Nscale Says $103 Billion Is Already Booked — the Neocloud IPO Window Just Cracked Open

A UK startup you've probably never invoiced says it has $103 billion in contracted revenue and wants to go public as soon as this month. Nscale, the Microsoft- and Nvidia-backed "neocloud," is telling prospective investors its total contracted revenue now sits around $103 billion — roughly double what it was before Anthropic signed its name to a $45 billion compute deal — with an IPO that could come as soon as this month (Reuters, citing The Information). Two days ago that number was ~$50 billion. The doubling took one customer.

Read past the eye-watering headline and this is the moment the AI infrastructure trade gets its public-market referendum. Contracted revenue — future commitments, not cash — has quietly become the valuation currency of the neocloud sector. Nscale's IPO would be the first big test of whether public investors will pay growth multiples for a business whose order book is measured in decades of future GPU rent.

What Actually Happened

• Nscale is telling prospective investors it holds about $103 billion in total contracted revenue, ahead of an IPO that could come as soon as this month, The Information reports (Reuters, Sept 2).
• That's roughly double its contracted revenue before the Anthropic deal — a single $45 billion, six-year agreement accounts for most of the jump (Yahoo Finance).
• The Anthropic deal commits roughly 460 megawatts of compute capacity at Nscale's flagship data center development in West Virginia (CNBC, Bloomberg).
• Anthropic signed a second $35 billion neocloud deal the same week — this one with Lambda, the Nvidia-backed operator, for a site in Nueces County, Texas (Bloomberg via MSN, Sept 2).
• Nscale's Microsoft relationship deepened to up to $30 billion in dedicated capacity at a ~1-gigawatt Fairmont, West Virginia site, which Nscale called "the largest contracted order in its history" (Data Center Frontier, Sept 1).
• Microsoft and Nvidia are already Nscale investors — meaning the startup's anchor customers and its backers are largely the same circle (Data Center Frontier).

Backlog Is the New Valuation Currency

The number doing all the work here is "contracted revenue," and it deserves a hard stare before anyone anchors on it.

$103 billion is not revenue in any sense a CFO from 2015 would recognize. It is the sum of signed multi-year commitments — mostly the Anthropic $45 billion, the Microsoft up-to-$30 billion, and other capacity agreements — that will be recognized over the lives of those contracts. A six-year deal recognized over six years contributes its value once, spread thin, with real costs (land, power, GPUs, depreciation) arriving far earlier in the curve. Neocloud economics are famously thin on margin and brutal on depreciation; the backlog number tells you demand exists, not that the business prints money.

Why use it anyway? Because that's the sector's chosen scoreboard. CoreWeave went public on the strength of its backlog and spent the following year whipsawing on every revision to it. The private-market equivalent — CoreWeave's own IPO, large compute deals, secondary rounds — has all been priced off contracted commitments. Nscale telling investors "$103 billion" ahead of a public listing is the logical endpoint: a company hoping to convert a backlog multiple into a market cap before anyone can audit the churn, the counterparty credit quality, or the actual gross margin hiding underneath.

To be fair to the model, the counterparties here are not vaporware. Microsoft pays its bills. Anthropic is generating real revenue and just posted a frontier-model release with enterprise pricing designed to win exactly these scale customers. This is a stronger order book than the sector's skeptics assume. The open question is what margin structure sits beneath it — and that's precisely what a prospectus would reveal and a private funding round never does.

Anthropic Is Buying Its Own Second Sources

Zoom out one week and the more interesting pattern emerges. Count Anthropic's newly reported third-party capacity commitments: $45 billion to Nscale. $35 billion to Lambda. Up to $30 billion in dedicated capacity through Microsoft. That's on the order of $110 billion in compute arrangements from a single AI lab in a matter of weeks — from a company that is not the largest spender in its own tier.

