Why NVIDIA Is Shifting Its Focus To AI Instead Of Cloud Services

Explore why NVIDIA is shifting its focus to AI over cloud services and how the decision reshapes its strategy and market position

NVIDIA is stepping back from the battle with hyperscale clouds and is shifting its position to DGX Cloud as internal AI infrastructure, not a publicly accessible cloud competitor. 

This move is intended to protect its top GPU franchise and to strengthen its relationships with AWS, Google Cloud, and Microsoft rather than threatening them.

Why DGX Cloud Is Shifting Focus

DGX Cloud launched as a GPU cloud service that is built on the hyperscaler data centers, aimed at renting NVIDIA-powered clusters directly to businesses. 

Two years later, the business is changing its focus so it can use the same technology that is utilized by NVIDIA’s own researchers, who use it to aid in designing as well as AI modeling, instead of chasing huge numbers of customers from outside.

Internal sources mentioned in numerous reports claim that the service was struggling to draw enough interest, particularly with a high price against the mainstream cloud services. 

Strategically, NVIDIA concluded that the greater prize is in securing an extended period of GPU usage and influence in all clouds, not creating a full-scale, parallel cloud platform that it has developed on its own.

Leadership Changes And Internal Alignment

The reorganization will include a change in leadership. According to reports, the chief of the business division cloud Alexis Black Bjorlin (joined NVIDIA from Meta in 2021), has been reassigned and is looking for a new position within the company, as the DGX Cloud’s mission is trimmed.

Moving forward in the future, it is expected that the DGX Cloud team will act more as one inside “platform group,” providing high-performance GPU clusters to NVIDIA engineers working on:

• New GPU systems and designs, as well as new GPU architectures
• Foundation, as well as free-source AI models that are designed to run at optimum speed on NVIDIA hardware
• Software stacks, such as CUDA networks, CUDA, as well as orchestration tools

This tighter alignment allows for easier design systems, chips and models into an integrated stack rather than trying to handle each external task directly.

Why Competing Was Risky

Jensen Huang has long had to find a way to maintain a balance between competition and cooperation with hyperscalers. Cloud providers both:

• The GPUs account for a large percentage of NVIDIA’s profits by purchasing GPUs in large quantities
• They are currently developing their own AI accelerations (like AWS Trainium/Inferentia, Google TPU, along with Microsoft Athena), which could reduce NVIDIA’s market share of the market over time.

According to reports, Huang was wary of expanding DGX Cloud into a full AWS-like rival due to the risk of:

• Afflicting pressure on key cloud partners who had already recognized DGX Cloud as a potential threat to their business
• The company is accelerating its move into proprietary chips to lessen dependence on NVIDIA
• Inflicting a loss on GPU revenue if the cloud decides to prefer its own silicon to counter

Instead of expanding the model that scared off customers and resulted in a limited amount of uptake, NVIDIA is choosing to remain the official “arms dealer” of the AI age, and has a roughly 80% part in the AI accelerator market that is used within cloud data centers.

Technical And Commercial Limitations Of DGX Cloud

Beyond politics with partners, DGX Cloud faced practical challenges in the market:

• Multi-cloud limitations: Since it was dependent on specific hardware footprints in various providers, it was unable to provide a truly integrated, cross-cloud experience. This made multi-cloud systems difficult to offer to customers.
• Prices and the overlap of services: Businesses can already lease NVIDIA GPUs via AWS, Azure, and Google Cloud. DGX Cloud, layered on top, was often more expensive and only compelling for niche, performance-sensitive workloads.
• Go-to-market difficulty: Offering the DGX Cloud as a top GPU cloud, while other partners offered similar infrastructure, caused channel conflicts and confused customers about the best time to select DGX Cloud over the native hyperscaler.

All in all, these problems led to DGX Cloud a strategically awkward business model to scale up compared to NVIDIA’s main strategy of distributing chips and systems into the cloud of everyone else.

NVIDIA’s New Cloud Strategy

NVIDIA isn’t leaving the cloud entirely, but is shifting its function. It has committed millions of dollars to long-term contracts for compute capacity leased from the top cloud providers, thereby being one of the largest single customers. 

A recent report outlined plans to invest between $26 and $30 billion in the coming few years to host servers by hyperscalers.

This capacity will be utilized to:

• The internal R&D department of NVIDIA is powered by model training at a large scale
• Help DGX Cloud for use as a service for partners instead of a complete retail cloud
• Create co-branded AI services where cloud providers offer services that are optimized for NVIDIA hardware

A spokesperson for the company has stated that NVIDIA is planning to “continue to invest in DGX Cloud” to aid internal research as well as to make available specific capabilities to partners instead of developing a stand-alone AWS competitor.

The Strategic Logic

In resigning itself away from cloud-based direct competition, and taking a stance to be the primary AI both in the software and hardware layers, NVIDIA is:

• Secure its GPU moat: keeping hyperscalers in high-volume, enthusiastic customers, rather than adversaries, boosts NVIDIA’s capacity to launch new chips and systems to all major clouds.
• Reduced risk to strategic plans: It decreases the motivation of AWS, Microsoft, and Google to invest more in home-grown accelerators to stay clear of an integrated vertical NVIDIA competitor.
• Focusing: This allows the company to concentrate its efforts on chip design, systems, and the CUDA/AI software platform, which are where its margins and defensibility are the highest.

Final Thought

DGX Cloud, in this new capacity, will become an internal engine and partner-enabling platform. It is not an all-out effort to create the next superscale cloud. 

A company that is already leading its share of the AI hardware stack, this is a standard “protect the core” move, which is to sacrifice a tempting, however risky source of revenue to protect the position that everything else relies on.