Remember when everyone was convinced Big Tech’s AI spending would eventually hit a wall? Yeah, about that. Goldman Sachs just threw a wrench into that narrative, and it’s a pretty hefty one.
The bank’s latest analysis suggests that Wall Street has been massively underestimating how much hyperscalers—think Google, Amazon, Microsoft—will actually spend on AI infrastructure. We’re talking roughly $1.1 trillion in 2027, compared to the $920 billion Wall Street is currently expecting. In a bullish scenario? Try $1.4 trillion. That’s not a rounding error; that’s a whole different ballgame.
Here’s the kicker: Goldman’s analysts believe the demand for AI computing power is still in its infancy. They’re forecasting that token consumption will explode 24 times by 2030, largely driven by enterprise agents—basically AI systems that can actually get stuff done for businesses. More tokens mean more computing power needed, which means more data centers, more chips, more networking equipment, and more power infrastructure. It’s a beautiful, expensive domino effect.
But wait, there’s more. Higher input costs are also pushing up the nominal dollars required to support all this token consumption. So even if the volume doesn’t skyrocket as much as expected, the price tag still does.
Now, here’s where it gets spicy. Several companies have recently started sweating about token expenses tied to their AI tools, raising the uncomfortable question: Will the productivity gains actually justify the astronomical costs? It’s the kind of question that keeps CFOs up at night.
The real evidence supporting Goldman’s bullish take? Cloud providers themselves. Google Cloud and Amazon Web Services reported a combined backlog of $832 billion as of Q1—up from $358 billion just six months earlier. That’s not speculation; that’s actual money already committed.
Goldman doesn’t expect AI supply and demand to balance out until at least the second half of 2027, meaning this spending spree could stay elevated way longer than most investors think. Historically speaking, AI-related investment hit about 1.5% of GDP in 2026. Compare that to the railroad, electrification, and automobile booms, which peaked at 2-3% of GDP. We might not be done yet.
The real constraint? It’s not money—it’s physical infrastructure. Data center projects are delayed, and memory, power, and labor are all bottlenecks. That’s actually good news for companies in the AI supply chain: semiconductors, networking, cooling, and power suppliers should keep seeing earnings growth.
The catch? Valuations for AI infrastructure stocks have gotten crowded and expensive. Share prices are outpacing earnings revisions, which means volatility could be lurking around the corner. And here’s the uncomfortable truth: While 54% of companies mentioned AI productivity during earnings calls, only 11% actually quantified the benefits, and just 2% showed an impact on earnings.
So yeah, the AI boom might be bigger than we thought. But whether it’s actually profitable? That’s still the $1.4 trillion question.
