Nscale vs Firmus: How to Read a Neocloud IPO ($103 Billion Backlog, $1.02 Billion Loss and a Downsized Offering)

By Carlos Montiel | Enterprise AI Specialist
Leer en español →
Published: 2026-10-12 | By: Carlos Montiel | Reading time: ~4 minutes

Two Nvidia-backed neoclouds are going public with the same pitch — decades-long GPU contracts — and this week the market answered them differently: Nscale filed an S-1 showing a US$1.02 billion loss in six months, and Firmus had to cut its offering.

What Nscale's S-1 Revealed

On September 18, 2026, Nscale filed its S-1 with the SEC to list on the NYSE under the ticker NSCL. The global coordinators are Goldman Sachs, J.P. Morgan and Morgan Stanley. The first-half 2026 numbers show the classic pattern of a neocloud in build-out mode: revenue multiplying, losses growing even faster.

Nscale (S-1, H1 2026 vs H1 2025) Revenue: US$140.6M (vs US$10.4M) +1,252% Net loss: US$1,020M (vs US$368.9M) Contracted backlog: US$103.4B as of Aug 31 (vs ~US$38B at end of 2025) of which, active contracts: ~US$2.6B Active or contracted GPUs: ~461,000 Average contract length: 5.7 years Largest customer: Anthropic, up to ~US$44.6B (agreements of Aug 25, 2026)

The number that jumps out is the gap between US$103.4 billion in backlog and only about US$2.6 billion in active contracts: the rest depends on securing financing and actually building the data centers. The Register also notes that Nscale warns in its prospectus that it has no binding financing commitments to fulfill the Anthropic agreements. At up to US$44.6 billion, a single customer accounts for roughly 43-44% of the backlog.

Firmus: The Same Bet in a Different Market

In Australia, Firmus (backed by Nvidia, with a US$10 billion debt facility led by Blackstone) is aiming to debut on the ASX under the code AIF around October 23. Its original plan: shares at A$11, a market capitalization of A$43.7 billion, and an offering of around US$5 billion, with more than 60% earmarked for servers and deployment. Its flagship project, Southgate, in Tasmania, plans for about 36,000 Nvidia GB300 GPUs and a first stage of 90 MW. Firmus expects a loss of A$77 million in the first half of its fiscal year.

This week's twist: The Nightly reported on October 8 that the price was cut by about 25% (it reports A$8.25 per share) and that the offering would shrink from US$5 billion to US$3 billion, with fund managers questioning the valuation and the structure. Other reports cite a new price of A$9, so the exact figure is not yet confirmed; what they agree on is that demand was weaker than expected. The Nightly also mentions bond yields at their highest since 2002, which makes the debt-funded plan more expensive.

Nscale vs Firmus: How to Read a Neocloud Before Buying Compute From It

Both companies share the same model: raise debt and equity against long-term contracts with a handful of AI customers, build data centers with Nvidia GPUs, and lease out the capacity. The differences matter if you depend on them:

Nscale Firmus Exchange NYSE (NSCL), ~US$30-35B ASX (AIF), A$43.7B (under review) Key customers Anthropic, Microsoft Reports mention OpenAI and Meta Status Public S-1, pricing pending Offering cut on weak demand Core risk Unfunded backlog Valuation and interest rates

A useful rule of thumb: contracted backlog is not revenue. To evaluate a compute provider, separate active contracts from conditional ones, look at concentration in a single customer, and check how much of the financing is binding.

What this means for a company buying AI: if your product depends on a neocloud's capacity (directly, or through a model that runs on it), ask about its counterparty risk the way you would with any critical vendor: what happens to your workloads if construction or financing is delayed? Avoid a single inference provider, keep portability clauses across clouds, and don't assume today's price per token will stay stable if interest rates rise or the IPOs go badly. Nscale's figures come from its S-1; Firmus's come from press reports and may still change.
Carlos Montiel
Enterprise AI Solutions Architect
Specialist in LLMs, Agents, and Orchestration
guatemalia.com/#contacto · info@guatemalia.com

Need to implement AI in your company?

Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.

Contact Carlos Montiel

info@guatemalia.com