During earnings calls this week, tech firms raised their capital expenditure, or capex, projections. Google’s parent company, Alphabet, said on Wednesday it plans to double capex in 2026 to nearly $185 billion. Amazon said Thursday it plans to devote a towering $200 billion to capex, well ahead of Wall Street estimates. Last week, Meta said full-year capex will rise to as much as $135 billion. Those firms’ spending, along with Microsoft’s growing projections, totals more than a staggering $630 billion.

And Big Tech is putting all of its eggs in one basket: Not only are the dollars dramatically higher, but the spend is more concentrated in a single purpose—scaling AI compute—rather than in a mix of strategic bets. 

The amount companies are spending on AI infrastructure now rivals that of some of the largest economies in the world and is comparable to the annual GDP of countries like Sweden and Israel. Capital expenditures fund such big-ticket infrastructure items as data centers, servers, and power systems that fuel the AI build-out race. Those data centers—some expected to be the size of a football field, or even four times the size of Central Park in Manhattan—require massive resources and energy to build, maintain, and operate.

“We’ve never invested this much in anything before,” Gil Luria, managing director and head of technology research at financial services firm D.A. Davidson, told Fortune. “But we’ve also never had a technology this promising before.”

Data centers in your shopping mall

As firms invest in physical data center infrastructure, some experts say the next round of build-outs could reach your town. “I firmly believe that the Stranger Things mall where they battle the creature will be converted to a data center,” Brent Thill, an analyst at investment banking firm Jefferies, told Fortune.

The magnitude of the current AI build-out is unlike any other investment in history. However, Luria said, the capex simply reflects existing demand. “It’s an unprecedented build-out,” Luria said. “But it’s really being done in conjunction with the growth in demand.” Luria points out that the demand backlog for Amazon, Meta, and Microsoft has reached new highs. Microsoft’s backlog, or the accumulation of orders the firm has accepted but not yet fulfilled, has doubled to $625 million thanks to OpenAI.

Thill said the build-out is addressing the AI industry’s existing bottleneck: physical infrastructure. He said the bottleneck has shifted from chips to energy, and now, it’s the physical shells that are lacking. “It went from a chip shortage, a GPU shortage,” Thill said. “Now, it’s a physical shell shortage.”

Market skepticism and the software squeeze

But as firms throw cash at AI infrastructure, it’s triggered a wariness of software valuations, causing a massive weeklong selloff of tech stocks and cryptocurrency as AI advancements cast doubt on the relevance of software technology. Although firms are bullish on AI’s potential, the technology has not yet paid off, and investors are reacting to uncertainty about its actual value. Coupled with weak jobs data, investor AI jitters spurred a wipeout of nearly $1 trillion from software and services stocks. But not everyone is concerned, including Nvidia CEO Jensen Huang, who has brushed off demands for short-term ROI.

Still, investors in tech giants are growing nervous because these firms are essentially exhausting their available capital to fund the infrastructure build-out, according to Luria. He said shareholders want to see returns, not added investment. “‘We understand that you want to invest all this money, but you’re investing all our money; you’re taking all your cash and all your cash flow and investing it,’” Luria said of the shareholder mindset.

Despite the selloff, Big Tech is betting on high ROI from AI. “We’re in a game of leapfrog now,” Thill said. “You have three to four big public vendors that are all lined up for this prize.”

As to why the build-out is taking off, Thill said that, given today’s demand for AI data centers, the only concern among tech firms is the risk of not doing enough. Any overbuild would grant some payoff.

“Even if you overbuild,” Thill said, “there’s so many people that would buy that overbuild even if they couldn’t sell it to their clients. Other people would want to procure it: state, local governments, [and] federal governments.”