Nvidia sold nearly 500000 A100 and H100 AI GPUs in Q3, with Meta and Microsoft becoming the largest buyers

On November 28th, it was reported that in the third quarter of this fiscal year, Nvidia earned $14.5 billion in data center hardware due to the high demand for AI GPUs.

As a leading global chip manufacturer, Nvidia has been committed to promoting the development of artificial intelligence technology. Its A100 and H100 AI GPUs are currently one of the most advanced chips in the company, widely used in fields such as machine learning, deep learning, and image processing. And this Omdia report provides us with some new information about the sales of these two chips.

Omdia believes that Meta and Microsoft are the largest buyers of Nvidia H100 GPUs. They have all purchased up to 150000 H100 GPUs, far exceeding the number of H100 processors purchased by Google, Amazon, Oracle, and Tencent (50000).

It is worth noting that most server GPUs are provided to large-scale cloud service providers. Omdia claims that server original equipment manufacturers (Dell, Lenovo, HPE) are still unable to obtain enough AI and HPC GPUs to fulfill server orders.

The analysis company believes that the sales of Nvidia H100 and A100 GPUs will also exceed 500000 in the fourth quarter. Meanwhile, there is a strong demand for H100 and A100, with a delivery cycle of 52 weeks for GPU servers. Meanwhile, Omdia stated that server revenue in 2023 grew between+6% and+8% year-on-year.

It is worth mentioning that Nvidia's A100 and H100 AI GPUs have achieved considerable success in the market. These chips have excellent performance in terms of performance, energy efficiency ratio, and have received high praise from many customers. And this Omdia report further proves Nvidia's leading position in this regard.

The report also states that almost all companies that purchase Nvidia H100 GPUs in large quantities are developing custom chips for AI, HPC, and video workloads. Therefore, as major companies shift towards their own chips, their purchases of Nvidia hardware may decrease over time.

The report predicts that the value of the server market is expected to reach an astonishing 195.6 billion US dollars (currently around 1.4 trillion yuan) by 2027. The report states that major companies are shifting towards customizing servers specifically for specific applications with different coprocessor arrays. For example, Amazon's dedicated server for AI inference has 16 Inferentia2 coprocessors, and Google's video transcoding server is equipped with 20 custom VCUs. Meta also followed closely and equipped the server with 12 custom processors for video processing.