Nscale Seeks $3.5 Billion in Financing Before Its IPO

Nscale, a two-year-old British AI infrastructure company, is in talks to raise as much as $3.5 billion before a planned initial public offering in New York, Bloomberg News reported on September 4, 2026, citing people familiar with the matter. The London-based firm, which builds and operates GPU-powered data centers for AI workloads, is seeking to close the financing ahead of a listing that could come as early as this month, according to TechCrunch and further reporting from The Next Web.

What Nscale actually sells

Nscale is not a household name like the hyperscalers it competes with. Founded in 2024, the company builds data centers and rents out the GPU capacity inside them — the raw computing power AI labs need to train and run large models — rather than building AI products of its own. It has grown quickly on that model: a $155 million Series A in December 2024 was followed by a $1.1 billion Series B in March 2026, led by investment fund Aker with Nvidia also participating, a round Nscale called ‘the largest Series B in European history.’ The new pre-IPO round would dwarf both: up to $1.5 billion in convertible notes, a form of debt that later converts into equity, led by hedge fund Third Point, plus about $2 billion directly from Nvidia, with Goldman Sachs advising on the raise.

Why AI compute demand justifies the number

The scale of the financing traces back to one deal. On August 26, 2026, Anthropic agreed to pay Nscale $45 billion for computing power, an arrangement that now anchors the company's planned flotation. Nscale is telling prospective investors that its total contracted revenue backlog has reached roughly $103 billion, up from about $51 billion just four weeks earlier, according to The Next Web's reporting on the Bloomberg story. That capacity had reportedly been offered elsewhere first: Semafor reported that Microsoft walked away from a similar arrangement over the summer during a review of its own data center portfolio, and that Google passed after examining its own capital spending plans. Both companies run some of the largest AI data center buildouts in the world, which underlines how tight GPU and data-center capacity has become even for the deepest-pocketed buyers.

• $1.5 billion in convertible notes led by hedge fund Third Point, priced at a double-digit discount to the IPO price that narrows as the valuation rises and stops adjusting at $30 billion
• About $2 billion directly from Nvidia, which also backed Nscale's $1.1 billion Series B in March 2026
• $103 billion in total contracted backlog, up from roughly $51 billion four weeks earlier
• $45 billion of that backlog from a single deal with Anthropic, signed August 26, 2026
• $790 million in committed debt from Nordic banks and Norway's export-credit agency Eksfin for Nscale's Narvik data center
• About 194,000 Nvidia Vera Rubin GPUs under contract

The gap between that backlog and Nscale's actual revenue is stark. The company is briefing investors that it could eventually produce annual revenue of about $18.1 billion and adjusted earnings of around $13.6 billion, figures it describes as ‘illustrative’ rather than formal guidance. By comparison, Nscale booked roughly $33 million in revenue across all of 2025, and its most recent quarter passed $100 million, according to The Next Web. The company's board includes former Meta executive Sheryl Sandberg and former UK deputy prime minister Nick Clegg.

The wider race for data centers, GPUs and energy

Nscale's fundraising sits inside a broader scramble among AI infrastructure providers to lock down chips, buildings and power at the same time. Its data center in Narvik, in northern Norway, has drawn $790 million of committed debt from ABN AMRO, DNB, Nordea, SEB and Eksfin, Norway's state export finance agency — financing tied to a site chosen specifically for its cold air and access to Nordic hydropower, with Microsoft as the customer there. Nvidia's roughly $2 billion contribution to the pre-IPO round would sit alongside Nscale's order for about 194,000 of Nvidia's own Vera Rubin GPUs, meaning the chipmaker would help finance a company that is, in large part, buying its chips — a financing pattern that has drawn scrutiny elsewhere in the AI infrastructure sector as investors try to distinguish genuine end-user demand from vendor-financed demand.

For companies that depend on third-party AI cloud compute rather than building their own data centers, Nscale's raise is a signal worth watching. It shows how concentrated the supply of AI computing power has become: a two-year-old company can now command a $45 billion contract from a single AI lab and a $2 billion check from the world's dominant chipmaker, while hyperscalers with far deeper resources decline similar deals over cost. That concentration cuts two ways for buyers. More capital flowing into GPU-dependent providers like Nscale should eventually mean more available compute and, potentially, downward pressure on prices as capacity expands. But a company whose contracted backlog vastly exceeds its current revenue, and whose largest financial backer is also its main chip supplier, carries execution and financing risk that a customer locked into a long-term compute contract inherits by association. Businesses that rely on a single external AI infrastructure provider should treat this moment as a reason to check contract terms for capacity guarantees and pricing protections, and to weigh whether depending on one still-unprofitable, IPO-bound supplier for critical compute is a risk worth carrying alone.