Nscale wants $35 billion on two customers and $103 billion

Nscale has $103 billion of contracts and two customers. The British neocloud plans to list on the NYSE at an expected $35 billion valuation, raising about $3 billion, and its IPO filing sets out a concentration risk in unusually plain numbers.

Roughly 85% of that contract book sits in two agreements: $43.8 billion to supply Microsoft with compute through 2033, and $44.6 billion with Anthropic. TechCrunch's read of the filing flags the condition attached to the second one.

The Anthropic deal has an exit

Anthropic's agreement is contingent on Nscale obtaining financing. The AI lab keeps the right to walk away or cancel if Nscale misses milestones that the filing itself describes as "stringent."

Read that sequence in order. Nscale needs capital to build the capacity. The contract that justifies the capital is conditional on raising the capital. And the customer can leave if the build runs late. That is not a backlog in the sense a manufacturer uses the word, it is an option the counterparty holds.

Which makes the $103 billion headline figure a poor guide to anything. The number is real, the contracts exist, and the enforceable portion of the larger one depends on Nscale executing a financing and construction programme it has not completed.

The financials are early-stage at scale

Revenue was $140.6 million for the six months to June 30, against $10.4 million in the same period a year earlier. That is a thirteenfold increase and it is a tiny base.

Net losses over the same period reached $1.02 billion, up from $369 million. So the company is losing roughly seven dollars for every dollar of revenue while asking public markets for a $35 billion valuation on the strength of contracts that mature through 2033.

There is a coherent version of this bet. Building data centre capacity is capital-intensive up front and generates revenue over long contracts, so early losses against booked future income is the expected shape. The question is whether the contracts hold, which returns to the cancellation clause.

Two pieces of arithmetic are worth doing before the pricing looks reasonable. Annualise the half-year revenue and you get about $281 million, which puts a $35 billion valuation at roughly 125 times sales. And the $3 billion raise, set against a half-year net loss of $1.02 billion, covers about 18 months at the current rate of burn before the company needs the next round. Neither number is disqualifying for infrastructure at this stage of a buildout. Both mean the listing is being priced off 2033 rather than off anything already happening.

Worth noting where the company came from: Nscale was spun out of the Australian cryptocurrency mining company Arkon Energy two years ago. Two years from crypto mining subsidiary to $35 billion AI infrastructure listing is a fast transition, and it means the operating history being underwritten is short.

Everyone in this business has the same problem

Nscale's concentration is not unusual, which is the more useful finding. A paper from credit hedge fund Sona Asset Management, covered by the Financial Times, found the pattern across AI infrastructure providers.

CoreWeave takes 67% of its revenue from Microsoft. Data centre builder Applied Digital takes 67% from Oracle and another 30% from CoreWeave, which means it is exposed to Microsoft twice, once directly and once through a customer. Sona's point is not that interdependence is inherently bad, but that a single setback or strategic change at one large buyer propagates through the whole sector at once.

That is the actual test this listing represents. Public investors have already bought concentrated AI infrastructure exposure through CoreWeave. Nscale asks them to do it again, with a shorter history, a conditional anchor contract, and a customer base of two.

The buyers doing the concentrating are not shy about their plans. OpenAI's own investor material projects $856 billion of compute spending and a $278 billion cash shortfall through 2030, and Nvidia's chief executive expects chip sales to double next year. Every one of those figures is a commitment rather than a delivery, which is the same category of asset as Nscale's backlog. Markets have shown they will fund an ambitious forecast, as they did when SpaceX shares rose on a revenue call nobody could verify.