The Debt-Funded AI Arms Race Is Running Out of Road

Meta ended 2025 with $81.59 billion in cash and $58.74 billion in long-term debt. A year earlier, that debt figure was $28.83 billion. In twelve months, Meta nearly doubled its long-term obligations, and then sold another $30 billion in corporate bonds in a single offering, the largest of 2025, the day after Zuckerberg told analysts the company would spend even more aggressively on AI. That is not a company funding growth from operations. That is a company making a leveraged bet on a technology whose commercial return timeline remains undefined.

The 2026 capital expenditure guidance tells you everything, $115 to $135 billion, nearly double the $72.2 billion Meta spent in all of 2025. Operating margins are already cracking under the weight of it, falling to 41% from 48% just one year ago as total costs rose 40% year over year. Free cash flow for full-year 2025 came in at $43.59 billion. At $135 billion in projected CapEx, the math does not work without continued borrowing. Analysts at Seeking Alpha have already modeled 2026 operating cash flows as insufficient to cover the buildout, meaning Meta enters negative free cash flow territory for the first time in its modern era.

The cost to build a hyperscale AI data center does not hold still when the components flow through a trade war. Meta’s Louisiana data center project, its Hyperion SPV, its third-party cloud dependencies, all of them face input cost inflation that was not priced into Q4 2025 guidance.

Morgan Stanley has estimated hyperscalers will spend roughly $3 trillion on AI infrastructure through the end of 2028, with approximately half financed through debt and structured vehicles rather than cash flow. When half of a $3 trillion buildout is debt-financed, and interest rates remain elevated, and tariffs are inflating input costs, and energy markets are unstable, you are no longer analyzing a technology cycle. You are analyzing a credit cycle dressed in a GPU.

The question is not whether Meta can service its debt today. It can. The question is what happens to a $135 billion annual CapEx commitment if advertising revenue softens in a recessionary environment, AI monetization lags consensus estimates, and refinancing costs rise. That is not a bear case. That is a sequence of events that has preceded every major technology bubble correction in the last thirty years.

And well let’s be truthful it’s not like Mark Z. has been wrong lately, Cough, Metaverse. Abject Failure. Very big gamble.