Most IT Leaders See AI Wins. Few Would Wake the CEO

Two-thirds of a room full of IT leaders say they can point to measurable AI results. Fewer than one in twenty say those results are big enough to interrupt the boss’s vacation. That gap sits right at the center of the AI bubble debate, The Decoder reports, and it tells us more about where enterprise AI is headed than most earnings calls do.

The story comes from Azeem Azhar, the British tech entrepreneur behind the research group Exponential View. He told it on a podcast with Nicholas Thompson of The Atlantic.

🧍 The Standing Test

Azhar was speaking to about 160 IT vice presidents in Las Vegas. He asked everyone who could point to measurable AI results to stand up. Two-thirds stayed on their feet, which was more than he expected.

Then he raised the bar. Who had results good enough to interrupt the CEO’s summer vacation? About eight people stayed standing.

It’s a room poll, not a study. Still, it captures the problem well. AI is producing value inside companies, but mostly in small increments. It isn’t transforming them yet. According to The Decoder, return on investment still can’t be documented across the whole economy. It shows up only in anecdotes like this one.

💰 Why This Matters for the Buildout

The bubble question comes down to one thing: is revenue at AI labs growing fast enough to pay for the huge data center buildout? A lot of smaller questions depend on the answer:

• Chip lifespan: Can GPUs stay useful in production for four, six, or eight years? Each option changes the depreciation math a lot.
• Enterprise value: Are companies getting enough real value to keep buying more AI, ideally at higher prices?
• Where the money goes: Even if usage grows, does the revenue flow back to the companies that paid for the infrastructure?

The third point is the one to watch. Azhar notes that many companies are moving from expensive frontier models to cheaper open-weight alternatives. That’s rational for any single company. But if it happens everywhere, you could get what he calls the “bear version of the story”: AI adoption keeps rising while the bubble still bursts, because too little money reaches the labs and cloud providers.

📈 The Case for Optimism

Azhar doesn’t only see risk. Some executives told him their boards are getting more ambitious after early wins. Even in slower markets like Italy, CEOs said trust is building and budgets are rising, despite mistakes along the way.

That pattern matters. Big enterprise technology shifts rarely start with dramatic results. Cloud computing and ERP systems both went through years of small, unglamorous wins before the big payoffs showed up.

⚠️ Discount the Self-Reporting

There’s a catch. The Decoder cites a Boston Consulting Group survey that found about 70 percent of CEOs worldwide say AI success affects how people see them in their roles. That gives leaders a strong reason to describe their AI results more generously than the facts support.

So even the two-thirds figure deserves some skepticism. Some of those people may have stood up because sitting down looked bad.

Azhar’s own conclusion is honest. He says he doesn’t have a simple answer and calls the situation “finely balanced.” He’s a credible voice here. Exponential View has tracked the economics of technology adoption for years, and he tends to lay out both the bull and bear case instead of picking a side.

🔮 What the Next 1-3 Years Look Like

My read: over the next two years, the market will shift from asking “do you have AI results?” to asking “are your results big enough to matter?” Boards will want proof of real business impact, not a list of pilots. Pricing pressure from open-weight models will also keep pushing enterprise AI costs down, which is good for buyers and tough for labs.

Here’s how practitioners and businesses can get ready:

1. Measure business outcomes, not activity. “Hours saved” won’t impress anyone for long. Tie AI work to revenue, margin, or cycle time.
2. Hunt for one result worth a vacation call. Ten small wins are fine. One big one changes the budget conversation.
3. Stay flexible on models. Build so you can swap frontier and open-weight models as prices and capabilities change.
4. Report honestly. Inflated AI claims catch up with you once the CFO starts asking for the numbers behind them.

The companies in that final group of eight will set the pace for everyone else. Whether enough others join them soon enough to justify the buildout is still the most important open question in AI. The full discussion is available at The Decoder.