do not require AI to fail for valuations to fall. They only require the financial outcome to be slightly less impressive than the assumptions already reflected in share prices.

The Lomond Logic

The most useful question is therefore not whether AI is a bubble or whether AI will transform the economy. Both ideas can be true at the same time.

History suggests that transformative technologies often attract excessive investment precisely because their long term potential is so compelling. Investors can correctly identify the future and still pay too much for it.

The next phase of the AI cycle is likely to be less about excitement and more about evidence. Companies will increasingly be judged on whether their AI investment creates cash flow, improves productivity and generates an acceptable return on capital.

AI does not need to fail for today’s valuations to correct. Expectations only need to have travelled further than reality.