Midjourney prompted by THE DECODER

The AI bubble debate boils down to one question: Is revenue at AI labs growing fast enough to pay for the massive data center buildout?

A lot of smaller questions hang off that big one. How long can purchased AI chips stay useful in production, for instance? Four, six, or eight years changes the math a lot. But the most important question is whether companies are getting enough real value from AI to keep buying more of it, ideally at rising prices. So far, the return on investment can't be documented at an economy-wide level and only shows up in anecdotes.

Whether AI pays off fast enough remains "finely balanced"

Azeem Azhar, a British tech entrepreneur and founder of the research group Exponential View, offers such an anecdote. In a podcast well worth listening to with Nicholas Thompson of The Atlantic, he describes speaking to about 160 IT vice presidents in Las Vegas. Azhar asked who could point to measurable AI results. Two-thirds stayed standing. That was more than he expected. Then he pushed further and asked who has results good enough to interrupt the CEO's summer vacation. Only about eight people remained on their feet.

Azhar's takeaway is that companies are making progress, but slowly. Whether that pace is fast enough to justify current investment levels remains an open question. Some executives say their boards are growing more ambitious after early wins. Even in slower markets like Italy, Azhar heard from CEOs that trust is building and budgets are rising despite missteps along the way.

At the same time, many companies are shifting away from expensive frontier models toward open-weight alternatives. In that scenario, AI usage might grow, but the bubble could still burst because not enough money flows back to fund the buildout.

Azhar points to what he calls the "bear version of the story." A Boston Consulting Group survey found that about 70 percent of CEOs worldwide say AI success matters for how they're perceived in their roles, giving them an incentive to paint a rosier picture than reality warrants. Azhar stresses that he doesn't have a simple answer to the AI bubble question either and that the situation is "finely balanced."

Atlantic Re:think