Nvidia has begun guaranteeing minimum revenue for neocloud GPU clusters and some datacenter leases, aiming to make AI compute financing viable beyond hyperscalers and large AI labs.
The move targets a funding bottleneck as AI debt needs are projected to reach about $7.1 trillion by 2029, with annual AI capex topping $2 trillion in 2028 and lenders still reliant on 5-year hyperscaler-backed contracts.
Under the typical 6-year structure, Nvidia offers a take-or-pay floor and shares in upside revenue above that level, letting lenders underwrite projects against Nvidia’s AA/Aa2 credit and usually size loans at roughly 70% to 80% LTV.
The program is designed to support shorter rental terms—especially sub-1-year contracts—for startups and inference providers that have struggled to secure capacity without heavy prepayments or long commitments.
Early disclosed deals are concentrated in Asia-Pacific, including SharonAI’s 40,000-GB300 Australia project with a $4.88 billion backstop and Firmus’s 360MW Indonesia cluster, with more negotiations underway.