Nscale has filed for a U.S. initial public offering, giving investors their first detailed look at an AI infrastructure company that has accumulated more than $100 billion of contracted business while generating only $141 million of revenue during the first half of 2026.
The Nvidia-backed British company plans to list on the New York Stock Exchange under the ticker NSCL. Goldman Sachs, JPMorgan and Morgan Stanley are leading the offering.
Nscale has not yet disclosed how many shares it will sell or an IPO price range.
Financial Times reporting suggests the company could seek a valuation as high as $35 billion, which would be more than double the $14.6 billion valuation established during its March funding round. The preliminary SEC filing does not confirm that target.
The valuation debate will center on an enormous gap between what Nscale operates today and what it has promised customers over the next several years.
As of August 31, the company had approximately 25,000 GPUs running across five active data-center sites. Including signed projects, that figure expands to 461,000 GPUs and 1.37 gigawatts of active and contracted capacity.
Nscale says it has line of sight to more than 10 gigawatts of potential power capacity.
Its contracts tell an even bigger story.
Active contracts represented about $2.6 billion of total contract value as of August 31. Adding capacity that has been contracted but not yet activated pushes that figure to $103.4 billion.
The company now has to build enough data centers, secure enough power, purchase hundreds of thousands of advanced Nvidia GPUs and arrange enough financing to turn that contract value into actual revenue.
Revenue Is Growing From an Extremely Small Base
Nscale’s reported growth rate is extraordinary.
Revenue reached $140.6 million during the first six months of 2026, up 1,252% from $10.4 million a year earlier.
But those percentages require context.
The company generated only $19.1 million of revenue in 2024 and $33 million during all of 2025.
Nscale is therefore scaling from a very small operating base while simultaneously committing to some of the largest AI infrastructure projects announced anywhere in the industry.
Losses are rising even faster.
Nscale posted a $1.02 billion net loss during the first half, compared with $368.9 million a year earlier. The headline figure includes $457.1 million of fair-value losses and $174 million of depreciation and amortization.
Even excluding many of those accounting effects, the business is not close to profitability.
Adjusted EBITDA was negative $199.2 million in the first half, while operating losses reached $492 million.
Management argues those margins should improve as recently built infrastructure begins generating revenue against a relatively fixed operating base.
That is plausible.
It also depends on projects coming online when expected.
Microsoft and Anthropic Dominate the Contract Story
A small number of customers account for most of Nscale’s current and future economics.
Its largest customer represented 52% of revenue during the first six months of 2026. That was already an improvement from 73% during 2025 and nearly all revenue in 2024.
Future contracts remain similarly concentrated.
Between September 2025 and April 2026, Nscale signed agreements with Microsoft that could generate as much as $43.8 billion through December 2033 if Nscale satisfies required delivery and availability conditions.
In August, the company signed four agreements with Anthropic worth up to another $44.6 billion.
Together, those two customers account for the vast majority of Nscale’s disclosed contract value.
The contracts are generally structured as multiyear take-or-pay arrangements, with an average contract life of roughly 5.7 years. Once infrastructure is delivered and accepted, those agreements can provide unusually strong revenue visibility.
The word “once” carries a lot of weight.
Nscale’s SEC filing says it has not yet secured binding financing commitments needed to fund its performance under the Anthropic agreements.
That makes the $44.6 billion Anthropic number very different from an ordinary software backlog.
The demand has been contracted.
Much of the physical infrastructure required to satisfy it does not yet exist.
West Virginia Is the Biggest Execution Test
The Anthropic deployment is tied to Nscale’s Monarch Compute Campus in Mason County, West Virginia.
Nscale acquired the project in March through its purchase of American Intelligence & Power Corporation.
The 2,250-acre campus is designed around its own large-scale power infrastructure and could ultimately support more than 8 gigawatts of gross generating capacity and more than 6.5 gigawatts of IT load.
The initial phase calls for approximately 2 gigawatts of gross power, including 1.37 gigawatts of IT capacity, during the first half of 2028.
Anthropic plans to use Nvidia Vera Rubin systems deployed there.
It is a huge project even by AI data-center standards.
Nscale will need financing for both the GPU hardware and physical infrastructure, and the contracts require the company to meet defined delivery and service-availability thresholds before the economics fully begin.
