Businesses could spend more than $1 trillion to develop generative artificial intelligence in the coming years, but a possible big payoff is decades away, if it comes at all, declared a noted professor at Massachusetts Institute of Technology.

“Given the focus and architecture of generative AI technology today... truly transformative changes won’t happen quickly and few—if any—will likely occur within the next 10 years,” MIT's Daron Acemoglu said in Goldman Sachs’ Top of Mind report, entitled “Gen AI: Too Much Spend, Too Little Benefit?”

Indeed, the heat wave that smothered much of the East Coast has recently abated. But the debate over the financial promise of generative AI still rages on Wall Street.

And again the arguments are sharply drawn: Love it or hate it. JP Morgan CEO Jamie Dimon compares AI to the printing press and electricity, while Berkshire Hathaway CEO Warren Buffett likens it to the development of nuclear weapons.

In its June report, Goldman presented arguments for both sides of the debate. While Acemoglu believes truly transformative changes from AI will be limited, three Goldman analysts are “more optimistic about AI’s economic potential and its ability to ultimately generate returns beyond the current ‘picks and shovels’ phase, even if AI’s ‘killer application’ has yet to emerge," the report said.

“Spending is certainly high today in absolute dollar terms. But this capex cycle seems more promising than even previous capex cycles,” argued Kash Rangan, U.S. software equity research analyst at Goldman.

The report also cited several barriers to AI growth, noting that demand for chips could outstrip supply and utilities might not be prepared to handle looming power surges required of chip development

Goldman, itself, though, takes a measured view of AI as an investment opportunity. Beyond reports of efficiency gains among developers, there’s little to show for the heavy spending so far on AI, the firm said. But despite the concerns and constraints, “we still see room for the AI theme to run, either because AI starts to deliver on its promise, or because bubbles take a long time to burst,” Goldman stated.

AI’s Limited Impact on Productivity
Acemoglu, whose latest book is “Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity,” argued that the upside to U.S. productivity and growth from generative AI in the next 10 years and beyond will likely be more limited than many expect. The professor allowed that the technology has the potential to fundamentally change the process of scientific discovery, research and development as well as create new products and platforms. But he said such fundamental changes won’t happen quickly, given the focus and architecture of generative AI technology today.

Acemoglu estimated that only a quarter of tasks exposed to AI will be cost-effective to automate within the next decade, suggesting that AI will impact less than 5% of all tasks. He’s also skeptical that AI adoption will create new tasks and products.

AI, the professor forecasted, will increase U.S. productivity by only 0.5% and GDP growth by only 0.9% cumulatively over the next 10 years.

“The largest impacts of the technology in the coming years will most likely revolve around pure mental tasks, which are non-trivial in number and size but not huge, either,” Acemoglu said.

The professor was joined in the skeptics’ camp by Jim Covello, Goldman’s head of global equity research. “AI technology is exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn’t designed to do,” Covello said in the report.

On the other hand, the Internet, even in its infancy, enabled low-cost solutions to disrupt high-cost solutions. That’s a “truly life-changing invention,” the analyst argued in the report.

Some Goldman Analysts See AI Payoff
In arguing for the long-term transformative and returns possibilities of generative AI, Rangan believes the potential returns from this capital expenditure cycle seem more promising than previous cycles. This cycle, he explained, is being led by incumbents with low costs of capital and massive distribution networks and customer bases. The management of these companies are also “very capable,” he said, adding that these factors lower the risks that the technology doesn’t become mainstream.

Unlike the current incumbents, the companies that led the late-1990s investment cycle “didn’t have the financing, reputation, or knowledge to succeed, resulting in a tremendous amount of underutilized capacity,” Rangan said in the report.

Another Goldman bull, Internet equity research analyst Eric Sheridan, also argued against irrational exuberance in AI spending, pointing out that current capex as a share of revenues doesn’t look markedly different from prior tech investment cycles. Furthermore, investors are rewarding only those companies that can tie a dollar of AI spending back to revenues.

“While I would never say I’m not concerned about the possibility of no payback, I’m not particularly worried about it today,” Sheridan said, “though I could become more concerned if scaled consumer applications don’t emerge over the next six to 18 months.”