NVIDIA-backed British new cloud computing company Nscale, which boasts a contract commitment of tens of billions of dollars, is facing the same difficulties as its Australian counterpart Firmus from a month ago: The company's valuation is being requested by investors to be cut nearly in half, and Microsoft has announced its withdrawal from a key data center project.

According to Bloomberg, Nscale previously sought to go public with a valuation of around $30 billion, but a group of investors has clearly stated that a valuation of $16 billion to $19 billion is "more likely to be accepted by buyers" – which represents a reduction of over 40% from the original valuation. Meanwhile, according to Reuters citing reports from The Telegraph, Microsoft has terminated its cooperation agreement with Nscale's data center campus in Loughton, Essex, UK (Loughton), as officials from the UK National Grid informed Nscale that the site may not receive power supply until the 2030s at the earliest. Microsoft has refused to comment on this matter.

This is the fifth case within less than a month where a AI data center listing plan has encountered setbacks. The previous four cases involved SoftBank's SB Energy, Firmus, Philippines' PLDT subsidiary Vitro, and Blackstone's AirTrunk. Goldman Sachs' credit team attributed this series of events to "funding cost issues with the AI transactions," while the credit market had already issued warnings several weeks earlier through deteriorating bond prices and credit default swaps.

Microsoft withdraws: Power supply uncertain, customers advised to leave first

The failure of the Lorton project reflects the core contradiction in the current infrastructure construction of AI: contracts can be signed within a day, but the connection to the power grid may be a long time off.

According to Reuters, the direct trigger for Microsoft to terminate the agreement was a notification from the State Grid that the site would not be able to access power supply until the 2030s. The party identified as Nscale responded that their "relationship with Microsoft remains stable and long-term," characterized this issue as a "power grid problem within London," and stated that they are "exploring alternative energy solutions." However, with key customers having already departed, such a statement is unlikely to make up for the substantial commercial losses.

Microsoft has chosen to withdraw from third-party AI computing power projects more than once this year – according to previous reports by Bloomberg, in June of this year, Microsoft rejected a cloud computing contract worth $3 billion with Oracle. As the world's largest cloud service providers, which have the most abundant cash reserves, begin to be picky about renting computing power, the core business logic of new cloud computing companies that claim to provide “computing power faster than anyone else” is directly challenged.

The situation in Norway's flagship park is also not optimistic. The facility located in Glomfjord ( Glomfjord ), north of the Arctic Circle, Nscale, has experienced delays of six months due to issues such as cooling system leaks, local regulations, construction delays, and fire incidents. Out of the planned capacity of 30 megawatts, only 6.4 megawatts are currently in operation.

The Real Book Under Trillions of Contracts: A 4% Implementation Rate and a 107-Fold Valuation Bubble

The prospectus for Nscale boasts impressive figures at the top, but upon closer inspection, the data reveals a significant gap between what was actually achieved and what was promised.

The total value of contracts disclosed by the company amounts to a staggering $103.4 billion, a sharp expansion from the approximately $100 million at the time of its establishment two and a half years ago. However, the actual value of contracts in operation is only $2.6 billion; the capacity that has been put into use is 55 megawatts (of which 48 megawatts are leased managed capacity), while the committed capacity under the IT agreements is about 1,346 megawatts, resulting in an actual implementation rate of only around 4%. Microsoft (with a contract cap of $43.8 billion) and Anthropic (with a contract cap of $44.6 billion, both from the unfunded West Virginia campus) together account for about 85% of the order reserves. Moreover, the contract terms for Anthropic allow for an exit without additional obligations, and the relevant financing has not yet been secured.

Financially, Nscale reported revenue of $140.6 million in the first half of 2026, with a net loss of $1.02 billion, which means for every $1 in revenue generated, there was a loss of approximately $7. If we annualize the revenue from the first half of the year, about $280 million in annual revenue would correspond to a valuation of $3 billion, indicating a price-to-sales ratio of around 107 times. Even if priced in the range of $16 billion to $19 billion, which is recognized by investors, the price-to-sales ratio would still be between 57 and 68 times.

In addition, the British newspaper "Financial Times" revealed that Nscale downplayed its relationship with ByteDance in its prospectus – which contributed nearly three-quarters of its sales last year. Nscale submitted its prospectus on September 18th, and the mandatory waiting period has long since expired. However, according to Bloomberg, the company is still "processing comments from the U.S. Securities and Exchange Commission" and has not yet commenced a formal roadshow.

