Product2 publishers3 min readPublished
Nscale's IPO filing shows just 2.5% of its $103.4 billion in contract value is active
Nscale's IPO filing counts $103.4 billion in active and contracted AI compute deals, with only about $2.6 billion of it active. Teams renting GPUs from smaller clouds should check whether their capacity is running yet and how the provider will pay to build the rest.
The Product Desk · Product desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Nscale says its customer contracts are mostly long-term take-or-pay deals, with a weighted average life of about 5.7 years.
- The contract total stood at $38 billion at the end of 2025, so it has grown about 2.7 times in eight months.
- Anthropic signed agreements on August 25 for dedicated GPU infrastructure at Nscale's Monarch Compute Campus, with payments of up to about $44.6 billion.
- Nscale's net loss reached $1.02 billion in the first half of 2026, up from $368.9 million in the same period of 2025.
- On September 15 Nscale agreed a financing of at least $3.1 billion, made up of $2.1 billion in unsecured convertible notes and $1 billion in notes or shares for NVIDIA.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- exposure If Anthropic's deals count in full, one lab accounts for about 43% of Nscale's contract value, and every smaller customer on the same hardware shares that dependence.
- cost At about $7.25 of net loss per revenue dollar, Nscale's build runs on outside money, so a customer on a long commitment relies on the next raise as much as on the hardware.
- decision A team whose launch needs new capacity in 2027 has to choose now between waiting on a site under construction and paying for running capacity elsewhere.
Picture an infrastructure lead pricing GPU time from Nscale this month. The company had about 25,000 GPUs running on August 31 [4]. Its preliminary S-1, filed on September 18 ahead of a planned NYSE listing under the ticker NSCL [1], counts 461,000 active and contracted GPUs [4]. Roughly 95% of that count was not yet active [17]. Nscale defines active capacity as online and generating revenue. Contracted capacity is IT capacity in facilities still under development [5].
The contract book follows the same pattern. Active TCV is about 2.5% of the total [19]. TCV counts revenue over the full committed terms of signed deals, not money already earned [3]. Spread evenly across the average contract life, the total comes to roughly $18 billion a year [20]. Nscale booked $140.6 million of revenue in the first half of 2026, up from $10.4 million a year earlier [6]. Two of the named deployments have dates. The Microsoft build of more than 66,000 Vera Rubin GPUs in Portugal begins in late 2027 [11], and Figure AI's first deployments are targeted for the second half of that year [12].
A backlog that size is easy to take as proof that a provider is safe to rent from. Take-or-pay terms commit the buyer to pay, but Nscale still has to finish the sites and install the hardware before contracted capacity earns revenue [9][5]. The filing itself lists customer concentration, access to financing and timely deployment of contracted infrastructure as business risks [13].
The reporting does not break out how much of the running fleet is sold as general cloud capacity and how much is dedicated to contract customers. Nscale sells to hyperscalers, AI labs, enterprises and government-backed programs, and its services include dedicated GPU clusters alongside cloud computing [15]. I'd expect a small team renting from a provider in this position to get whatever the long contracts leave, at least until the new sites come online.
For anyone choosing a smaller GPU cloud, I'd sort the decision on two axes. The first is whether the capacity on offer is active in Nscale's sense, meaning online and billing, or contracted, meaning a site still being built. The second is whether your workload could move to another provider within a few weeks.
Active and portable: rent on short terms. You give up whatever discount a longer commitment would have bought, and you keep the option to leave. Active but hard to move: ask how the provider pays for its buildout before you sign, because a slow exit means you stay through any funding shortfall. Contracted but portable: treat the delivery date as a forecast and keep a second quote current. Contracted and hard to move is the take-or-pay position the large buyers in this filing accepted [9]. A team that cannot absorb a late site should not take it.
What to watch
- How the NYSE listing prices against the possible valuation of up to $35 billion that has been reported, after a September 29 report questioned the buildout.
- Whether the planned Anyscale acquisition closes, putting the Ray-based compute platform that other teams build on under a GPU provider's ownership.