The server industry is having a weak 2019 due to high inventory levels at clients, but global server shipments are expected to grow around 5% on year in 2020 as clients' inventory disgestion has almost reached an end. There are three major factors driving the server market from 2019 to 2024: more enterprises have begun pushing for digital transformation, AI applications are growingly popular and cloud datacenter firms are eagerly promoting hybrid cloud systems, according to Digitimes Research's latest 5-year server forecast report.
Surging 5G data transmission volumes are also expected to boost server demand starting 2023.
Global server shipments are expected to see a CAGR of 6.5% from 2019-2024 with Microsoft, Amazon Web Services (AWS), Google and Facebook all planning to establish new datacenters worldwide, eyeing the potential demand, Digitimes Research's data shows.
As cloud computing services have become common, multiple cloud management and hybrid cloud platforms have become a new battlefield among server players. The players have also been unveiling their hyper-converged infrastructure (HCI) products that highly integrate servers' hardware architecture with virtualization technologies to cut down costs and time for clients learning heterogeneous resource integration.
HCI has already become a technological development trend of the server industry, while many have started connecting their software defined datacenter (SDDC) with HCI as a way to assist enterprise to conduct digital transformation.
To achieve a higher AI computing efficiency, chipmakers are looking to divide related chips into two different categories: training and inference, and the move is expected to separate AI-specific servers into two different specification settings.
More enterprises are expected to choose to use AI-specific servers to develop their AI models due to concerns over costs and security and with 5G to come online in the near future, demand for AI-specific servers is expected to surpass that for AI model training servers between 2020-2022.
In terms of network data processing capability, the server market currently has two different markets: large-size datacenter and standard enterprise with both having 25GE as the basic network speed. As AI applications and multimedia and game streaming are becoming popular, large-size cloud computing datacenters are mainly seeking stability and high expansibility, making SerDes (serializer/deserializer) interface a new spec standard.
The network speed of 400GE has already been applied in large datacenters at the moment and 800GE is expected to become available by 2021. For regular enterprises, demand for 100GE is currently rising stably.