Jim Covello saw the warning signs early. As head of global equity research at Goldman Sachs, he questioned the flood of money pouring into artificial intelligence years before most on Wall Street voiced similar doubts. His 2024 report laid out a simple challenge. Would the coming trillion dollars in spending deliver returns that justified the cost?
Two years later the question has grown sharper. Not softer. Covello appeared on Goldman Sachs’ Exchanges podcast in late May 2026. His tone carried fresh urgency. “In a lot of ways, companies are losing more money today implementing this technology than they were two years ago,” he said. “The hill that has to get climbed is even steeper today than it was before, because we’ve spent more money.”
The economics look more questionable now. Enterprise buyers show scant returns. Model developers burn cash. Hyperscalers keep writing big checks without clear payoff. Value has concentrated in semiconductor makers. Nvidia and its peers have captured nearly all the profit created so far. That pattern stands out. Covello spent 16 years covering chip stocks. He knows the usual cycle. Suppliers thrive when their customers thrive. This time the customers bleed while suppliers boom. “That can’t go on forever,” he noted.
So Covello now recommends investors favor the big spenders over the chipmakers. Long hyperscalers. Underweight semis. In two of three likely outcomes the big cloud and social media operators come out ahead. Only if the current imbalance persists indefinitely do chip stocks keep winning. Few expect that status quo to hold.
Fortune reported on the podcast the same day it dropped. The piece captured Covello’s blunt warning. “At some point you’ve got to make money.” He added that companies make investments to generate returns. “And we’ve gotten further away from that over the last couple years instead of closer to it.”
Consumer adoption surprised him. It proved magnificent. Far beyond expectations. The underlying models advanced quickly too. Yet those bright spots do not erase the central problem. Enterprises must make or save real money. If they do, the technology fulfills its promise. If not, the debate grows uncomfortable. “If we’re having the same debate two years from now and we’re still saying, ‘Well, it’s early,’ then we might have a challenge,” Covello said. “Because at some point, when does the short-term become the long-term?”
Data backs his caution. MIT researchers found 95 percent of organizations reported zero return on AI pilots. An EY survey showed 99 percent of companies faced financial losses tied to AI risks. Those losses averaged $4.4 million each. Surveys from Cognizant reveal 93 percent of jobs already disrupted by the technology. Six years faster than once projected. Productivity gains that should follow have not appeared. Executives call it an activation gap.
Fear of missing out fuels the spending. FOMO runs through every layer of the supply chain. Hyperscalers raised capital expenditures even when their stocks lagged the market. Covello once predicted sustained underperformance would force discipline. It did not. “There’s a tremendous amount of FOMO at every level of the supply chain,” he observed. That insecurity, not optimism alone, keeps the money flowing. And insecurity often signals bubbles.
Inside companies the picture looks messier. C-suite leaders express enthusiasm. Line workers see less benefit. Data readiness lags. Models and agents perform well in tests yet encounter corporate information that is siloed, inconsistent, or unprepared for automation. Legacy systems add drag. Incumbents face structural hurdles that AI-native startups avoid. The result? Slower realization of gains. Higher costs. Persistent losses.
George Lee, co-head of the Goldman Sachs Global Institute, takes a more optimistic long-term stance. Even he sees the math as daunting. He estimates total infrastructure spending could reach $7 trillion to $8 trillion. Simply rearranging existing profit pools will not suffice. Net new economic activity must emerge. Otherwise the numbers never work.
Political headwinds gather too. Populist resentment toward AI has grown, especially in the United States. Commencement speakers face boos when they mention the technology. Data centers consume more than 4 percent of national electricity. Projections show that share could triple by 2028. Local communities push back against rising power costs. President Trump brought hyperscalers to the White House in March 2026 to sign a voluntary Ratepayer Protection Pledge. The gesture highlighted how fast the issue escalated.
Covello folds these pressures into the broader economic case. “I would put all of it in the broader economics bucket,” he said. “How much of it is going to flow back to the individual? Can we make a case that individuals are benefiting economically from using the technology?” The answer so far feels incomplete.
Ray Dalio offered his own perspective days earlier. The billionaire investor told Bloomberg Television on June 3, 2026, that bubbles accompany all great technological changes. “All great technology changes produce bubbles,” he said. “Nobody can get it exactly right. You have to either spend a ton of money to capture your market share and don’t worry about whether it’s too much or not, or you don’t spend enough money and you lose your market share.” His comments arrived as OpenAI and Anthropic near potential IPOs valued near $1 trillion. Neither company turns a profit.
The bull market itself owes partial credit to AI optimism. That creates a circular dynamic. Strong stock prices support continued spending. Continued spending feeds the narrative. Yet Covello sees limits. The market grants a long leash for now. That leash is not infinite. Three clocks tick at once. Economic. Operational. Political. Stakes rise with each passing quarter.
None of this means the technology fails forever. Covello has never claimed as much. “That doesn’t mean it’s never going to happen,” he said. “It just means the stakes are higher.” The distinction matters. Progress in the lab does not guarantee profit in the enterprise. Breakthroughs in capability do not automatically translate into balance-sheet wins.
History offers parallels. Overbuilding what the world does not yet need often ends badly. Dot-com investors learned that lesson the hard way. NASDAQ fell around 70 percent from peak to trough. Covello witnessed those cycles. His semiconductor coverage spanned multiple booms and busts. The current setup carries echoes. Massive capital intensity. Concentrated winners. Questions about sustainable returns.
Goldman itself produced an October 2025 note arguing the market was not in a bubble yet. Fundamentals drove tech gains more than pure speculation. Incumbents dominated AI so far, unlike past bubbles that drew floods of new entrants. Those differences provided comfort. They have not silenced the skeptic inside the firm.
Recent coverage sharpens the focus. Bloomberg highlighted Dalio’s bubble warning just before Covello’s latest remarks. Other outlets have revisited the trillion-dollar question with fresh data on electricity demand and local opposition. The conversation has shifted. From whether AI will transform society to whether the current investment pace can continue without proof of payback.
Executives at industry events echo the tension. Cisco’s chief people officer described teams that lean heavily on AI watching trust erode after nine months. Fear drives decisions more than measured analysis. One chief brand officer at a consumer company put it plainly. FOMO begins with fear. Fear rarely produces the clearest thinking about customer needs.
Yale professor Jeffrey Sonnenfeld examined related bottlenecks last month. Data infrastructure, not model sophistication, will decide whether agentic systems scale. Most large companies underestimate the work required to make their information ready for autonomous agents. Silos, poor governance, and pre-AI architecture create friction that models alone cannot overcome.
Covello’s persistence stands out because of his position. Goldman advises many of the hyperscalers. It underwrites offerings for chip companies. The firm sits at the center of the financing apparatus that powers the boom. A prominent skeptic inside that apparatus commands attention. His views have not softened with time or technological progress. If anything they have hardened.
The coming months will test the thesis. Major AI companies approach public markets. Investors will demand clearer paths to profit. Enterprises will face another round of budget reviews. Power grids will strain under added load. Political rhetoric around job displacement and energy costs will intensify.
Short-term has lasted longer than many expected. The question Covello keeps returning to grows louder. When does it become the long term? Companies cannot lose money on these implementations indefinitely. At some point the bill comes due. When it does, the winners and losers will sort themselves with brutal clarity. The semiconductor party may quiet. Hyperscalers may finally see returns. Or the entire spending wave could crest and retreat. No one knows the precise timing. But the debate has moved past speculation. It now centers on measurable economics. And those economics, two years deeper into the bet, look tougher than before.
WebProNews is an iEntry Publication