Hyperscalers on $1T AI Spending Spree
Google, Meta, and Oracle are embarking on a collective $1 trillion borrowing spree to fund a massive arms race in AI infrastructure. This capital expenditure boom is fueling fears of a shakeout similar to the dot-com era's fiber-optic overbuild. For startups, it means investors are now demanding ruthless capital efficiency and a defensible moat beyond just access to large models.
The projected 2026 capital expenditure for the top five hyperscalers—Microsoft, Amazon, Alphabet, Meta, and Oracle—is estimated to be between $685 billion and $690 billion. This represents a significant increase from the combined $381 billion spent in 2025. Approximately 75% of this 2026 spending, or around $450 billion, is expected to be dedicated specifically to AI infrastructure. This level of investment is consuming a massive portion of the companies' cash flow, with projections showing that the top five will spend about 90% of their operating cash flow on capex in 2026, a sharp rise from 65% in 2025. This has led to an increased reliance on debt markets to fund the expansion. For instance, investor Michael Burry noted that this spending could lead to negative free cash flow for Amazon in 2026 and a 90% drop for Alphabet. The scale of this AI-related capital expenditure has surpassed the peak of the dot-com era's fiber-optic buildout as a percentage of U.S. GDP. During the late 1990s, telecom companies invested over $100 billion, which led to a massive oversupply of fiber-optic cable and numerous high-profile bankruptcies, including WorldCom and Global Crossing. For startups, this capital concentration in the hands of a few tech giants is creating a bifurcated venture capital market. While overall VC funding for AI startups is at a record high, investors are increasingly concentrating their bets on fewer, more established AI companies. To secure funding, startups now need to demonstrate clear product-market fit and sustainable unit economics, with a focus on a 3:1 LTV/CAC ratio and a 12-month payback period.
