AI can't grow itself into profitability

In today’s Sunday Times, Andrew McDonald from Uber shares his “head-exploding moment” from April 2026. Andrew is the CEO of Uber and he learned in April that his $140 billion ride-hailing in one quarter. And then he discovered the AI had not produced anything of real discernible benefit to the company’s bottom line.

Based on my teaching AI to medium-sized businesses, this is the same conclusion. Newly-learned AI skills will help you do your job more efficiently but faster speed of work doesn’t result in a big leap for the company’s benefit.

We started hammering home the importance of making efficient prompts so the AI doesn’t burn through a lot of tokens to start thinking.

Danny Fortson wrote a half page in the business pages of the The Sunday Times explaining how the real cost of tokens is high and immediate. “The cost of a token in the industry and the value that token generates are not completely in equilibrium yet. The cost is much higher than the value it generates,” said Jeetu Patel, president of Cisco. “This is an area I do worry about. I worry that if the token costs remain higher than the value they generate, then people will pull back.”

This realisation lands directly in front of AI companies going public with record-breaking valuations that bear no relation to traditional financial metrics. All three companies hitting the market are losing billions of dollars annually because, despite their breakneck revenue and user growth, they are spending far more to develop new models and build data centres. They are, in effect, subsidising their customers.

It appears to me that the explosive growth they have experienced will not continue unabated and that these companies won’t grow their way to profit. For the purported epoch-defining run of trillion-dollar stock market floats coming down the track this sudden bout of financial sobriety from Al’s biggest spenders could mean that those astronomical sales curves could level off - or even drop. Al rationing could, in fact, become the pin that pops the feared Al bubble.

This month Uber rolled out a monthly AI spending cap of $1,500 per employee. Microsoft made a similar move last month, cancelling subscriptions to Anthropic’s powerful but pricey Claude coding tool in lieu of a cheaper internal alternative.

By the end of 2026, we will have a few frothy companies buried in the AI graveyard.