Sam Altman Says AI Has Explained Itself Badly. He's Right.

Sam Altman told Bloomberg Television this month that the AI industry has done a bad job of explaining what the technology is for. "The industry has done a terrible job of this on the whole," he said. "I think we have done a bad job ourselves, maybe better than some others, worse than some others."

A week earlier he had told Time something blunter: "Clearly, people hate data centers — right now, at least."

It is rare to hear the person at the centre of the hype say it out loud. He's right, and that failure has cost the industry.

The backlash has a number on it

In the first three months of 2026, local opposition blocked or delayed at least 75 data centre projects in the US, worth about US$130 billion. That comes from Data Center Watch, which has tracked these fights since 2023. It is the most it has recorded in a single quarter. Over the same stretch, the number of active local opposition groups more than doubled, from 396 to 833.

It is not a noisy minority either. A Gallup poll in May found seven in ten Americans oppose a data centre being built in their area.

That is what a communication failure looks like when it reaches a council meeting. People were told AI would change everything. Nobody told them what it would do for them. What they could see was a very large building, a very large power bill and not many local jobs.

In Europe it is not a story problem

Europe has a different constraint, and better messaging will not fix it.

In March, Denmark's grid operator, Energinet, paused new grid connection agreements. Around 60 gigawatts of projects were waiting to connect. Denmark's peak electricity demand is around 7 gigawatts. That is nearly nine times what the whole system carries at its busiest, queued up and asking to be plugged in. Data centres made up about 14GW of it.

No amount of storytelling shifts that ratio. You can win a community over. You cannot win over a transformer.

Everyone is running on FOMO

Meanwhile the decisions are being made in a hurry. Countries worry they cannot compete without it. Boards worry their competitors have already started. Plenty of the spending is a fear of missing out, dressed up as a strategy.

The smarter move is to sit back and ask a boring question: what would we actually get?

Not what the keynote promised. What would show up in your week, in your team's hours, in your numbers.

For most businesses I work with, the honest answer comes down to two things.

One: the say-it-again work

This is the work of explaining the same thing, over and over, to different people.

The customer question you have answered two hundred times. The onboarding email. The quote that is 80% the same as the last one. The status update the client asked for because they could not find the last status update. The proposal where only the name, the numbers and one paragraph change.

None of this is hard. That is the point. It is repeated, it follows a pattern you already know, and it eats hours because it arrives in small pieces all day. AI is genuinely good at it, because you can show it what "right" looks like once and it can say it again, in your voice, for the next person.

When I wrote about where AI actually took cost out of Biz365, I called this translation waste, and it was where most of the money was. It is also why my virtual assistant's job slowly moved across. None of it was dramatic.

Two: the check-it work

This is the work of checking something against rules you already have.

Is the form complete? Does the invoice match the purchase order? Has every required field in the compliance record been filled in? Does this timesheet fit the roster? Has the document got the clauses it is supposed to have?

People do this checking because it has to be done, not because they are good at it. They are tired by the fortieth one, and the error they miss is the one that costs money. AI can do the first pass against a clear set of rules and flag the exceptions, so a person spends their attention on the three that look wrong rather than the forty that are fine.

In our own business this was verification waste, and it is the one place where the quality went up as well as the cost coming down. That is a narrow claim, and I mean it to be. I have also written that AI takes away the thinking and hands you back the checking. Both are true. The fix is to design the checking step on purpose, and an agent doing the first pass is a good part of that design.

What you will not get

You will not get the massive efficiencies from the hype end. You will not get a PhD-level researcher who runs your strategy while you sleep.

You will get real efficiencies. Hours back every week, on work that was never the reason anyone joined your business. It is less exciting than the pitch. It is also real, and it shows up in the numbers.

The maths is not your problem

The maths behind the hype is coming undone. Enormous sums have been committed to AI, and the value realised is nowhere near it. MIT's research found that 95% of enterprise generative AI pilots produced no measurable profit and loss impact. The data centre fights are the physical version of the same gap: huge spending that the people around it cannot see the point of.

But that is not your disconnect to fix. It belongs to the companies that made the bets.

Here is what that gap means for you. While the money is being spent, the tools keep getting better, the vendors keep competing hard for your business, and the capability on offer is more than most businesses will ever use. You have a window to take the savings in your own repeated work, built on infrastructure someone else paid for.

That window does not need you to believe the hype. It needs you to find the say-it-again work and the check-it work in your own business, and fix them one at a time.

If you want help finding where that work is hiding in your business, come and have a chat with us, or see what this looks like running in production. Or get this kind of thinking weekly.