About 107 of 160 IT vice presidents in Las Vegas said they could point to measurable results from AI, but only around eight said the results were strong enough to interrupt the CEO's vacation. The account comes from Azeem Azhar, founder of Exponential View, in a podcast with Nicholas Thompson of The Atlantic. Azhar said the first result exceeded his expectations, while the second showed how few cases look decisive. The gap matters because AI labs need buyers to keep expanding purchases to fund the data center buildout.
Las Vegas poll and the data center question
Azhar described addressing a room of about 160 IT vice presidents and asking who had measurable AI outcomes to report. Two-thirds stayed standing, a share higher than he had expected before the session. He then raised the bar and asked who had outcomes compelling enough to justify calling the chief executive during a summer break. About eight people remained on their feet after the second question. The exchange was recounted in his conversation with Thompson, and Azhar presented it as evidence of slow but real enterprise progress rather than broad transformation.
The central business question behind the anecdote is whether revenue at AI labs is growing fast enough to pay for the data center buildout now underway. That calculation turns on how long purchased AI chips remain useful in production, with four, six or eight years producing very different outcomes. It also depends on whether enterprise customers see enough value to keep buying more capacity, ideally at rising prices. Without that continued demand, infrastructure spending lacks a funding base. Documented return at the level of the whole economy is still missing, with evidence limited to individual anecdotes.
Azhar said some executives report that boards have become more ambitious after early wins, which supports further spending. He cited Italy as a slower market where chief executives still described growing trust and rising budgets despite missteps along the way. At the same time, many companies are moving from expensive frontier models toward open-weight alternatives to control costs. In that pattern, overall AI usage can keep growing while less money flows back to the providers building large-scale infrastructure. Azhar stressed that he has no simple answer to the bubble question and described the balance as finely balanced.
What this means for AI buyers
For companies deploying AI, the distinction between wider usage and higher spending is becoming the point to track. Open-weight models offer a way to expand pilots and departmental tools without committing to the highest-priced frontier systems. That route fits smaller firms that need contained costs and faster deployment across a limited set of tasks. Larger organizations face different pressure because boards are growing more ambitious after initial successes and expect scaling. Their AI budgets may keep rising even as the mix shifts toward cheaper models.
The evidence base calls for caution when interpreting upbeat claims from peers and suppliers. A Boston Consulting Group survey found that about 70 percent of CEOs worldwide say AI success affects how their performance is perceived, which creates an incentive to describe outcomes in favorable terms. Buyers should therefore separate measurable operational or financial gains from general statements about progress. Useful checks include the lifespan assumed for chips, the price path for the models in use, and whether savings or revenue can be tied to specific deployments. Projects that cannot pass that test are unlikely to justify larger commitments.
The marker to follow is whether the number of projects that pass the vacation test starts to grow. If more IT leaders can link deployments to budget increases approved after verified wins, and if lab revenue keeps pace with buildout costs, the business case firms up. If growth concentrates in lower-cost open-weight use while frontier spending stalls, funding pressure will persist.
