AllDataSI
Neoclouds · IPO filings

Can Nscale outlast the hype cycle?

The newest neocloud to file for an IPO has real customers and real cash. It also has a $103 billion backlog that is 97% unbuilt, and a walk-away clause sitting right where its risk is highest.

Nscale filed its S-1 on September 18, asking the NYSE to list it under the ticker NSCL at a valuation reported at up to $35 billion. That's more than double its $14.6 billion valuation from a funding round in March. Goldman Sachs, J.P. Morgan and Morgan Stanley are leading the deal.

My first reaction was that this looked like a company riding a wave of buzzwords: a two-year-old spinout of a crypto miner, now an "AI infrastructure platform" with a backlog bigger than most of its listed peers. After reading through the filing and the coverage, I think that label is wrong, but the skepticism behind it is right.

$140.6MRevenue, first half of 2026 (up from $10.4M)
$1.02BNet loss in the same six months
$103.4BContracted backlog, of which $2.6B is live
85%Share of backlog from Microsoft and Anthropic

What's real

Customers are paying cash up front. Microsoft and Anthropic have signed contracts and prepay a meaningful share of them, which is why Nscale reported about $1.7 billion of positive operating cash flow in a half-year where it lost a billion dollars. Large buyers don't prepay billions for slideware.

The hardware exists too. About 25,000 GPUs are running across 17 sites, and the revenue is actual rental income. Even the crypto origin is normal for this sector: IREN, Cipher, TeraWulf and Core Scientific all started as miners, because miners already held the power contracts and sites that AI data centers need.

What isn't built yet

Almost all of the headline number is still to be built. Of 461,000 contracted GPUs, roughly 5% are live. To turn the backlog into revenue, Nscale has to grow its fleet about twentyfold. It has around $29 billion of capital commitments against $1.5 billion of cash, its gross margin is negative 35% even before depreciation, and the filing carries going-concern language: substantial doubt about continuing without new financing. It also discloses material weaknesses in its accounting controls.

Then there's Nvidia, which is Nscale's chip supplier, an investor, and a lender through roughly $1 billion of convertible notes issued in September. Nscale has no long-term GPU supply contract behind $103 billion of commitments.

The walk-away clause

The detail I keep coming back to: Anthropic can terminate tranches of its contract without liability if Nscale delivers late.

Delivery clauses are common in capacity contracts. What makes this one different is where it sits. With only 5% of contracted capacity live, late delivery is Nscale's single biggest risk, so the protection kicks in exactly when Nscale is most likely to need the contract. Anthropic's $44.6 billion works less like a firm commitment and more like an option Anthropic holds.

A backlog is only as firm as its weakest exit clause. For Nscale, the exit clause sits on the largest single customer and triggers on the most likely failure.

The power question

Anthropic's capacity is planned for the Monarch Compute Campus, 2,000 acres in Mason County, West Virginia: 1.3 gigawatts to start, expandable to 8 GW, opening in 2027. Anthropic's 460 MW is about a third of the first phase.

West Virginia gets roughly 86% of its electricity from coal. A 2025 state law, HB 2014, lets certified data centers build their own power systems outside local zoning control, and the governor's plan is to grow generation with coal and gas. Utilities have told lawmakers they won't build new coal plants, so new data center power will most likely come from the state's abundant natural gas. Cheap power and fast permitting are the strongest parts of the site.

But Anthropic wasn't the campus's first tenant. Microsoft had a letter of intent for the full capacity and Google was in talks; both fell through over the summer. That leaves one customer, a 2027 build date and financing that, by the S-1's own account, isn't committed.

Will it survive the cycle?

It's tempting to say any company can be kept alive with more funding rounds. That's true in good times, and the real test of a hype cycle is what happens when they end. Nscale's plan assumes the capital markets stay open for years. If the cycle turns, the question becomes whether anyone will lend to it at all.

The closest parallel I know is the telecom buildout around 2000. Companies like Global Crossing laid real fiber, much of which is still in use. The companies went bankrupt, their shareholders were wiped out, and buyers picked up the assets cheaply. The infrastructure survived; the equity didn't. Neoclouds face a harsher version, because fiber lasted decades while GPUs are economically outdated in four to six years. A late build or a softer market could leave Nscale with debt on aging chips and a customer who has already walked.

What does outlast a bust is power: grid connections, permitted land, cooling. If Nscale fails, those are the assets a hyperscaler buys out of bankruptcy.

What I'm watching

I track these companies' SEC filings daily on the Filings Desk; Nscale's amendments and pricing will show up there as they're filed.

Sources

  1. Nscale Limited, Form S-1 (SEC EDGAR, Sept 18, 2026)
  2. Reuters: Nvidia-backed AI cloud firm Nscale reveals revenue surge in US IPO filing
  3. Mostly Metrics: Nscale IPO S-1 breakdown
  4. Quartz: Microsoft and Anthropic dominate Nscale's $103B backlog
  5. Startup Fortune: Nscale files for a $35 billion IPO
  6. West Virginia Public Broadcasting: Mason County data center nets multi-billion-dollar client
  7. Low-Carbon Power: West Virginia electricity mix
  8. Charleston Gazette-Mail: WV data center moves are fossil-fueled
  9. West Virginia Public Broadcasting: No new coal to make electricity, utilities tell lawmakers

This post is my own analysis for information only. It isn't investment advice.