$27 Billion in Off-Balance-Sheet Exposure Lives in a Footnote. Footnotes Never Cause Problems, Right? Meta and Blue Owl Capital built a $27 billion joint venture called Hyperion to develop AI data centers. Instead of carrying that debt, Meta structured the deal as an operating lease. Clean books. Happy shareholders. The lease is only four years with options to extend. Short for a hyperscale deal. But there's a Residual Value Guarantee. If Meta walks away during the first 16 years, it writes a capped, declining check based on the gap between what the campus is worth and what was promised. Blue Owl gets a backstop. Meta keeps $27 billion off the liability column. Accounting rules say unless that payout is "probable," you don't record it. Ernst & Young flagged the lease accounting as a Critical Audit Matter, meaning it required significant judgment. Both sides agreed the payout was unlikely and moved on. I've been in commercial real estate for 30 years. I understand long-term guarantees on physical assets. A data center is trickier. The shell, power, and cooling hold value. Hyperion is getting 3 GW of dedicated power. That electrical capacity could be the most durable asset on campus no matter what hardware sits inside. The chips are a different story. Nvidia ships new GPU architectures every 12 to 18 months, each roughly doubling performance. I can't tell you what AI compute looks like in 2030. Neither can Meta. So the question is whether Hyperion was designed to swap hardware without gutting the building. If yes, the risk shrinks. If no, 16 years is a long backstop on today's architecture. Meta isn't alone. Moody's reported in February 2026 that the five largest hyperscalers have $662 billion in future data center lease obligations sitting off their balance sheets. That's 113% of their combined adjusted debt. The investors on the other side are pension funds, insurance companies, and sovereign wealth. They bought what looks like infrastructure-grade fixed income. The question is whether they priced in technology risk that traditional infrastructure doesn't carry. Each deal looks sound alone. The systemic pattern deserves scrutiny. What happens when 16-year financial guarantees collide with 18-month hardware cycles across an entire industry at once?