Jim Covello counted the artificial intelligence billboards he saw while speeding down Highway 101 from San Jose to San Francisco this month. Of the nearly 40 billboards he passed, including ones advertising things like Writer Enterprise AI and Speech AI, he saw them as further evidence of an economic bubble.
“Not that long ago, all the signs were in code,” Covello said, “and now they’re all AI.”
Covello, head of equity research at Goldman Sachs, has become Wall Street’s leading AI skeptic. Three months ago, he shocked the market by publishing a research paper that cast doubt on whether companies would get a full return on an estimated $1 trillion in AI spending over the next few years. He said generative AI, which can summarize text and write software code, makes so many mistakes that he doubts it can reliably solve complex problems.
Goldman’s paper came days after partners at venture firm Sequoia Capital raised similar questions in a blog post about AI. Their skepticism marked a turning point for AI stocks, leading to a rerating of some of Wall Street’s hottest trades.
Goldman’s basket of AI stocks, managed by a separate unit of the firm and including Nvidia, Microsoft, Apple, Alphabet, Amazon, Meta and Oracle, has fallen 7 percent from a peak on July 10 as investors and business leaders debate whether AI justifies its huge costs.
This pause comes at an early stage in the AI arms race: The tech industry has a history of investing heavily in enabling technological transformations, such as the personal computer and internet revolutions, that took more than five years of buildup to produce results.
But Mr. Covello, 51, has seen the boom and bust of the tech industry. As a semiconductor analyst, he witnessed the dot-com bubble burst and was heartbroken to see colleagues lose their jobs. More recently, the Goldman veteran joined an in-house team that evaluated AI services for the firm to use. He said the services he evaluated were costly, cumbersome and “not smart enough to make our people smarter.”
Some say Mr. Covello’s warning is premature, given the industry’s history. Shortly after Goldman’s paper was published, George Lee, co-head of the firm’s geopolitical advisory practice, pushed back against Mr. Covello in an email, saying AI would save workers time and make them more productive. Mr. Lee urged Mr. Covello to be patient.
“The long-term impact of the platform transition is that applications will emerge over time as the technology becomes more sophisticated, more readily available and cheaper,” Lee said in an interview about the email.
Goldman’s clients asked to hear more, and in response, the firm began hosting private bull-bear roundtables in which Lee, as the bull, would present his optimistic view of AI, and Covello, as the bear, would present his pessimistic view.
The conversation was long overdue, said Jim Morrow, CEO of Boston-based Calodine Group, a Goldman client. “AI had captured the market zeitgeist,” he said. “Having people at firms like Goldman sounding the alarm and saying, ‘This isn’t going to happen the way everyone thinks it will,’ made people ask important questions about what was actually going on.”
Covello was a born skeptic. Before he enrolled at Georgetown University, the first in his Philadelphia family to attend college, his father questioned whether a four-year degree would be worth the expense. He went on to play first base for the school’s baseball team, and brought that same skepticism to the plate, batting .270 with a stellar career.
He joined Goldman Sachs as a technology analyst in 2000. That summer, the firm gathered in a Napa Valley vineyard for a passionate internal meeting to discuss the tech industry. But the internet boom, already at its peak, began to crash within the next few months.
A few companies, like Google and Amazon, survived and became incredibly wealthy, but Covello stuck to the devastation. “It was a very scary time,” he says. “I didn’t know if I’d still have a job.”
Mr. Covello kept his job. At the time, Goldman was cutting costs by replacing experienced analysts with younger staff. The firm promoted him to lead semiconductor analyst in 2001 and to head of global equity research in 2021.
After ChatGPT was released in 2022, the tech industry began comparing the arrival of AI to the dawn of the public internet, a comparison that caught Covello’s attention. “That’s not something anyone should be rooting for,” he said, recalling the millions of jobs that have been lost.
Experts predicted that launching an AI business would require spending $1 trillion on data centers, utilities and applications — costs that Covello believed would prevent the industry from cheaply solving real-world problems like internet companies did decades ago.
As a member of Goldman’s working group on AI, he reviewed a service that used generative AI to automatically update analysts’ spreadsheets with companies’ financial results, which he said saved analysts about 20 minutes of their time per company but came at six times the cost.
Mr. Covello’s skepticism was widespread within the company. Alison Nathan, editor of Top of Mind, a monthly investigative report, was planning a feature on AI and, at the urging of a colleague, she met with Mr. Covello.
“For about 35 minutes, I was glued to his story and his opinions,” she said.
Nathan decided to interview Covello for the report, and that conversation helped define the 31-page report’s title: “Gen AI: Too Much Spending, Too Little Benefit?”
Covello disputed the idea that AI costs would fall, pointing out that the costs of some advanced technologies, such as the machines that make semiconductors, are rising, and he also criticized AI’s capabilities.
“When you overbuild things that the world doesn’t need or isn’t ready for, it usually ends badly,” he said.
It was one of the most widely read reports in the publication’s 12-year history.
At Goldman’s annual tech conference in San Francisco this month, Covello and Li presented their competing views on AI to a crowd of several hundred people. Covello cited a Business Insider article about a pharmaceutical company canceling a contract with Microsoft because it deemed the company’s AI services “good for a middle school presentation” and highlighted the technology’s shortcomings.
Lee shook his head. He pointed to a Princeton University paper that found AI helped improve the productivity of 5,000 developers across 100 companies by 20 percent.
“It’s not perfect,” Lee said, but added that “people are getting some productivity savings.”
Some in the audience questioned whether Goldman was playing it safe by spotlighting its own AI pessimists at a conference featuring AI evangelists like Jensen Huang, CEO of Nvidia, the world’s leading AI chipmaker.But many found the discussion constructive.
“This is the latest version of the question: ‘If you build it, will they come?'” said David Liederman, a portfolio manager at Endurance Capital Partners.
Covello predicts the AI boom will lose steam as companies adopting the technology cut spending in the face of declining profits. He doesn’t think it will trigger another dot-com bust, but he reevaluates his position daily.
“If you have some kind of outlandish view, you’re always going to have this kind of delusional state that AI is going to be as big as everyone thinks it is,” he says, “so I really look every day at my blind spots. What could I be doing wrong?”