That makes construction schedules, financing costs, power availability and Nvidia’s hardware delivery cadence central variables in Nscale’s future earnings.
Nvidia Is Both Supplier and Financial Backer
Nvidia’s relationship with Nscale extends well beyond being a recognizable shareholder.
The company is building much of its compute infrastructure around Nvidia’s latest GPU platforms, including Grace Blackwell Ultra and future Vera Rubin systems.
Nvidia has also repeatedly invested in Nscale.
On September 15, Nscale signed agreements covering at least $3.1 billion of additional financing, including $1 billion that will come from Nvidia through convertible notes or non-voting equity depending on when the transaction closes relative to the IPO.
Those securities will ultimately convert around the public offering.
Nscale separately raised roughly $2 billion in its Series C earlier this year, valuing the company at $14.6 billion. Investors included Nvidia, Dell, Nokia, Citadel, Jane Street and Point72.
That backing gives Nscale access to capital and strategic hardware relationships that would be difficult for a smaller data-center developer to replicate.
It also shows how financially interconnected the AI infrastructure ecosystem has become.
Nscale buys Nvidia computing systems. Nvidia owns Nscale securities. Nscale then rents the computing capacity to companies such as Microsoft, Anthropic and other AI developers whose own spending helps sustain demand for Nvidia chips.
Those relationships are not inherently problematic, but they become important when investors assess how much demand is independently funded and how dependent the infrastructure boom is on continued capital recycling throughout the AI sector.
The Balance Sheet Will Need to Keep Growing
Nscale held approximately $1.48 billion of cash at June 30.
That is substantial for a company that barely existed as an independent business two years ago.
It is small relative to what Nscale intends to build.
The company has already assembled a collection of GPU financing facilities, revolving credit lines, data-center loans and equipment-financing arrangements worth several billion dollars.
Management says its existing cash, committed borrowing capacity and customer prepayments should cover at least the next 12 months.
Beyond that, Nscale explicitly expects to keep financing growth through debt, equity, strategic partners and customer prepayments.
The IPO is therefore not simply a liquidity event for an established cloud provider.
Public equity would become another source of capital for a business whose revenue opportunity is expanding much faster than its current balance sheet.
Anyscale Pushes Nscale Beyond Renting GPUs
Nscale is also trying to avoid becoming just another capital-intensive GPU landlord.
In July it agreed to acquire Anyscale, the company founded by the creators of Ray, an open-source system used to distribute AI workloads across computing clusters.
The acquisition will add roughly 200 employees and give Nscale software that helps customers train, fine-tune and run AI models across large numbers of GPUs.
Financial terms were not officially disclosed, although reporting has placed the transaction near $1.65 billion.
The strategy resembles a broader shift among AI infrastructure providers.
Raw GPU capacity can eventually become more interchangeable, particularly as more data centers come online.
Owning the software layer gives Nscale an opportunity to capture higher-margin revenue and make customers less likely to move their workloads elsewhere solely because another provider offers cheaper computing.
It also adds integration costs and another business that needs capital before Nscale has demonstrated profitability in its existing operation.
The IPO Will Be Priced on What Has Not Been Built Yet
Nscale’s filing presents unusually large numbers for a company with only $140.6 million of first-half revenue.
Its $103.4 billion of active and contracted value is more than 700 times that six-month revenue figure.
That comparison does not mean Nscale is undervalued or overvalued.
It shows where the risk sits.
Most of the company’s value is tied to infrastructure and revenue that are scheduled for the future rather than assets generating cash today.
Nscale has legitimate evidence behind that expansion: Microsoft and Anthropic have signed enormous contracts, Nvidia is committing additional capital, more than 460,000 GPUs are already active or under contract, and customers are agreeing to multiyear take-or-pay structures.
The company also has a $1.02 billion first-half net loss, negative adjusted EBITDA, high customer concentration and significant financing still to arrange for its largest project.
The next version of the IPO prospectus will supply the missing number that brings all of those facts together: the actual valuation public investors will be asked to pay.
If Nscale ultimately seeks something close to the reported $30 billion to $35 billion range, investors will not primarily be paying for the 25,000 GPUs operating today.
They will be paying for Nscale’s ability to finance, construct and operate the other hundreds of thousands that have already been promised.