AI’s soaring costs are outpacing the returns

And the gap is growing

Enterprise AI spending is accelerating, but productivity gains have yet to generate the new markets and revenue streams needed to make the economics sustainable.

AI costs are rising sharply for enterprise, but those enterprises face a problem: new markets and revenue streams aren’t being created fast enough to pay the bill.

Two new reports from separate consulting firms (itself a sector on which data suggests AI has already had a considerable impact) indicate concerns about whether current levels of AI spending are sustainable.

A $4.2 trillion gap

The first of these is a report by global management consultancy Bain & Company which says that the aggregate compute demand from the AI industry demands $6 trillion in annual revenues within the next five years. The consultancy estimates that the consumer and enterprise AI market could total between $1.2 trillion and $1.8 trillion within the same time frame.

This means that the AI industry needs $4.2 trillion from markets and products that, as of today, no not exist, to pay its compute bills.

93% enterprises exceeded AI budget

At the same time, enterprises are increasingly aware of and concerned about rising AI costs. A McKinsey report which focuses on US government organisations (many of which have, since January 2025, been drastically downsized in terms of people and upsized in terms of agentic AI deployment), concludes that whilst the cost per token of AI has reduced, the total bill has shot up because the amount of tokens required for all the agentic activity has also shot up.

This is why Tokenmaxxing is no longer a thing.

The same report from McKinsey found that 93% of 120 private enterprises who were considered reasonably mature in terms of AI adoption, has exceeded their planned AI budgets.

There isn’t enough data available yet to be able to explain away exactly why agentic AI isn’t delivering in the way that analysts and markets seemed to assume it would.

More time hasn’t led to more innovation

In his forward to the report’s main findings, Bain & Company’s Chairman of Global technology, Media and Telecommunications Practice David Crawford gets to the heart of the problem, saying:

“AI must do more than boost labor productivity on existing tasks; it needs to unlock new sources of growth and value. Superintelligence should be used for superhuman endeavors, not just better execution of existing, mundane work.”

It’s an interesting take because for several years now, generative and agentic AI has been sold with just that message – automate the mundane and free up time for the strategic brilliance that will reimagine your business model and create new markets and boost revenue streams.

The problem is that the first part hasn’t led to the second. Most enterprises and government organisations probably are a little bit more productive, but being able whistle up a deck in five minutes as opposed to spending half a day on the same task hasn’t yet led to the wave of innovation necessary to make all this infrastructure financially viable.

Unless organisations can turn what still look like relatively modest efficiency gains into genuinely new products, services and revenue streams, the economics of AI are likely to become more not less challenging.

AI deployments will be judged by a single metric. Can it create enough value to pay for itself?