Joining forces with Google to support CoreWeave, NVIDIA is fully committed to building the "AI Cloud".

In 2023, NVIDIA encountered generative AI and was recognized as the biggest "gold digger" in the global AI boom. Huang Renxun has repeatedly referred to this moment as the "iPhone moment" of AI. Now, Huang Renxun has set his sights on an opportunity to become a hunter rather than prey - the cloud.

On August 31st, it was reported that CoreWeave, a computing power newcomer "supported" by NVIDIA, is becoming a hot commodity. Currently, they are working with advisors to sell 10% of the company's shares, with a valuation estimated to reach between 5 billion and 8 billion US dollars.

Until a few months ago, CoreWeave was relatively unknown, but riding the wave of AI computing power, NVIDIA took notice of CoreWeave.

CoreWeave was founded in 2017, initially focusing on cryptocurrency, but in 2018 they expanded their projects to GPU acceleration. According to CoreWeave's official website, their services are now 80% cheaper than traditional cloud computing providers.

According to media reports, in order to support CoreWeave, NVIDIA allocated a large number of new cards to CoreWeave, even in the face of an H100 shortage, and directly participated in the investment. In April of this year, NVIDIA became a major participant in CoreWeave's $421 million Series B financing, raising CoreWeave's valuation to $2 billion.

In June of this year, CoreWeave announced that in order to build 10 new data centers and acquire more NVIDIA GPUs by the end of the year, they decided to mortgage their NVIDIA H100 GPUs to obtain a debt financing of $2.3 billion (approximately 16.5 billion RMB).

Supporting CoreWeave in the cloud computing field is just part of NVIDIA's deployment in the AI cloud domain.

On August 28th, at the Google Cloud Next event, NVIDIA CEO Huang Renxun made a surprise appearance and announced that the partnership between NVIDIA and Google Cloud will expand to advanced cloud computing. NVIDIA's stock price rose 4.2% to a new all-time high.

Through this collaboration, Google's PaxML large-scale language model framework can now immediately leverage the performance of NVIDIA's H100 and A100 GPUs. In addition, Google Cloud will be one of the first companies to receive NVIDIA's next-generation DGX GH200 AI supercomputer, which is powered by the Grace Hopper superchip. This supercomputer was launched earlier this month by NVIDIA. Huang Renxun said, Google and NVIDIA are collaborating to reshape the generative AI cloud infrastructure:

Our expanded partnership with Google Cloud will help developers accelerate their work in infrastructure, software, and services, thereby improving energy efficiency and reducing costs.

NVIDIA's ambition in the AI cloud market is evident. In such a new generative AI market, the distribution of interests between NVIDIA and cloud service providers is constantly changing. NVIDIA's layout in the cloud business has already disrupted the cake of cloud service providers.

NVIDIA's Ambition in the "AI Cloud"

Amazon, Microsoft, and Google firmly control the cloud computing market, making it difficult to get a piece of the pie.

According to Statista data, in 2022, Amazon AWS has a market share of 32%, Microsoft Azure has 23%, and Google Cloud has 10%. The combined market share of these three giants reaches 65%.

Some analysts explain that the essence of cloud services is to virtualize the hardware resources in data centers and then lease them to the market. Traditional data center servers are almost 100% based on Intel and AMD CPUs, but CPUs are not good at large-scale data processing and parallel computing, while NVIDIA GPUs can make up for the shortcomings of CPUs in this regard.

In March of this year, at the GTC conference, NVIDIA officially launched the NVIDIA DGX Cloud AI cloud service and announced cooperation with multiple cloud service providers, allowing enterprises to host DGX Cloud infrastructure through cooperative cloud service providers without the need to purchase and own servers.

However, NVIDIA's choice is not without risks: on one hand, there is fierce competition among peers, with many companies striving to enter the AI track and seize development opportunities; on the other hand, this new business model of directly selling AI services to large companies and governments may cause conflicts between NVIDIA and its largest customers, large technology companies.

Perhaps Everything is Just Beginning

To some extent, NVIDIA's launch of DGX Cloud has formed a competitive relationship with service providers including Google and Microsoft, but NVIDIA still wants to maintain close cooperation with cloud service providers. Therefore, it has decided to host DGX Cloud on the cloud platforms of various cloud service providers.

Oracle is the first cloud service provider to announce cooperation with NVIDIA. In March, Oracle announced the strategic deployment of NVIDIA AI applications on its new Oracle Cloud Infrastructure (OCI).

On August 28th, Google Cloud and NVIDIA announced an expanded partnership to advance the development of AI computing, software, and services. Google Cloud will be one of the first companies to receive NVIDIA's next-generation DGX GH200 AI supercomputer, which is powered by the Grace Hopper superchip. NVIDIA DGX Cloud AI supercomputing and software will be directly provided to Google Cloud customers through the browser, allowing them to access the AI computing power of NVIDIA's supercomputer DGX. Huang Renxun said:

"We will deploy NVIDIA DGX Cloud in the Google Cloud platform. This is where we conduct AI research, optimize software stacks, and all the work we do will immediately benefit Google Cloud and all users of Google Cloud."

However, another cloud giant, Amazon AWS, has refused to cooperate with NVIDIA on the DGX Cloud product.

Dave Brown, Vice President of Amazon Elastic Compute Cloud, said in a media interview: "NVIDIA contacted us and we studied the business model. But for AWS, it doesn't make much sense." He believes that AWS has long-term experience in building reliable servers and has existing supply chain expertise.

AWS started purchasing NVIDIA's H100 chips in March of this year, but only as part of its self-developed system. AWS is also considering using AMD's latest AI chip, MI300, but has not made a final decision yet.

Analysts point out that in the face of an emerging generative AI market, the distribution of interests between NVIDIA and cloud service providers is constantly changing, but NVIDIA has already made a move on the cheese of cloud service providers.