Nvidia’s Cloud Spending Soars, as It Looks Beyond Chips
Commitments more than doubled in three months earlier this year as It builds out its DGX Cloud service.
Nvidia has become the world's third-most-valuable company by selling chips to power artificial intelligence. Now it's also one of the biggest corporate buyers of cloud computing services.
Last week, Nvidia said it has committed to spend nearly $9 billion on cloud computing in the coming years, which involves spending with Amazon, Microsoft, Google and Oracle. That’s a huge jump from the $3.5 billion Nvidia said it had committed to cloud services as of January.
The Takeaway
• Nvidia has contracts to spend $8.8 billion on cloud services over the coming years
• The planned spending is to support R&D and Nvidia’s DGX Cloud service
• Nvidia wants to sell its own cloud services as well as chips
Nvidia didn’t disclose the lengths of its contracts, but two people familiar with contracts for cloud services said they typically span five to seven years. Even stretched out over seven years, Nvidia would be spending more than $1 billion annually, making it one of the top three spenders in The Information's Cloud Database.
It’s not clear why Nvidia’s planned spending increased so dramatically, but the company offered another hint with its quarterly financial results. Previously, Nvidia had said that its cloud spending was to support its internal research and development. But last week, Chief Financial Officer Colette Kress said in a statement accompanying the results that the cloud deals also would help power cloud services Nvidia sells under the name DGX Cloud.
DGX Cloud is part of Nvidia’s complicated relationships with the biggest cloud providers, Amazon, Microsoft, Google and Oracle. Through DGX Cloud, Nvidia rents servers with its chips located in those companies’ data centers. It then turns around and rents those servers to its own customers, including ServiceNow and Amgen.
S. “Soma” Somasegar, a managing director at Madrona Ventures, said Nvidia’s mention of DGX Cloud shows the company is “not happy being just a hardware play,” and wants to capture more of the revenue generated by companies that rent out its chips.
“Nvidia doesn't want you to buy [graphics processing units] and then spend all of your cloud money on AWS, [Google Cloud], and [Microsoft] Azure,” he said. So, Somasegar said, Nvidia is seeking some of the billions companies spend on services such as software that runs in the cloud. Nvidia has begun to sell AI-related software to customers including Adobe, Getty Images and Shutterstock, which say they are using it to build AI models.
DGX Cloud helps Nvidia get closer to some of its customers and guard against a future where its largest customers compete with it. Amazon, Google and Microsoft are working on their own AI chips that could lessen their dependence on Nvidia. Nvidia sells customers on DGX Cloud by promising better performance.
Nvidia's DGX Cloud service relies on AWS, Microsoft Azure, Google Cloud and Oracle, so it’s unclear which providers will benefit from Nvidia’s uptick in cloud spending. A spokesperson for Nvidia declined to provide more details beyond the company's public filings but said the cloud agreements are “long-term contracts over several years.”
The Information previously reported that Nvidia planned to rent servers containing around 16,000 H100s—one of its most advanced chips—to power DGX Cloud on AWS; that’s roughly as much computing power as a large AI startup might use. At Oracle, Nvidia is the second largest customer of servers that contain H100 chips, behind xAI, according to someone with direct knowledge.
Somasegar said the big cloud providers have similar symbiotic relationships with companies such as Snowflake, which spends hundreds of millions of dollars annually on cloud computing services and sells database software that runs on top of it. The cloud providers offer products that compete with Snowflake, but they’re also happy to work with the firm because it drives use of their servers, he said.