More than two years after unveiling its vision to build a cloud service capable of competing with AWS, NVIDIA has made significant adjustments to its cloud business strategy. The company recently restructured its cloud team, integrating DGX Cloud into the engineering and operations system, with a focus on meeting internal R&D demands for NVIDIA chips rather than prioritizing the sale of cloud services to external enterprise customers.
Media reports indicate that, $NVIDIA (NVDA.US)$ more than two years after NVIDIA CEO Jensen Huang proposed the vision of building a cloud service capable of competing with $Amazon (AMZN.US)$ Amazon Web Services (AWS), he has now abandoned the idea of direct competition with these companies and reorganized the company’s cloud team last week.
Several individuals familiar with the organizational changes, as well as an internal memo, have revealed this shift. These sources told the media that in the restructuring memo shared with some employees last week, the head of NVIDIA’s cloud business unit and several related executives were reassigned, while others left the company.
According to a memo, one of the executives, Alexis Black Bjorlin, joined NVIDIA from $Meta Platforms (META.US)$ in 2023 and reported directly to Huang. She is currently seeking a new position within the company. NVIDIA has integrated its cloud team, which comprises hundreds of employees, into the engineering and operations organization led by Senior Vice President Dwight Diercks, who oversees software engineering and also reports directly to Huang.
The memo, along with several former members of the department, informed the media that this cloud team, known as DGX Cloud, will primarily serve NVIDIA engineers’ demand for NVIDIA chips used to develop open-source AI models, rather than focusing on selling cloud services to external enterprise clients.
As part of the restructuring, NVIDIA’s newer cloud service, DGX Cloud Lepton, will also be integrated into the engineering team. This service allows cloud providers to list unused NVIDIA server computing power in a marketplace operated by NVIDIA, but the project has faced challenges since its launch.
Jensen Huang first unveiled the DGX Cloud service at NVIDIA’s flagship annual conference for developers and customers in March 2023. The initiative aimed to create a new revenue stream and help the company establish direct connections with AI developers who rent NVIDIA chips from cloud providers like AWS, Google, and Microsoft.
NVIDIA’s selling point at the time was that using chips through DGX Cloud would deliver better performance compared to the configurations offered by the cloud providers themselves.
However, in practice, NVIDIA was concerned that as Google, Microsoft, and Amazon developed their own AI chips, they might reduce their reliance on NVIDIA chips and steer customers toward these alternatives. Establishing direct relationships with AI developers was viewed by NVIDIA as a crucial strategy to hedge against this risk.
This initiative, at least on paper, holds significant potential. NVIDIA had previously highlighted its early customers, including $ServiceNow (NOW.US)$ 、 $SAP SE (SAP.US)$ and $Amdocs (DOX.US)$ . To build its cloud services, NVIDIA leased servers from major cloud service providers, customized them according to its extremely stringent standards, and then rented them out to AI developers.
Conflicting Priorities in Competition
However, several individuals who previously worked in the department told the media that the DGX team has consistently struggled to attract sufficient customers. One of these individuals pointed out to the media that, since DGX Cloud actually operates in data centers run by different cloud service providers such as AWS, NVIDIA faces difficulties providing troubleshooting support to its clients. Fixes implemented in one cloud provider’s data center may not be applicable to another company’s facilities.
One individual told the media that Jensen Huang was also reluctant to significantly expand this business to avoid alienating cloud service providers, who are among NVIDIA's largest chip customers. Meanwhile, NVIDIA has taken several measures to provide financial support to emerging cloud service providers such as $CoreWeave (CRWV.US)$ and Lambda, whose businesses actually compete with DGX Cloud.
NVIDIA began scaling back from this still nascent cloud business earlier this year. The company had previously informed investors that this business could potentially generate $150 billion in revenue in the future, surpassing AWS's current annual revenue.
Despite $Alphabet-C (GOOG.US)$ As companies attempt to capture a portion of the market share, NVIDIA's absolute dominance in the AI chip sales sector remains firmly intact. AWS has significantly reduced the price of its Trainium AI chips and is in discussions with OpenAI on a collaboration plan that would see this leading AI developer, which primarily uses NVIDIA chips, also adopt Trainium chips. Additionally, Meta Platforms is considering spending billions of dollars to purchase Google’s AI chips, known as Tensor Processing Units (TPUs).
Amidst these highly interconnected relationships among tech companies, NVIDIA itself has become one of the largest renters of NVIDIA servers, which are procured by cloud service providers such as AWS and Google. On one hand, NVIDIA has attempted to lease some of these servers to DGX Cloud customers; on the other hand, it has utilized them to develop various AI models, including those related to robotics and autonomous driving technologies.
NVIDIA stated that it plans to invest $26 billion over the next few years to lease such servers.
A spokesperson for NVIDIA told the media:
"We will continue to invest in DGX Cloud to provide world-class infrastructure for cutting-edge R&D and offer our cloud partners the software capabilities they need to succeed. Our goal has always been to nurture DGX Cloud as a pilot project through which we learn how to better build systems for our ecosystem partners, and that has not changed."
Editor/Rocky