According to people familiar with the matter, NVIDIA is in talks with the emerging artificial intelligence cloud service provider Nscale and intends to provide them with a debt credit guarantee of several billion dollars for the development of a 460-megawatt data center; this data center is planned to be rented by the artificial intelligence company Anthropic.
According to sources, the deal has not yet been finalized, and the specific terms may still change. However, this once again indicates that NVIDIA continues to provide financial support to so-called “new cloud companies” such as Nscale, CoreWeave, and Firmus. These companies have incurred significant debt while financing data centers dedicated to artificial intelligence and still need to raise more funds, hence they are under pressure from investors.
NVIDIA has provided support to these “new cloud companies” and other large customers through approximately $170 billion in credit support and investments. Last week, there were reports that NVIDIA was also in talks to invest hundreds of millions of dollars in the Australian artificial intelligence cloud service provider Firmus, which suddenly halted its initial public offering (IPO) plans at the beginning of this month.
NVIDIA's credit support provided for the Nscale facility located in West Virginia is primarily used for the power supply and construction of the facility. This is different from the large-scale credit support NVIDIA provides to other customers, which usually covers the rental costs of data centers or financing for chip purchases.
As disclosed in the IPO prospectus submitted last month by Nscale, the company signed a contract worth $45 billion with Anthropic to lease its first facility with a capacity of 460 megawatts (MW) within the Monarch data center park located in West Virginia; this facility will be equipped with NVIDIA's latest Vera Rubin graphics processors (GPU). According to industry estimates, the cost of constructing a data center with a power capacity of 1 gigawatt (GW) is approximately $12 billion. Based on this, the credit support provided by NVIDIA could exceed $5 billion.
According to one of the individuals familiar with the matter, the agreement will be structured in the form of a “residual value guarantee” specifically for the data center itself, rather than being tied to lease contracts or chips. This will enable Nscale to raise investment-grade financing for the project construction. The individual stated that if Nscale defaults on the payment of construction funds, NVIDIA will step in to make the payments and temporarily take over the facility; during this period, NVIDIA will also assist in arranging the sale and re-lease of the site.
According to another informed source, Anthropic also plans to lease its second-phase facilities located in a park in West Virginia under a contract of roughly comparable scale.
An investor in the field speculates that the reason NVIDIA is keen to ensure the success of Nscale is that this startup company has a connection with Anthropic, and Anthropic is one of the largest users of artificial intelligence (AI) server chips. Unlike most of its peers, Anthropic has adopted a diversified strategy, using products from multiple chip suppliers rather than relying primarily on NVIDIA.
Many "new cloud companies," including Nscale, have not yet been tested by the market; they need to achieve explosive growth in revenue in order to repay the debts incurred from purchasing a large number of servers (mainly NVIDIA servers). As interest rates rise, this problem becomes increasingly severe.
For NVIDIA, these “new cloud companies” are of paramount importance, as several of its key customers—Google, Microsoft, and Amazon—are developing their own server chips in order to reduce their dependence on NVIDIA’s products. This summer, NVIDIA even adjusted its revenue reporting methodology to further highlight the significance of these “new cloud companies.” NVIDIA began to strongly support such enterprises starting in 2023, which coincided with a surge in demand for GPU.
Whether NVIDIA will continue to support this ecosystem may determine the survival of certain “new cloud” companies, and it could also trigger integration within the industry – some industry executives believe that such integration is inevitable.
NVIDIA seems eager to prove to those who are skeptical that they are wrong. To this end, NVIDIA has provided various forms of support to these “new cloud companies,” including equity investments, chip quotas, credit support, and the connection of customer resources. According to executives from NVIDIA and these cloud service providers, a team within NVIDIA (led jointly by Jensen Huang’s son-in-law, Nico Caprez) has even begun to assist these cloud service providers in finding locations that meet power supply requirements. Such locations are very difficult to find, as either seeking an independent power source or connecting to the public grid often takes a considerable amount of time.