Successive setbacks: New cloud computing IPO platforms are being shut down one after another

The predicament of Nscale is not an isolated case, but rather a microcosm of the retreat of the financing wave for AI data centers over the past month.

This wave of setbacks began on September 22: SoftBank's SB Energy was originally planning to raise funds for the world's largest data center, but due to investors being deterred by its valuation of over $50 billion, IPO was forced to postpone its launch and has yet to initiate formal marketing. Subsequently, Firmus first cut the price of IPO by 25%, and then completely withdrew its listing plans; Vitro, a subsidiary of PLDT in the Philippines, postponed the listing of its data center REIT until 2027; even though Blackstone's AirTrunk increased its yield to about 6%, its $1.5 billion IPO in Singapore was also postponed until 2027.

The credit market has already sounded the alarm first. The credit default swap spreads for CoreWeave have soared to record levels, implying a default probability of about 63% within five years; the yield on its unsecured bonds in 2032 has reached 15%. Gartner predicts that over 80% of new cloud computing companies will cease independent operations before 2030. Goldman Sachs' credit team wrote in their weekend report:

"Capital has not disappeared; it has merely become more selective... The market is increasingly distinguishing between companies that can finance their growth on their own and those that rely on continuous access to financing for their growth."

Goldman Sachs' high-yield credit team also noted that the average price of financing bonds for sample high-revenue data center projects has dropped from $101.5 in June to $94.7 in October. "Construction risks, refinancing exposures, tenant guarantees, power supply, and approval permits" are becoming key considerations in credit underwriting. It is precisely these power supply and approval issues that directly led to Nscale losing Microsoft as a key client in Lorton.

Valuation repricing spreads to the stock market: Tech sector suffers consecutive sell-offs

The pressures in the credit market have fully been transmitted to equity assets. CoreWeave saw its market value evaporate by about a quarter last week, and Nebius has fallen by 32% in two weeks. According to a report from Goldman Sachs' global equity derivatives team, the credit spreads of hyperscale cloud service providers have widened to their widest level since 2025. Historically, when spreads were in similar ranges, the price levels of related stocks were about 15% lower than they are now.

Goldman Sachs experts in technology, media, and telecommunications Peter Callahan pointed out that the NASDAQ 100 index recorded its worst single-week performance since April 2025 last week, with a decline of about 550 basis points, with the semiconductor sector leading the drop by about 11.5%. The Philadelphia Semiconductor Index (SOX) has been declining for eight consecutive trading days, setting a new record for the longest consecutive decline since 2008; the valuation of the semiconductor sector has compressed from about 29 times the forward price-earnings ratio this summer to about 17 times. The basket of stocks favored by Goldman Sachs (AI) has fallen by 25% relative to the market high in June.

Next step: The private equity market may be the only way out.

For now, Nscale faces two paths. According to Bloomberg, discussions are still ongoing, and the details of IPO's plan may change; no final decision has been made yet.

One option is to accept a discounted valuation and list the company at a price of $16 billion to $19 billion—this is only slightly higher than the $14.6 billion valuation during the financing in March of this year by Series C. For the investors of Pre-IPO in September of this year, led by Third Point under Dan Loeb, with NVIDIA, Apollo, and Citadel also participating, injecting a total of $3.36 billion, this will be a much more disappointing outcome than initially anticipated. The second option is to follow the example of Firmus and turn to the private market—but even Blue Owl was unable to find third-party funding for a $4 billion data center project.

The larger market context cannot be ignored either. Bankers are facing a situation of "interwoven multiple risks," and the IPO market is seeing almost all AI projects freeze, except for Anthropic. As one of the two main pillars of Nscale's order reserves, Anthropic's own listing plans may also face delays. By then, Nscale will be competing with one of its largest customers for the funds of the same group of AI thematic investors—a rather delicate situation.

When the speed at which contracts worth hundreds of billions are signed far exceeds the speed at which grid connections are implemented, the three pillars supporting this wave of infrastructure construction frenzy for AI – procurement commitments from ultra-large-scale cloud service providers, low financing costs, and available power supply – are each being re-priced by the market.