Goldman forecasts $1 trillion AI spend

AI spending is turning into an infrastructure story. Not just a software story, and not just a chip story. Goldman Sachs is now framing the build-out in numbers that look more like railroads or telecom booms than a normal tech cycle — with potential AI infrastructure needs in the trillions and hyperscaler capex still climbing fast. ### What did Goldman actually say? Goldman put out two closely related messages in early May. One was a May 1 framework piece arguing that the total AI build-out could land anywhere from $4 trillion to $8 trillion over five years, depending on a few physical assumptions about how the infrastructure gets built and replaced. The other was a fresh strategist note, highlighted May 11, saying analysts now expect U.S. megacap hyperscalers to spend $755 billion on capital expenditures in 2026, up 83% year over year. (goldmansachs.com) ### Why is the range so huge? Because this is not really a forecast in the usual sense. Goldman’s point is that small changes in a few engineering assumptions swing the total by hundreds of billions of dollars. How long AI chips remain economically useful matters. So does how expensive next-generation data centers become as power density rises. And then there are bottlenecks — power, labor, and equipment — that can stretch timelines and change replacement cycles. (goldmansachs.com) ### Why are investors focused on hyperscalers? Because they are the ones writing the checks right now. The Yahoo-cited Goldman note says 2026 capex could reach 100% of cash flow from operations for the megacap cloud group. That is the key pressure point. If operating cash is getting absorbed by data centers, chips, networking gear, and power connections, there is much less room for buybacks unless companies dip into cash piles or add debt. (goldmansachs.com) ### Which companies are pushing the numbers up? The same handful of giants at the center of the AI race. Goldman’s figures, as summarized May 11, showed Microsoft lifting its 2026 capex outlook to $190 billion, Amazon to $200 billion, Alphabet to roughly $180 billion to $190 billion, and Meta to $125 billion to $145 billion. Those are not side projects anymore. They are balance-sheet-defining commitments. (finance.yahoo.com) ### Why do buybacks matter here? Because buybacks were the easy way Big Tech returned surplus cash to shareholders. Goldman’s strategist note says hyperscalers cut buybacks 64% year over year in the first quarter, and now devote 20% of total spending to buybacks and dividends versus a 2017-2022 average of 34%. Basically, AI capex is crowding out the old shareholder-friendly default. (finance.yahoo.com) ### Is this just Goldman getting more bullish? Not exactly. Turns out Goldman has also been saying Wall Street keeps underestimating AI capex. Back in December, its research team said consensus 2026 hyperscaler capex had already risen to $527 billion from $465 billion. By January, Goldman was still saying the biggest cloud companies were likely to spend more than half a trillion dollars in 2026. The newer $755 billion figure shows how fast estimates are still moving. (finance.yahoo.com) ### What is the real economic question? Whether this build-out becomes self-funding before investors lose patience. If AI revenue ramps fast enough, the spending looks rational. If not, the sector starts to look like a utility build-out with tech multiples attached. That is the catch — the physical AI stack now needs chips, cooling, cabling, and power at a scale that can overwhelm even giant cash flows. (goldmansachs.com) ### So what should readers watch next? Watch three things: capex revisions, free cash flow, and power constraints. If spending keeps rising while cash returns stay weak, the market will get pickier about who is actually turning AI infrastructure into revenue. Goldman’s own framing has shifted in that direction already. The story is no longer “AI is big.” It is “who can afford the build-out, and for how long?” (goldmansachs.com) ### Bottom line The headline number is flashy, but the important shift is simpler. AI is becoming a capital-allocation regime. And once that happens, every investor question changes with it. (goldmansachs.com)