Nvidia Restructures Cloud Team After Retreating From AWS Competition
The Takeaway
- Nvidia reorganizes cloud division, effectively ending direct competition with AWS.
- DGX Cloud team shifts focus to internal Nvidia engineering needs.
- Cloud division head Alexis Black Bjorlin seeks new internal role.
More than two years after Nvidia CEO Jensen Huang shared his vision to develop a cloud service that could rival Amazon Web Services, he has backed away from competing with such firms and last week reorganized his company’s cloud team, according to several people with knowledge of the organizational change and an internal memo about it.
The restructuring, shared with some employees in a memo last week, reassigned the head of Nvidia’s cloud division and several other related executives, while others departed, these people said.
One of the executives, Alexis Black Bjorlin, who joined Nvidia from Meta in 2023 and reports to Huang, plans to take a new role within the company, according to the memo. Nvidia is merging the cloud team, which consists of several hundred people, with Nvidia’s engineering and operations organization, run by Dwight Diercks, a senior vice president of software engineering who reports to Huang.
The cloud team, known as DGX Cloud, will primarily serve Nvidia engineers’ demand for Nvidia chips, which they use to develop open source AI models, and will no longer focus on selling the cloud service to external enterprise customers, according to the memo and people who have worked in the division.
As part of the reorg, a newer Nvidia cloud service, DGX Cloud Lepton, is also moving into the engineering team. It allows cloud providers to list unused Nvidia server capacity in a marketplace run by Nvidia, but it didn’t get off to a fast start.
Huang unveiled the original DGX Cloud service at its flagship annual conference for developers and customers in March 2023. The effort aimed to create a new source of revenue and help the company forge direct ties with AI developers that rent Nvidia chips from cloud providers like AWS, Google and Microsoft.
Nvidia’s selling point was that the chips would perform better through DGX Cloud, compared to the way cloud providers configured the chips.
In reality, Nvidia was worried that as Google, Microsoft and Amazon each developed their own AI chips, they would lessen reliance on Nvidia’s and push customers to those alternatives. Gaining direct relationships with AI developers was Nvidia’s hedge on that risk.
The plan had a lot of potential, at least on paper, and Nvidia touted early customers such as ServiceNow, SAP and Amdocs. To create its cloud service, Nvidia rented servers from major cloud providers and customized them to its own stringent specifications before renting them out to AI developers.
But the DGX team struggled to attract customers, according to multiple people who worked in the division. And it was hard for Nvidia to provide customers with troubleshooting support given that DGX Cloud actually ran in the data centers of different cloud providers such as AWS, meaning that making a fix to one facility might not work in facilities owned by other firms, one of the people said.
Huang also has been reluctant to grow the business and risk irking cloud providers that are among Nvidia’s biggest chip customers, one of these people said. And Nvidia has taken numerous steps to financially back young cloud providers such as CoreWeave and Lambda, whose businesses effectively competed with DGX Cloud.
Nvidia earlier this year stepped back from the nascent cloud effort, which it previously told investors could generate $150 billion in revenue—more than AWS generates annually.
Nvidia’s overwhelming dominance in AI chip sales has persisted even as Google and others have tried to take some share. AWS has heavily discounted its Trainium AI chips and has been talking to OpenAI about a deal in which the AI developer, which primarily uses Nvidia chips, will use Trainium chips too. And Meta Platforms is considering spending billions of dollars on Google’s AI chips, known as tensor processing units.
In another example of how intertwined these technology firms are, Nvidia has become one of the biggest renters of Nvidia-powered servers purchased by cloud providers such as AWS and Google. While it tried to rent out some of those servers to DGX Cloud customers, Nvidia also uses the servers to develop many types of AI models, including for robotics and self-driving technology.
It said it plans to spend $26 billion renting such servers in the coming years.
“We continue to invest in DGX Cloud to deliver world class infrastructure for our cutting-edge research and development, and to provide our cloud partners with the software capabilities to succeed,” an Nvidia spokesperson said. “[Our] goal has always been to pilot and nurture DGX Cloud as a way to learn how to build systems better for ecosystem partners, that is not changing.”