Anthropic is executing a second-source strategy for the compute layer the way a supply-chain officer would. No single landlord. No single site. Multiple operators (Nscale, Lambda, the Azure backbone) across multiple geographies (West Virginia, Texas, and Microsoft's footprint). This is what compute procurement looks like when a lab's roadmap depends on capacity it doesn't own and can't afford to queue for: buy the same capability several times, from different vendors, and let them compete on the next renewal.

The implications run downhill:

• The hyperscalers are being hedged. Anthropic's Google and Amazon relationships remain, but the marginal dollar of compute is going to specialist operators who will build to spec. Every new neocloud deal slightly reduces the strategic leverage of the big three clouds over their own AI tenants.
• Capacity is being locked up years in advance. These are not spot-market purchases; they are 460-megawatt-class, multi-year reservations. Anyone planning to rent frontier-scale GPU capacity in 2027–2028 from the merchant market should assume the best blocks are spoken for.
• The financing round-trip is getting tighter. Nvidia backs Nscale and Lambda; Microsoft backs Nscale and buys from Nscale; Anthropic buys from all of them while itself being the revenue engine that justifies the build. Nobody in this loop is doing anything improper — but the circularity means the sector's demand signals are increasingly self-referential, and a public listing will force that structure into daylight.

The IPO Will Price the Circular Question

If Nscale lists this month, it becomes the first pure-play neocloud with an Anthropic-scale anchor tenant to face public markets. The prospectus will have to answer, in audited numbers, the questions the private market never asked:

1. What's the actual gross margin on contracted compute? Power costs, GPU depreciation schedules, and utilization assumptions — the three numbers that decide whether $103 billion of backlog is a fortune or a treadmill.
2. How much concentration risk sits in two customers? Anthropic and Microsoft are the book. If either slows expansion, renegotiates, or builds its own capacity, the backlog revalues overnight.
3. What does the capex curve look like against the revenue curve? Neoclouds buy next-generation hardware on the promise of future rents. Public markets are less patient with that timing than venture investors were.

The bear case is simple: this is CoreWeave's story with a bigger headline number, and CoreWeave's stock has spent its public life teaching investors how volatile backlog-driven AI infrastructure really is. The bull case is equally simple: the demand is real, the counterparties are creditworthy, and the market is underestimating how long the compute shortage lasts. Nscale's listing will force one of those narratives to win in public, with daily pricing — and every other neocloud, from Lambda to the next Nvidia-backed entrant, will be marked to that result.

What Builders Should Take From It

• Read the prospectus when it lands — it's free market research. It will be the first audited look at real neocloud unit economics. Whatever it says about margins, depreciation, and customer concentration applies to the entire GPU-rental layer you're building on.
• Treat long-horizon GPU pricing as a market that's being locked. Multi-year capacity reservations at this scale mean spot prices stay volatile but baseline availability tightens. If your product depends on heavy inference compute, contract earlier than feels comfortable.
• Copy Anthropic's second-source pattern for anything critical. One vendor per critical dependency — model, cloud, payment, or compute — is a single point of failure wearing a contract badge. Anthropic is spending $110 billion to avoid being anyone's captive customer.
• Discount "contracted revenue" in any AI pitch you see. Founders, investors, and buyers should ask the same three questions: recognized over what period, from how many counterparties, at what margin. A backlog is a claim, not a fact of cash.
• Watch the IPO outcome as a sector signal. If public markets pay up for backlog-backed neoclouds, expect a wave of similar listings and accelerated build-outs. If the offering stumbles, the private-market valuations underneath the whole compute layer get retested — and so does the pace of new data center supply.

A company that didn't exist in its current form a few years ago is about to list on the strength of future rent from two customers, backed by the investors who sell it the chips. That's not necessarily a bubble — the demand lines are real and the counterparties are serious. But it is the moment the AI infrastructure trade stops being a private-market story you read about and becomes a public one you can price. Pay attention to the first week of trading. It'll tell you what the market thinks the next three years of AI compute are worth.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.