Meta Platforms and other major technology companies are facing a widening financial gap between the cost of producing artificial intelligence and the massive capital expenditures required to build the necessary infrastructure.
The article highlights a divergence in costs: inference-token costs have fallen by approximately 47% per quarter, or roughly 13 times annually. This rapid deflation benefits users, but it creates a challenge for developers who must cover the high upfront costs of data center construction.
According to the report, hyperscalers—including Meta, Amazon, Alphabet, Microsoft, and Oracle—plan to spend about $750 billion on data centers this year. This spending is projected to exceed $1 trillion next year and surpass $5 trillion over the next four years.
Wharton finance professor and former SEC chief economist Jessica Wachter warns that these companies must achieve 2.7 times productivity gains by 2030 to cover depreciation, the cost of capital, and a standard 15% hurdle rate. Without such gains, Wachter characterizes the current buildout as potentially the largest misallocation of capital in history.
Financial constraints are already impacting major players. Infrastructure spending consumed nearly all of Alphabet’s nearly $120 billion in quarterly revenue, resulting in a $5.9 billion free-cash deficit for the first time since its 2004 IPO. Morgan Stanley estimates that more than half of the $2.9 trillion in spending by hyperscalers through 2028 will be financed with external capital.
Meta’s financial engineering illustrates this trend. The company handed 80% of its $30 billion Hyperion campus in Louisiana to Blue Owl Capital through a joint venture. The subsidiary then leases the buildings back to Meta on four-year terms, matching the expected lifespan of the GPUs inside. Stijn Van Nieuwerburgh, a finance professor at Columbia, noted that if Meta were to walk away from the deal, investors would be left with an empty building and no cash flow.
The article also points to a disconnect between AI investment and economic productivity. A Stanford survey of about 6,000 senior executives found that about 90% reported no productivity increase over three years. Nobel laureate Daron Acemoglu suggested that if productivity gains do not materialize, public sentiment toward AI may sour.
Regarding the broader macroeconomic picture, Oaktree co-founder Howard Marks warned that multi-trillion-dollar AI capital requirements are colliding with structural U.S. federal deficits running near 6% of GDP. He noted that increased demand for capital causes the cost of money to rise.
MIT Sloan professor and former SEC Chair Gary Gensler described the entire AI buildout as a "parlay bet" by the capital markets. To deliver on current valuations, hyperscalers must simultaneously extract trillions in direct software revenue, achieve widespread macro productivity, and successfully defend frontier model pricing against low-cost alternatives.
