Hyperscalers will spend over $1 trillion on AI data centers next year. They need a 2.7x productivity boost by 2030 to break even. Current AI revenues are only $150–200 billion this year. Alphabet reported its first free cash deficit since 2004 due to AI spending. Over half of future data center costs will be borrowed. Economists see little economy-wide productivity growth from AI so far. Executives expect a 1.45% productivity boost over the next three years. Hyperscalers must also fend off cheaper AI models. Borrowed funds expose financial institutions to data center risks. Meta’s Hyperion data center cost rose from $10 billion to $30 billion. Its financing now involves a joint venture with Blue Owl Capital. The venture uses complex leases and guarantees to fund construction. Financial risks are spreading through pension funds and insurance policies.
What changed
Hyperscalers previously funded AI data centers primarily with accumulated cash rather than borrowed capital.
What it unlocks
Sustaining AI infrastructure investments requires achieving unprecedented productivity gains and revenue growth to avoid financial strain.
- $1.1 trillion hyperscaler AI data center spending by 2027
- $750 billion spent this year on AI data centers
- $5 trillion projected AI capital investments over next four years
- $150–200 billion total AI revenues this year
Sources