NVIDIA CEO Jensen Huang has abandoned plans to directly compete with Amazon Web Services (AWS) in cloud services, opting instead to restructure the company’s cloud team last week. This marks a strategic pivot over two years after Huang initially envisioned challenging AWS in the cloud market.
The reorganization was revealed through an internal memo and confirmed by multiple sources familiar with the changes. According to these sources, the restructuring reassigned roles for NVIDIA’s cloud business leaders and several executives, with some departing the company.
Alexis Black Bjorlin, a senior executive who joined NVIDIA from Meta in 2023 and reported directly to Huang, is now seeking a new internal role. The cloud team, comprising hundreds of employees, has been merged into the Engineering and Operations division led by Senior VP of Software Engineering Dwight Diercks, who also reports to Huang.
The memo and former team members disclosed that the cloud unit, known as DGX Cloud, will now primarily support NVIDIA engineers in developing open-source AI models using the company’s chips, rather than focusing on selling cloud services to external enterprise clients.
As part of the restructuring, DGX Cloud Lepton—a newer service enabling cloud providers to list unused NVIDIA server capacity—will also be absorbed into the engineering team. The initiative has faced early challenges since its launch.
Huang first unveiled DGX Cloud at NVIDIA’s annual developer conference in March 2023, aiming to diversify revenue streams and forge direct ties with AI developers who typically rent NVIDIA chips from AWS, Google, and Microsoft. NVIDIA pitched DGX Cloud as offering superior chip performance compared to traditional cloud providers.
However, NVIDIA grew concerned that major cloud players like Google, Microsoft, and AWS—all developing in-house AI chips—might reduce reliance on NVIDIA hardware and steer customers toward alternatives. Direct engagement with AI developers was seen as a hedge against this risk.
On paper, the plan held significant potential, with early adopters including ServiceNow, SAP, and Amdocs. NVIDIA leased servers from large cloud providers, customized them to stringent standards, and subleased them to AI developers.
**Competitive Conflicts and Challenges** Former team members revealed DGX Cloud struggled to attract sufficient customers. One issue was troubleshooting across multiple cloud data centers (e.g., AWS), where fixes in one environment didn’t always apply elsewhere.
Huang reportedly hesitated to aggressively expand the service to avoid alienating cloud providers that are NVIDIA’s largest chip buyers. Meanwhile, NVIDIA funded emerging cloud rivals like CoreWeave and Lambda, which indirectly competed with DGX Cloud.
Earlier this year, NVIDIA began scaling back its nascent cloud business, despite previously projecting $150 billion in future revenue—surpassing AWS’s current annual sales.
NVIDIA maintains dominance in AI chip sales, even as rivals like AWS slash prices for their Trainium chips and court clients like OpenAI. Meta Platforms is also considering multibillion-dollar purchases of Google’s Tensor Processing Units (TPUs).
In this complex ecosystem, NVIDIA itself has become one of the largest lessees of its own servers—procured through AWS and Google—using them both for DGX Cloud and in-house AI model development, including robotics and autonomous driving projects. The company plans to invest $26 billion in server leases over the coming years.
A NVIDIA spokesperson stated: *“We continue investing in DGX Cloud to deliver cutting-edge infrastructure for R&D and empower cloud partners with essential software capabilities. DGX Cloud remains a pilot project to refine system-building for ecosystem partners—a goal unchanged.”*
Comments