AI hardware financing

Training and serving frontier models requires clusters of tens of thousands of GPUs or TPUs — hardware that costs hundreds of millions to billions of dollars to acquire. Rather than record that liability directly, buyers route the purchase through a separate legal entity (a special-purpose vehicle or joint venture), treat the payments as operating expenses, or lease capacity from a partner who holds the asset. The hard part is disclosure: accounting rules differ by jurisdiction, and the same arrangement can be a transparent partnership or an opaque liability depending on how it is structured and what the auditors accept.