Only $2.6 billion of Nscale’s $103.4 billion in active-plus-contracted total contract value was active at August 31. About 25,000 of 461,000 active-plus-contracted GPUs were in service.
Those gaps define the proposed initial public offering. The signed contracts are commercially important, but most of the capacity behind them still needs to be financed, built, delivered and accepted before the service term starts. Nscale’s filing gives investors a contract-conversion problem, not a conventional backlog multiple.
The income statement is just as easy to misread. For the six months to June 30, Nscale reported $140.6 million of revenue, a $1.02 billion net loss and negative $199.2 million of adjusted EBITDA. The net loss includes large non-cash and financing-related charges. The adjusted result still shows a young operator spending ahead of its live estate.
The contract total starts before the service does
Nscale defines active capacity as IT megawatts that are online and generating revenue. Contracted capacity is under development pursuant to signed customer contracts. Its committed contracts reserve specified capacity for fixed terms, generally on a take-or-pay basis. Customers are billed for reserved capacity regardless of actual use, but revenue recognition begins as service is provided.
TCV is broader. Nscale defines it as aggregate revenue across the full committed term of signed agreements, measured when a contract is signed. It excludes optional extensions, renewals, unexercised capacity and significant financing components tied to upfront payments.
That makes $103.4 billion a valid measure of signed commercial scale. It does not make it revenue. It is almost 40 times the $2.6 billion attached to active contracts. Active GPUs were 5.4% of the active-plus-contracted fleet. The two percentages are not interchangeable because chip generations, prices and contract lives vary, but both locate the same bottleneck: conversion.
The site ledger points the same way. Five data-center sites were active and 12 were contracted, with 1.37 GW of active-plus-contracted capacity. Site count and megawatts do not map directly to GPU count, but both show how much of the reported estate remains under development.
Remaining performance obligations provide another view. Nscale reported $56.4 billion of RPO at June 30, defined as contracted revenue not yet recognized under service contracts. That figure includes deferred revenue and non-cancellable amounts to be invoiced. It also includes a $7.1 billion significant financing component from customer prepayments, which TCV excludes. RPO and TCV therefore cannot be treated as two labels for one backlog number.
Prepayments move cash forward
Nscale’s first-half cash flow looks stronger than its earnings. Operating activities provided $1.686 billion, primarily because customers paid upfront against long-term contracts. Investing activities used $3.286 billion, largely for GPUs and technology equipment. Financing provided another $1.628 billion, mainly from the Series C raise.
Customer cash is useful funding. It is not earned revenue at receipt. Nscale records a contract liability and recognizes revenue over the service term. The filing says the weighted-average prepayment across active and contracted TCV was 23% at August 31. This structure shifts part of the build cost to customers, but it also pulls cash forward from services that remain to be delivered.
The $1.02 billion net loss should not be called cash burn. It included a $457.1 million fair-value loss, mainly from Sandton and Nvidia warrant liabilities, $174 million of depreciation and amortization, and $113.8 million of share-based compensation. Yet the adjusted result is not comfortable. Cost of revenue excluding depreciation was $189.6 million, already above $140.6 million of revenue. Adjusted EBITDA was negative $199.2 million.
Capital commitments show why funding remains central. At June 30, Nscale disclosed $24 billion of purchase commitments for undelivered technology equipment, $3.5 billion for data-center construction and $1.4 billion of undiscounted payments for leases not yet commenced. The equipment and construction amounts were payable primarily during 2026 and 2027.
Take-or-pay does not remove delivery risk
The contracts reduce utilization risk after capacity goes live. They do not remove construction risk before that point. Contract terms commence only after successful delivery of GPU clusters.
Microsoft agreements covering payments of up to $43.8 billion through 2033 are subject to delivery and service-availability requirements. Anthropic agreements provide for up to $44.6 billion, also subject to delivery and availability. For Anthropic, Nscale must use best efforts to obtain qualifying finance for the GPUs and data-center infrastructure. The filing states that no binding financing commitments for that performance had been obtained as of the prospectus date. Missed delivery windows and sustained uptime failures can give Anthropic termination rights over affected tranches.
Customer concentration runs through the live business as well. The filing reports that its largest customer supplied 52% of first-half revenue, while the corresponding concentration measure for 2025 was 73%. It does not identify whether those figures refer to the same counterparty. The maximum Microsoft and Anthropic payment amounts sum to $88.4 billion, about 85.5% of August TCV. That calculation is not revenue guidance. It shows that conversion timing, customer acceptance and counterparty performance matter almost as much as the aggregate contract total.
Nscale has raised substantial capital. A September subscription agreement covers at least $3.1 billion: $2.1 billion of unsecured convertible loan notes plus $1 billion from Nvidia in either notes or non-voting shares, depending on closing timing. The Nvidia commitment is expected to close around November 16. That is committed capital with execution conditions, not cash already available for the full Anthropic build.
Take-or-pay protects the economics of delivered capacity. The decisive risk sits between signature and acceptance.
The filing still leaves the IPO price range and offered share count blank. Without them, investors cannot assess valuation or dilution. They can assess the operating test.
The most useful quarterly scorecard will not be TCV alone. It will be active TCV, active GPUs, revenue, adjusted EBITDA, capital spending and the share of contracted capacity that reaches customer acceptance. If those measures rise together, the contract book is converting into an operating business. If the headline total grows faster than the active base, financing remains the product Nscale must deliver first.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.