Nvidia, the chipmaking giant, has witnessed an extraordinary surge in its data center revenue, reaching a staggering $14.5 billion in the third quarter of 2023, as reported by research firm Omdia. The company's significant growth is attributed to the shipment of nearly half a million H100 and A100 GPUs during this period.
In a Market Snapshot Report, Vlad Galabov, Director of Omdia's cloud and data center research practice, and Manoj Sukumaran, Principal Analyst, Data Center Computing and Networking, highlighted that Nvidia's Q3 data center revenue has quadrupled compared to the same quarter in the previous year.
The majority of the GPU server shipments in Q3 were directed towards hyperscale cloud service providers, with Meta emerging as one of the largest clients. Microsoft also made substantial orders for H100 GPUs, presumably to power their expanding portfolio of AI products and Copilots.
Following Meta and Microsoft, other tech giants such as Google, Amazon, Oracle, and Tencent secured notable positions in the list of Nvidia's clients. However, it's worth noting that Tencent faces challenges in obtaining H100s due to stringent export restrictions imposed by the Biden administration.
The analysts at Omdia predict that Nvidia is on track to surpass the half-a-million mark in GPU shipments by the end of Q4 2023, highlighting the ongoing high demand for hardware in the market. However, server manufacturers like Dell, Lenovo, and HPE are facing delays in fulfilling their H100 server orders due to a shortage of GPU allocation from Nvidia, leading to wait times of approximately 36 to 52 weeks.
Looking ahead, Galabov and Sukumaran anticipate significant growth in the server market, projecting it to be worth $195.6 billion in 2027, more than doubling from a decade ago. This growth is attributed to contributions from server processors and co-processors, particularly in the realm of hyper-heterogeneous computing, where application-optimized server configurations with numerous co-processors are gaining traction.
In the domain of AI, the report highlights popular server configurations for tasks such as large language model training and video transcoding. Nvidia's DGX server, equipped with eight H100/A100 GPUs, is a preferred choice for large language model training, while Amazon's servers for AI inference, with 16 custom-built co-processors, lead the way in that domain.
The report emphasizes the increasing importance of hyper-heterogeneous computing, noting its impact on data center physical infrastructure. Rack power distribution revenue has outpaced expectations, and data center thermal management revenue is projected to grow by 17% in 2023, driven by higher rack densities requiring liquid cooling solutions.
As the demand for applications like generative AI continues to grow, the authors caution that the only potential limitation to the current rate of AI deployment is power availability. Notably, companies are leveraging sought-after GPUs as debt leverage, indicating the strategic importance of these components in the rapidly evolving landscape of data center technology.
