Nscale announces $3.36 billion in convertible financing for AI infrastructure

Nscale announced $3.36 billion in pre-IPO convertible loan note financing on 25 September 2026. The number is the announced size of a financing arrangement, with two different funding moments. Nscale describes a $2.36 billion initial tranche at closing and a further $1 billion commitment from NVIDIA, whose funding it expects in mid-November. It would be misleading to read the headline as confirmation that all $3.36 billion had already reached Nscale’s bank account.

Third Point led the financing. Nscale also names NVIDIA, funds managed by Apollo, Citadel, Hudson Bay Capital, Abu Dhabi Investment Council and 8090 Industries among its new and existing backers, with further participants listed in the release. Goldman Sachs & Co. LLC acted as placement agent. The transaction gives the infrastructure company a potentially large source of capital as it expands facilities for AI computing, but the release offers no allocation of dollars to individual sites or a timetable for when each site will become operational.

How the financing works

The securities are convertible loan notes, rather than a sale of ordinary shares today. According to Nscale, the notes will convert automatically into ordinary shares when an initial public offering is completed; NVIDIA’s notes would convert into non-voting shares. That condition describes what happens if and when an IPO is completed. Neither the phrase “pre-IPO” nor the automatic conversion provision means that Nscale has already listed on an exchange or announced an IPO date.

Nscale’s release includes forward-looking language about the proposed financing and the expected closing of tranches. It does not disclose a coupon, maturity, conversion price or valuation for these notes. Without those terms, the announcement cannot establish the future ownership percentage or cost of this capital. The funding schedule is also a reason to keep the initial tranche distinct from NVIDIA’s later commitment when comparing the deal with other AI infrastructure raises.

Why an AI cloud provider needs this kind of capital

Nscale describes its business as a vertically integrated stack: power supply, data centers, GPU clusters and cloud software for training and inference. The planned uses in its financing announcement stretch from behind-the-meter power plants to liquid-cooled facilities and large GPU installations. Each layer can constrain the others. A GPU order is useful only when a site can supply enough electricity, remove the resulting heat and connect the machines to customers. Construction, electrical equipment and cooling systems also require spending before the compute capacity can be sold at scale.

The company’s infrastructure page illustrates that physical scope. It describes Glomfjord in Norway as delivering 30 MW of operational compute capacity, with room to expand to 60 MW. It calls the Ward County project in Texas a roughly 240 MW data center designed with closed-loop direct-liquid cooling and rear-door heat exchangers, with a larger expansion plan. These are Nscale’s own descriptions of specific sites, not a promise that the financing will be spent there. They show why “AI cloud” involves power and thermal engineering as well as GPUs and software.

Liquid cooling moves heat from dense servers through a fluid loop rather than relying only on room air. It can support high-density GPU systems, but the complete system still needs pumps, heat exchangers, power distribution, network links and operating procedures. More financing can help pay for those pieces; it does not itself prove that a proposed campus is built, energized or serving customers. Nscale does not provide enough project-level information in this release to measure how much new capacity the notes will produce.

Reading the $103 billion contract figure

Nscale says it has more than $103 billion in total contracted value (TCV). TCV generally represents the aggregate value of contracts over their terms. The release does not break that figure down by customer, contract duration, delivered capacity, payment schedule or recognized revenue. It should therefore not be treated as $103 billion in annual sales, cash already collected or a measure of current profit. Contracts may support a case for expansion, while actual revenue depends on service delivery and the accounting treatment of each arrangement.

The near-term questions are straightforward: whether the anticipated NVIDIA tranche funds on the stated schedule, how Nscale deploys the capital across power and data center projects, and what operational capacity those projects deliver. The announcement establishes a significant financing plan and its conversion trigger. It leaves those execution results to be demonstrated later.

Source and availability checked on 26 September 2026.