In artificial intelligence, compute is the processing capacity used to train and run models: clusters of graphics processing units (GPUs) and other accelerators, the servers and networking around them, and the data-centre space and electricity they need. Model developers buy it directly, rent it from a hyperscaler or specialist cloud provider, or contract for capacity in an AI data center.
Because the hardware is costly and resellable, compute is increasingly financed like other equipment, through leases, an equipment finance agreement or asset-based lending secured on GPUs and the rental contracts behind them. Lenders focus on depreciation, since new chip generations can cut the value of older units quickly, and on the strength of customer commitments. Crypto projects also market access to compute through networks of distributed hardware, an area often grouped with DePIN.
Ref 1300000375