Big Tech hyperscalers are piling on debt at a record pace, with $121 billion in new borrowing this year alone to bankroll artificial intelligence infrastructure. That figure comes from UBS, which also projects an additional $900 billion in debt for 2026.
The scale of the borrowing
The $121 billion raised in 2025 marks a sharp increase from previous years. The money is flowing into data centers, specialized chips, and the energy systems needed to power them. Hyperscalers — the handful of companies that operate massive cloud and AI computing platforms — are racing to build capacity before demand outstrips supply.
What the debt funds
AI models require enormous computational resources. Training a single large language model can cost tens of millions of dollars, and running inference at scale demands even more hardware. The new debt is earmarked for expanding server farms, purchasing graphics processing units, and securing long-term energy contracts. Some of the borrowing also goes toward acquisitions of AI startups and talent.
Investor appetite remains strong
Despite the eye-popping totals, bond buyers have shown little resistance. The hyperscalers carry investment-grade credit ratings, and their AI investments are seen as a bet on future revenue growth. The debt is being issued in both investment-grade and high-yield tranches, with maturities ranging from three to 30 years.
The 2026 projection
UBS expects the pace to accelerate. The bank's $900 billion forecast for next year would bring the two-year total past $1 trillion. That level of borrowing would test market capacity, especially if interest rates remain elevated or if AI revenue growth disappoints. For now, though, the hyperscalers are pressing ahead.
