Nvidia AI platform for cloud GPU providers widens supply
The vendor is addressing the AI chip supply and demand challenge while also helping some lower-tier cloud GPU providers that are facing challenges in the GPU-as-a-service market.
Nvidia's new GPU marketplace marks a strategic direction for the AI chip giant as it looks to fulfill worldwide market demand for lower-cost AI compute power.
The vendor introduced DGX Cloud Lepton Monday at Computex, a tech trade show in Taiwan. The platform connects developers who need compute resources to build agentic and physical AI applications with tens of thousands of available GPUs, Nvidia said.
The vendor said Nvidia Cloud Partners such as CoreWeave, Lambda and Yotta Data Services will offer Nvidia Blackwell, Nvidia's most powerful AI chip, and its other advanced GPUs on the DGX Lepton marketplace.
Addressing challenges
This launch reflects some challenges in the AI hardware market and for vendors that are trying to provide cloud-based GPUs.
Not just Nvidia, but Nvidia's supply chain can barely keep up with the demand, and demand far exceeds supply.
Chirag DekateAnalyst, Gartner
"Not just Nvidia, but Nvidia's supply chain can barely keep up with the demand, and demand far exceeds supply," Dekate said.
The lack of supply has been fruitful for GPU cloud providers.
In the past few years, two types of GPU cloud providers have popped up in the marketplace.
Leading GPU-as-a-service companies such as CoreWeave, Crusoe and Lambda that boast extreme scale and strategic partnerships with the cloud provider giants, Dekate said.
On the other hand, some emerging GPU-as-a-service companies struggle to find customers because the service is not cheap. These vendors must shoulder the cost of deploying the GPUs and data centers and manage them all.
This year, Nvidia began switching from its Hopper chip to Blackwell and is now readying to replace Blackwell with Rubin.
"You are now seeing GPU-as-a-service companies struggle," Dekate said. "[They] are struggling because every generation switch not only comes with capability. With that comes not just the additive cost structure of acquiring, procuring and applying the systems, but also the problem of trying to support and power many of these ecosystems."
He added that moving from one GPU generation to another triggers a need for more power.
While the leading GPU-as-a-service providers can deal with these costs, emerging providers can't.
This is where DGX Cloud Lepton comes in, Dekate said.
"For end users that want access to GPUs and GPU-as-a-service to spin up AI projects or products, I think DGX Cloud Lepton enables them to discover and, more importantly, abstract away the complexity of building these clusters themselves," he said. "They can easily access GPUs through these providers in a simplified manner."
The strategy shows how Nvidia is trying to help smaller GPU service companies navigate the market changes.
"It is widening the pool of providers," said Nick Patience, an analyst at The Futurum Group.
Moreover, in a world in which AI factories are emerging, this product could provide the layer needed to "power some of the AI products of the future," Dekate said.
Developers and lock-in
Patience said the platform also gives Nvidia a more direct relationship with developers.
While Nvidia has its large developer community, the GPU-as-a-service cloud providers also have their own developer communities.
"This kind of inserts Nvidia into that community," Patience continued. "It's a direct developer engagement, which can only be good for the kind of stickiness of Nvidia and its GPUs and the entire software stack."
Even more so, the move could lead to Nvidia vendor lock-in more than ever.
"This bringing together in a single interface and a single customer reference point means Nvidia, in theory, knows more about what the customers want than they did before," Patience said. "It also tightens that relationship that they have with developers."
However, while this approach appears to be a stopgap measure to deal with Nvidia's supply-demand problem and the challenges some GPU cloud providers face, it might not be convincing enough for enterprises serious about using AI technology on a large scale.
Instead, those enterprises might use an AI factory powered by AWS, Azure or Google or build something on-premises.
"If you are a serious user, chances are that you're defaulting to one of these routes or creating a hybrid cloud strategy," Dekate said.
Esther Shittu is an Informa TechTarget news writer and podcast host covering AI software and systems.
