$3bn · Debt · Cloud · London, UK
Nscale has closed approximately $3bn in aggregate commitments across two senior secured delayed-draw term loan facilities for AI infrastructure in Texas and North Carolina. The package provides up to $1.85bn for the company's Ward County campus and up to $1.2bn for its Madison site. J.P. Morgan and Goldman Sachs served as joint lead arrangers, joint bookrunners and co-structuring agents for both facilities.
Both facilities received investment-grade ratings with stable outlooks, according to Nscale. The structure matters: these are committed delayed-draw facilities, so the headline describes available financing rather than cash necessarily drawn in full on announcement. The debt is tied to defined capital expenditure at two US projects rather than general-purpose venture funding.
Financing follows the hardware
At Ward County, the facility will fund NVIDIA GB300 and VR200 systems supporting approximately 200MW of IT load, along with networking, storage and liquid-cooling equipment. Nscale describes the site as a purpose-built campus for high-density AI workloads. The North Carolina facility will finance a retrofit of a 96-acre colocation site with up to 40MW of IT load, plus the GPU and networking infrastructure needed to run high-performance compute deployments.
That split shows why AI-cloud expansion is increasingly an infrastructure-finance problem. Nscale sells compute capacity and software services, but delivering them requires power, buildings, cooling, networks and expensive accelerators to be installed together. Project-level delayed-draw debt can align borrowing with equipment and construction milestones; it also leaves execution risk concentrated in whether each campus is completed, energised and put into service on schedule.
Capacity has to become service
Nscale's model spans data centres, GPU clusters and a cloud software layer for AI training and inference. The Ward County build is also linked to the company's expanded Microsoft agreement, which included plans to deliver roughly 104,000 NVIDIA GB300 GPUs from the Texas campus in phases from the third quarter of 2026. The financing therefore supports capacity attached to a named commercial deployment, rather than an entirely speculative site build.
A subsequent SEC filing confirms the two facilities and describes their use for GPU-infrastructure capital expenditure. It also underlines the distinction between committed capital and operating output. Closing the debt package funds the build-out, but the commercial proof comes from installing the systems, bringing power and cooling online, and converting contracted capacity into reliable compute service.
