Wallstreetcn
2023.09.12 01:46
portai
I'm PortAI, I can summarize articles.

Huang Renxun's "300 billion new market," Nvidia's cloud ambitions

NVIDIA's ambition extends beyond selling chips to include the cloud services market in the hands of its customers.

Global Enterprise Service Giant Oracle's New Quarterly Performance Slows Down More Than Expected, among a number of weak businesses, only AI-related businesses have exceeded expectations in growth. Cloud services and technology licensing support businesses have seen a 13% increase in revenue compared to the same period last year. So far, Oracle has received orders for the second-generation cloud capacity worth over $4 billion from AI development companies, which is twice the amount from the previous quarter.

Due to the weak performance in the new quarter, Oracle's overnight stock price fell more than 9% after the US market closed, but its increase since the beginning of the year still exceeds 51%. In order to seek a foothold in the global competition of generative AI, Oracle announced that it will spend billions of dollars this year to purchase NVIDIA chips to accelerate the expansion of cloud computing services.

As the king of AI hardware, NVIDIA has always been a key AI chip supplier for Oracle, Amazon AWS, Microsoft, and other cloud service platforms. However, with the increasing demand for NVIDIA chips from AI software development companies, the demand from cloud service providers quickly outstrips supply.

At this critical moment, NVIDIA sees a new $300 billion opportunity: leveraging its dominant position in hardware to sell cloud computing services and software to AI development companies.

Last year, NVIDIA proposed to these major customers to rent NVIDIA chip-supported servers in their data centers, so that it can lease these servers to AI software development companies. According to the latest reports from the media, Microsoft, Google, and Oracle have agreed to NVIDIA's proposal, while AWS has not.

Analysts believe that this model of obtaining cloud software revenue by establishing direct relationships with enterprise chip users may not necessarily take away the existing revenue of cloud providers, because NVIDIA pays to rent servers from them and then charges a premium to its own customers. However, this may weaken the influence of cloud providers and help NVIDIA sell its own AI software to AWS, Microsoft, Google, and Oracle customers.

In response, Joshua Bernstein, former manager of AWS and Google Cloud, said:

It puts NVIDIA's brand above the cloud provider's brand. This shows that NVIDIA is aware of its position in the market and what measures it needs to take to maintain this position.

NVIDIA's cloud service DGX Cloud is the latest evidence that its ambitions no longer stop at selling chips. Recently, NVIDIA has also started selling AI software separately to customers such as Adobe, Getty Images, and Shutterstock, who say they are using it to build AI models. In addition, NVIDIA can leverage its relationship with DGX Cloud customers to sell more AI software of this kind.

$300 Billion Opportunity

In the quarterly earnings conference call held in August, when asked about this new cloud service, NVIDIA CEO Jensen Huang candidly stated that cloud providers "really like it." However, according to reports, spokespersons from the cloud providers who accepted the proposal declined to comment, and there was no promotion of the service on their official websites, except for a press release issued by Oracle in March. An AWS spokesperson also declined to comment.

Bernstein analyst Stacy Rasgon said:

I can completely understand why Amazon rejected (NVIDIA's cloud service), because ultimately, NVIDIA is the one with the customer relationships.

Rasgon stated that, in comparison, the cloud providers who agreed to launch DGX Cloud are smaller in scale than Amazon, so theoretically, they can compete with Amazon for market share in cloud computing.

Charlie Boyle, Vice President and General Manager of NVIDIA, refuted the notion that NVIDIA is competing with traditional cloud providers in an interview, stating that the new cloud service is "not about NVIDIA taking customers away, it's about NVIDIA and cloud providers sharing." When asked about Amazon's rejection of NVIDIA's service, he said that AWS has always been a "great partner... but every company has its own limitations."

NVIDIA's cloud computing service helps customers develop AI models and provides them with customized pre-trained models. These pre-trained models include Megatron 530B, which NVIDIA uses for text generation (similar to ChatGPT), and People Net, a visual model for identifying individuals in video clips.

Currently, NVIDIA's DGX cloud service has gained favor from some major buyers, such as IT software giant ServiceNow, biopharmaceutical company Amgen, and insurance company CCC Intelligence Solutions. Jensen Huang told analysts last month that DGX Cloud has achieved "tremendous success," but did not disclose specific details of its revenue.

Equally important, the cloud service enables NVIDIA to sell software for managing large datasets used in developing AI models, competing with products similar to Databricks, Snowflake, and Microsoft.

In the quarterly earnings report announced in August, NVIDIA informed investors that selling software to companies developing AI or VR applications could bring a potential revenue opportunity of $300 billion.

Chief Financial Officer Colette Kress told analysts during the quarterly earnings conference call in August that the software business generates hundreds of millions of dollars in revenue each year, which, although a fraction compared to the chip business, still has the potential for growth.

NVIDIA's Trojan Horse

However, for traditional cloud providers, DGX Cloud is like a Trojan horse that could turn them into intermediaries for NVIDIA. For example, ServiceNow uses DGX Cloud to develop AI software, consolidating IT requests and providing support for chatbots. John Sigler, Senior Vice President of this IT software giant, stated that NVIDIA's services enable ServiceNow to easily run its new AI software in its own data centers and across multiple cloud providers simultaneously. It can utilize a "single software platform" and have NVIDIA manage the processes.

Therefore, ServiceNow does not need to keep track of which traditional cloud provider is providing servers for its AI development behind the scenes.

Sigler said:

If NVIDIA's cloud services use Microsoft or Google's data centers, that's completely fine.

According to The Information's cloud database, ServiceNow spends at least $75 million annually on Microsoft's cloud services.

From NVIDIA's perspective, it is fair to launch this type of cloud service within traditional cloud providers' data centers. Although NVIDIA and cloud providers need each other, this relationship is becoming more complex and controversial. AWS, Microsoft, and Google have been selling or developing their own AI server chips to reduce their reliance on NVIDIA's chips, so NVIDIA naturally sees them as competitors.