Nscale arrived in the capital markets with two numbers that summarize the tension in the race for artificial intelligence infrastructure. In the first half of 2026, revenue rose 1,252%, to US$ 140.6 million, while net loss reached US$ 1.02 billion. At the same time, the company reported US$ 103.4 billion in total value of active and contracted contracts.
The S-1 filed for a future NYSE listing under the ticker NSCL, however, reveals something more important than growth or loss in isolation: a huge portion of the business sold by Nscale still needs to be financed, built, powered and equipped before it turns into revenue. It is this gap between contracted demand and operational capacity that could define the economic sustainability of the new generation of AI compute providers.
US$ 103 billion contracted does not mean US$ 103 billion in revenue
On August 31, Nscale had approximately US$ 2.6 billion in active contracts, compared with US$ 103.4 billion considering active and contracted contracts. The difference also appears in infrastructure: about 25,000 GPUs were active, while 461,000 were classified as active or contracted.
The filing itself offers a more restrictive accounting measure. Remaining performance obligations, or RPOs, totaled US$ 56.4 billion as of June 30. TCV and RPO, therefore, are not equivalent to already recognized revenue, nor do they mean that all the capacity needed to deliver these contracts is available today.
This detail changes the reading of growth. Nscale has already proven it can sell large blocks of future capacity. The challenge now is to turn these commitments into operational facilities within the expected timelines and costs.
Growth does not yet absorb the cost of capacity
The net loss of US$ 1.02 billion also needs to be broken down. About US$ 457.1 million came from fair value adjustments, while operating loss was US$ 492 million. Even excluding some of these effects, the operation remains far from equilibrium: adjusted EBITDA was negative US$ 199.2 million in the six-month period.
There is an even more direct sign. Nscale recorded US$ 189.6 million in cost of revenue, excluding depreciation and amortization, against US$ 140.6 million in revenue. The direct cost rose 2,331% in a year, above the 1,252% growth in revenue, mainly due to the expansion of sites, rent and energy.
This does not demonstrate that the model is structurally unviable. A significant portion of the infrastructure is still in the initial phase of use, and long-term contracts can change the economics of the assets as more capacity comes into production. But the filing shows that contracted scale, by itself, has not yet produced economic scale.
The bottleneck shifts from demand to financing
The scale of the challenge appears in future commitments. In June, Nscale had US$ 24 billion in purchase commitments for technology equipment not yet delivered, mainly for payment in 2026 and 2027. Another US$ 3.5 billion was committed to data center construction and related services in the same period.
This puts financing at the center of the model. The company had US$ 1.5 billion in cash as of June 30 and says it intends to finance its expansion with a combination of additional debt, equity, strategic partnerships and customer prepayments. The S-1 acknowledges that expansion will continue to require substantial investments.
The contract with Anthropic makes this dependency especially visible. The agreement provides for dedicated infrastructure at the Monarch campus in West Virginia, but requires Nscale to obtain qualified financing for GPUs and data centers. As of the date of the prospectus, the company said it did not have binding commitments for all the financing necessary to execute these contracts.
Nscale has been expanding its fundraising capacity. A few days before the filing, it closed US$ 3.1 billion in convertible notes with investors, including US$ 1 billion from Nvidia. Even so, the economic mechanism remains the same: new contracts create the need for infrastructure, which creates the need for new capital before producing revenue at scale.
Nscale makes public a model that already appears at CoreWeave
The dynamic is not exclusive to Nscale. CoreWeave ended 2025 with US$ 5.1 billion in revenue and a net loss of US$ 1.2 billion. In the same period, it used approximately US$ 10.3 billion in investing activities and ended the year with US$ 21.6 billion in debt. More than 98% of its revenue came from committed contracts.
Nebius follows a less leveraged structure, but it has also been bringing contracts and infrastructure financing closer together. In July, it raised US$ 775 million in debt secured by GPUs and contracted cash flows, in a structure the company intends to use to finance additional customer commitments.
The examples show how the AI cloud market is bringing together two traditionally distinct industries: cloud computing and project finance. Multi-year compute contracts help justify financing for GPUs, power and data centers; these assets, in turn, need to become operational so that the contracts generate enough cash to remunerate the capital raised.
The test will be converting backlog into margin before capital becomes more expensive
Nscale has contracted demand on a scale far larger than its current operation, but it also concentrates risks. Its largest customer accounted for 52% of first-half revenue, and the prospectus states that Microsoft and Anthropic should become significant customers in future periods.
The sustainability of the model, therefore, will depend less on the ability to announce new contracts and more on the speed with which already signed contracts pass through four stages: financing, construction, activation and margin generation.
The next relevant indicators will be progress on the financing needed for the Anthropic deal, the number of GPUs actually put into operation, the evolution of cost of revenue, customer concentration and the pace of debt growth relative to cash generated by facilities already active.
Nscale's IPO turns these variables into a public market question. The S-1 does not show a lack of demand for AI infrastructure. It shows precisely the opposite. The question that now begins to be tested is how much capital needs to be mobilized to deliver that demand and whether, when the data centers are finally operating at scale, revenue will grow fast enough to sustain the financial structure built before it.
