Hyperscalers Now Plan to Spend $1.4 Trillion on AI by 2028 — That Number Should Give You Pause

Every quarter for the past two years, the projected bill for AI infrastructure has gone up, not down. The latest revision, out this week from a major Wall Street research shop, now pegs combined 2028 capital spending by the five biggest hyperscale computing companies at roughly $1.4 trillion — up from $1.2 trillion just a few months ago. The justification isn't more ambition. It's that building the stuff got more expensive.

The cost per gigawatt of data center capacity is rising, driven by pricier memory chips and the specialized power infrastructure — what the industry calls "powered shells" — needed to run them. Timelines from groundbreaking to operational data center now stretch up to three years, partly because of supply bottlenecks and partly because of a less obvious factor: local political pushback. Communities are increasingly wary of data centers landing in their backyards, consuming water and electricity and delivering relatively few permanent jobs. That friction is pushing developers to rush project starts and lean harder on the jobs pitch, which tends to raise costs rather than lower them.

This matters well beyond the technology sector. These five companies' capital spending has become one of the single largest swing factors in US GDP calculations. When a handful of firms commit sums approaching the size of a G7 government budget to physical infrastructure, the multiplier effects — construction employment, electrical equipment manufacturing, industrial real estate, and yes, local utility rate structures — ripple into the broader economy in ways that ordinary corporate investment doesn't. It's also why AI capex has become a genuine input into inflation forecasts: demand for electricity, memory chips, and skilled construction labor competes directly with demand from every other sector of the economy.

There's a version of this story that's straightforwardly bullish: strong balance sheets funding productivity-enhancing infrastructure is how economies grow. There's another version that's more cautionary: an industry spending at a pace where cost overruns are being absorbed rather than fought, with revenue models still being worked out in real time, is not obviously disciplined capital allocation. Both things can be true. The point for readers isn't to pick a side in a stock debate — that's not this column's job — it's to understand that decisions being made in boardrooms in Menlo Park and Redmond are now macro variables that show up in your region's electricity rates, in the inflation prints that shape mortgage pricing, and in the local labor market if a data center ever gets sited nearby.

Long Island doesn't have a hyperscale data center campus, and probably won't — land and power costs here work against it, which is itself instructive. But Long Islanders are exposed to this story anyway, through PSEG Long Island's regional grid interconnections, through 401(k)s and pension funds heavily weighted toward the very companies making these bets, and through a national inflation picture that the Fed uses to set the rate that determines your mortgage refinance math. When a forecast for corporate capital spending moves by $200 billion in a single revision, it's worth asking not whether the stocks go up, but whether the underlying bet — that AI compute demand will justify this spending before the money runs out — is being tested by anything other than continued optimism.

The honest answer right now is that nobody knows, including the people writing the checks. What we do know is that the scale of this buildout means its outcome, whichever way it goes, will be felt by people who never bought a share of any of these companies.