On December 6th, Caixin News Agency reported that the H100 GPU is considered the "new gold" in the technology industry because it is the preferred choice for providing computing power for generative artificial intelligence.
Which companies did Nvidia's best-selling H100 GPUs go to this year? The following image clearly lists Nvidia's largest customers this year.
According to the latest report from market research firm Omdia Research, Meta and Microsoft both ranked first with a purchase volume of 150000 H100 GPUs.
It is not surprising that Microsoft has such a large order volume, as the company needs to provide computing power for its growing artificial intelligence products and existing artificial intelligence assistant Copilot (formerly Bing Chat).
Meta also has ambitions in the field of artificial intelligence. A few months ago, there were reports that Meta was continuously purchasing AI chips and creating data centers with the aim of developing a chatbot as powerful as ChatGPT. It is reported that Meta is developing a new large-scale language model, which is expected to be several times more powerful than the Llama 2 model launched in July this year, and aims to start training the new model next year.
However, there is a slight disparity in purchase volume between other companies and Meta and Microsoft.
Google, Amazon, Oracle, and Tencent jointly ranked third with a purchase volume of 50000 H100 GPUs; The estimated purchase quantity of cloud service provider CoreWeave is 40000 H100 yuan; Baidu and Alibaba respectively purchased 30000 and 25000 H100 yuan; ByteDance and cloud service provider Lambda Labs followed with the purchase of 20000 H100; Tesla purchased 15000 H100 yuan.
Most of them have gone to cloud providers
The report from Omdia Research also shows that Nvidia supplies most of its GPU servers to large cloud service providers. This may also mean that Nvidia is moving towards cloud services.
It is worth mentioning that among NVIDIA's top 12 clients in the H100, only CoreWeave and Lambda Labs are startups, but they have a strong connection with NVIDIA.
As early as the launch of the H100 chip this year, Nvidia chose CoreWeave and Lambda as the first companies to use the chip. And in their financing process, Nvidia's investment presence can be found: in April this year, Nvidia participated in the $221 million Series B financing of CoreWeave; In July this year, Nvidia invested $300 million in Lambda Labs.
Currently, CoreWeave is valued at approximately $2 billion, while Lambda's valuation is expected to exceed $1 billion.
Previously, analysts pointed out that Nvidia sold its GPUs to these cloud startups in order to expand the target customer base of GPUs, no longer limited to cloud giants such as Amazon, Microsoft, and Google.
Future outlook
In addition, new data recently released by Omdia shows that chip manufacturing giant Nvidia shipped nearly 500000 H100 and A100 GPUs in the third quarter. Previous financial reports showed that Nvidia's data center revenue for the quarter was $14.5 billion, almost four times that of the same period last year.
The analysis company also predicts that by the fourth quarter of 2023, sales of Nvidia H100 and A100 GPUs will exceed 500000 units.
However, almost all companies that purchase Nvidia H100 GPUs in large quantities are also developing custom chips for AI, HPC, and video workloads themselves. Therefore, over time, as Nvidia's customers turn to their own chips, their purchases of Nvidia hardware may decrease.