Goldman Sachs Spots Cracks in the AI Spending Boom as Returns Stay Elusive

Wall Street has poured hundreds of billions into artificial intelligence infrastructure. The numbers stun. U.S. investment alone heads toward $600 billion next year. That sum equals nearly 2% of GDP. Yet evidence that the outlays deliver matching economic gains remains thin. Goldman Sachs analysts laid out the tension in fresh research published just yesterday.

Jessica Rindels, a Goldman Sachs analyst, put the figure at almost $600 billion for 2026. She noted it has already topped 10% of business fixed investment in recent quarters. The scale invites hard questions. Companies redirect capital. Construction crews shift to data centers. Financing flows one way. But does any of it pay off?

The bank examined whether hyperscalers sacrifice other projects to bankroll their AI push. Answer? Not really. These giants cut share buybacks instead. They borrow freely. High interest rates fail to slow them. Strong cash flows and open capital markets keep the machine running. So the AI surge avoids deep cuts elsewhere inside those balance sheets. The Yahoo Finance report captured the nuance.

Buyers of AI services tell another story. Goldman Sachs surveyed them. Costs stay modest for most. Still, roughly two-thirds of that spending comes from trimming other budgets. Direct trade-offs appear. The sums, however, look small next to the infrastructure binge. Little sign of broad displacement surfaces yet. The analysis offers cautious reassurance even as it flags risks.

James Covello sees deeper trouble. As Goldman Sachs’ head of global equity research he has questioned the boom for years. In updates this spring he conceded some errors. The core doubt sharpened. Returns stay elusive. He pointed to $30 billion to $40 billion already spent on enterprise generative AI. Results? A MIT study cited by Fortune found 95% of company pilots yield zero return. An EY survey added that 99% of sampled firms logged financial losses averaging $4.4 million each.

Workers notice the gap too. A Wall Street Journal survey revealed executives praise efficiency gains while rank-and-file employees report scant change. One AI hiring startup tested frontier agents on 480 routine tasks performed by bankers, consultants and lawyers. Every system failed most assignments. The disconnect grows.

Costs climb instead of fall. Gartner projects worldwide IT spending will hit $6.15 trillion in 2026, up from $5 trillion in 2024. Harvard Business Review researchers tracked “workslop,” the errors AI introduces. A 10,000-person organization loses more than $9 million a year in wasted effort. New headaches replace old ones. Productivity metrics refuse to budge in line with the hype.

Hyperscalers keep writing big checks. Microsoft, Meta, Google and Amazon line up over $300 billion in combined capital expenditure for 2025, according to earlier tallies. Projections for 2026 push higher. Meta alone may reach $135 billion. Debt now finances part of the tab. Goldman Sachs strategists expect credit to cover 35% of hyperscaler outlays by 2027. Equity markets once carried the load. Credit markets join the frenzy.

Broader economic impact looks muted. Goldman Sachs Research once forecast AI would lift U.S. growth measurably by 2027. That timeline slips. Recent calculations show the massive investment added basically zero to 2025 economic expansion. Barclays offered a slightly brighter view. Even after adjustments, AI spending lifted output by an annualized 0.8% in the first half of last year. GDP grew faster. The contribution still falls short of promises.

Power demand surges alongside chips. Data centers devour electricity. Some hyperscalers pursue behind-the-meter generation to bypass grid constraints. The choice affects timing more than total capital required. Training still dominates spending. A faster shift to inference could speed revenue and improve utilization. Yet the infrastructure bill barely shrinks. Assumptions about returns hinge on a handful of variables few fully grasp.

Comparisons to past cycles feel inevitable. The dot-com era saw similar exuberance followed by brutal reckoning. Today’s build-out differs in speed and concentration. A few mega-cap names shoulder most expense. Their valuations reflect sky-high expectations. Microsoft joined Nvidia in the $4 trillion market-cap club last year. Meta hovers near $2 trillion. Any slowdown in perceived progress hits hard.

Enterprise adoption accelerates on paper. Surveys show rising use. Actual bottom-line proof lags. Only a small share of firms can tie AI to specific profit improvement. J.P. Morgan Asset Management estimates AI could add 1.4% to 2.7% annual growth over the next decade. That gain would matter. Delivery remains uncertain. Demand for compute outruns supply for now. The imbalance props up suppliers but cannot last forever.

Goldman Sachs itself leans into the technology. The firm rolls out its own AI assistant to boost analyst productivity. It targets fee growth, efficiency and a better operating ratio. Management expresses long-term confidence. Near-term expenses stay elevated. The bet mirrors clients’ dilemmas. Spend now. Pray for payoff later.

Emerging markets watch closely. Goldman Sachs economists see limited job displacement in India. Forty percent of the workforce sits in construction and retail, areas less exposed. Service roles face more pressure. Productivity could rise 0.4 percentage point over a decade. Wage gains may follow for users of the tools. Whether the pattern repeats elsewhere stays unclear.

Investors sift the signals. Some Wall Street outlooks for 2026 treat AI as a powerful growth engine. Others flag bubble risks. BCA Research stays neutral on stocks despite recession fears, crediting the capital-expenditure tailwind. NatWest calls it an unusual propellant at this business-cycle stage. Divergence widens across the value chain. Chip makers thrive. Downstream software faces tougher scrutiny.

The crowding-out question refuses to vanish. Imported equipment makes up a large slice of U.S. AI spend. Dollars flow abroad. Domestic multipliers shrink. Construction resources grow scarce in key regions. Financing competes with other corporate needs. So far the effect stays contained. Hyperscalers’ financial flexibility explains much of that resilience. Smaller buyers enjoy no such luxury.

Depreciation looms. Goldman Sachs analysts warned that rising expenses will erode some return-on-equity gains for mega-cap tech in coming years. The S&P 500 could feel ripples. Upside exists too. Higher asset turnover and margins from genuine productivity advances would lift broader equities. The path from capex to those outcomes stretches longer than many models assume.

Power, talent and data remain bottlenecks. No easy fixes appear. Regulators eye energy consumption and concentration risks. Executives juggle shareholder demands for returns against fear of falling behind. FOMO, Goldman Sachs noted, often outweighs poor near-term performance as a decision driver. The insecurity fuels the cycle.

Analysts keep refining assumptions. Small changes in utilization rates or inference mix swing projected returns dramatically. Most debate centers on timing and distribution rather than absolute capital scale. The trillions still must be spent. Who captures value afterward matters more than ever.

History offers no guarantee. Past technologies eventually delivered. The wait tested patience. Today’s AI wave arrives with higher stakes and steeper valuations. Goldman Sachs’ latest observations inject measured skepticism at a moment when optimism dominates boardrooms. The data say wait. Markets say bet bigger. The tension defines the era.