Big Tech has committed over $3 trillion to AI infrastructure—and the returns are still nowhere in sight

What happened

Following Q3 earnings calls from Amazon, Microsoft, Alphabet, and Meta late last month, two realities crystallized: Big Tech is accelerating—not easing—its AI data center spending, and the promised revenue payoff remains elusive.

According to a Bloomberg tally reported by The Next Web, five tech giants—Alphabet, Meta, Microsoft, Amazon, and Oracle—have now committed more than $3 trillion in future spending on data centers, leases, equipment, and energy. That figure is up from approximately $2.4 trillion in a similar count just months earlier in July (which did not include Oracle). Future leases account for $1.3 trillion of the total, with the remaining $1.9 trillion in other orders. Some leases run as long as 30 years.

Alphabet leads the pack with more than $1 trillion lined up—nearly eight times its level a year ago. Meta follows with over $800 billion, roughly half of which is tied to data center leases that have not yet commenced. Meta CFO Susan Li told analysts the company expects to "continue investing aggressively to support both our core business and emerging AI initiatives."

Separate tracking from CapexIndex (via nexi.fund) puts combined capital expenditure across 45 public companies at $941.5 billion as of July 2026, with an estimated $399.5 billion directly attributable to AI. The Big Four—Amazon, Alphabet, Microsoft, and Meta—account for roughly $580 billion of that total, a 77% increase from 2025. Notably, the spending is no longer going primarily to GPUs; power, cooling, and data center construction now consume over 60% of the budget.

The scale is historically unprecedented. The buildout surpasses the combined inflation-adjusted spending of the U.S. Interstate Highway System and the Apollo program.

Why it matters

The numbers are staggering, but the core tension is simple: no earnings call has yet answered what happens when the spending stops accelerating. Big Tech is pouring capital into physical infrastructure—power plants, cooling systems, buildings, and decades-long leases—at a pace that assumes AI demand will grow fast enough to fill it. If that demand materializes, the companies that locked in capacity early win. If it doesn't, they are left with enormous fixed costs and lease obligations that could take years to unwind.

The shift in spending composition is also significant. With over 60% of capex now going to power, cooling, and construction rather than chips, the buildout has become a real-estate and energy story as much as a computing one. That means the bottleneck is no longer just GPU supply—it's grid capacity, permitting timelines, and construction labor.

For investors, the question is whether these companies can generate AI revenue that justifies capital expenditure growing at 77% year-over-year. For competitors and startups, the question is whether they can access compute at all when hyperscalers have pre-purchased decades of capacity.

What to watch

• Whether any of the Big Four revise their capex guidance downward in subsequent quarters—a signal that demand assumptions are softening.
• AI revenue disclosure: companies have been vague about how much revenue is directly attributable to AI products versus repackaged cloud services.
• Energy and grid constraints: if power availability becomes the binding limit, spending may shift from data centers to energy infrastructure deals.
• Regulatory attention: the scale of pre-purchased capacity could attract antitrust or competition scrutiny, particularly if smaller players are priced out of compute access.
• Oracle's inclusion in the $3 trillion tally suggests the spending club is expanding—watch for additional entrants.