Report: NVIDIA is scaling back its emerging cloud computing business and shifting its focus to internal research and development

On Friday, The Information reported that NVIDIA is gradually scaling back its nascent cloud computing business. According to insiders, NVIDIA has reduced its efforts to attract enterprise customers to use its DGX Cloud services and plans to primarily use the service for itself, including supporting internal researchers.

This strategic adjustment indicates that NVIDIA is facing limited demand in the cloud services sector. NVIDIA had previously committed to spending $13 billion to rent back its AI chips from major cloud service providers and planned to achieve a revenue target of $150 billion through its leasing business.

The contraction of NVIDIA's cloud business may reflect market resistance to its pricing strategy. According to individuals who assisted multiple companies in negotiating cloud services, AI developers are resistant to the high prices of DGX Cloud servers, which are typically higher than those of traditional cloud service providers.

This business had positioned it in a quasi-competitive relationship with Amazon Web Services, and this change also alleviates the competitive pressure between NVIDIA and its largest customers, particularly the major cloud service providers that account for half of its revenue.

Shift in Focus to Internal R&D

According to individuals with direct knowledge of the situation, NVIDIA is now using most of the server capacity of DGX Cloud for internal research and development, with employees utilizing these servers for various tasks ranging from designing new chips to developing AI models.

Alexis Black Bjorlin, head of NVIDIA DGX Cloud, denied the claim of a strategic shift in an interview. She stated, "Our internal researchers need a lot of computing resources, and so do our customers, so our strategy remains the same."

Nevertheless, NVIDIA no longer specifies in its latest quarterly earnings report that its cloud spending commitments are partially allocated to DGX Cloud, despite having made such statements in previous quarters. This disclosure change suggests that NVIDIA is no longer prioritizing providing the service to external customers.

Cloud Ambitions Encounter Real-World Challenges

CEO Jensen Huang had ambitiously outlined the prospects of DGX Cloud at the beginning of 2023, hoping that large enterprises such as banks and pharmaceutical companies would rent AI chip servers directly from NVIDIA, just as they would from Amazon and Microsoft.

In the earnings call at the beginning of 2023, Jensen Huang stated, "We will democratize access to this infrastructure, truly making this technology and capability easy to obtain." He hoped that the service would alleviate the shortage of NVIDIA chip servers faced by enterprise customers at that time.

By August 2023, Huang told investors that DGX Cloud was a "huge success." By the end of 2024, CFO Colette Kress stated that the company's software business (which she indicated includes DGX Cloud) had an annualized revenue of $2 billion.

Seeking Breakthroughs with New Service Models

NVIDIA still hopes that some enterprise customers will rent GPUs directly from it. This summer, NVIDIA launched another cloud service, DGX Cloud Lepton, which is a "marketplace" where companies can access GPUs through a network of cloud service providers Unlike the original DGX Cloud, cloud service providers on the Lepton platform can independently choose to list available GPU server capacity, rather than signing multi-year contracts with NVIDIA to lease capacity and then sublease it to customers.

According to previous reports, the initial development of this market platform has been slow. Executives from several cloud service providers that have launched on NVIDIA's Lepton platform stated that they believe this new service competes with their own business, as it allows chip designers to establish direct relationships with their customers