• MIT economist warns AI infrastructure investments may not meet investor profit expectations.
• Only 5% of jobs are likely to be significantly impacted by AI in the next decade, according to the economist.
• "A lot of money is going to get wasted," MIT's Daron Acemoglu told Bloomberg.

A bulk of the hundreds of billions of dollars being poured into artificial intelligence infrastructure investments could go up in smoke.

That's according to MIT Economist Daron Acemoglu, who told Bloomberg in an interview that the hype surrounding AI may not meet its lofty expectations.

"A lot of money is going to get wasted," Acemoglu said.

Acemoglu estimates that only 5% of jobs are ripe for being taken over or heavily assisted by AI technologies over the next decade.

That suggests the estimated economic benefits of massive efficiencies and productivity gains unlocked via AI technology may not come to fruition, at least not for a very long time.

"You're not going to get an economic revolution out of that 5%," Acemoglu said.

One of the big concerns is that cloud hyperscalers like Microsoft, Amazon, and Meta Platforms' massive investments in Nvidia's AI-enabled GPUs will not lead to a corresponding revenue surge.

And that could lead to a sudden cool down in the AI story if investors start to scrutinize profit margins and the expected payoff time of such investments.

Acemoglu ultimately sees three potential scenarios for the AI story, and none are particularly bullish.

Acemoglue sees a combination of the second and third scenarios as the most likely.

"When the hype gets intensified, the fall is unlikely to be soft," Acemoglue said.

While the MIT economist is impressed by what large language models like ChatGPT can do, reliability issues means they won't replace humans in the work place for a very long time.

"You need highly reliable information or the ability of these models to faithfully implement certain steps that previously workers were doing," Acemoglu said.

He added: "They can do that in a few places with some human supervisory oversight" like coding, "but in most places they cannot. That's a reality check for where we are right now."