The first few months of 2018 has led to great leaps in the US cloud hyperscalers in terms of deployments and testing of next-generation solutions, such as 400Gbit/s. However, Chinese hyperscalers are building their infrastructure with notable differences, some of which are materializing in the type of equipment being deployed. While this has often been attributed to a technological gap with Western cloud providers, I believe some of the difference is actually driven by physical limitations of the cloud in China. Three major differences have emerged in the past six months:
First, many cloud buildouts in China are occurring in data centers that have less power availability than in the United States. In the US, we’re seeing an increase in the amount of power many cloud customers want per rack. This is allowing for ever denser deployments. However in China, some regions of the country simply don’t have the same power budget. The impact on the data center is significant as the amount of compute per rack is often limited by power. This is causing the Chinese market to often need slower port speeds or to have one switch span multiple racks.
Second, data center interconnect (DCI) remains significantly different between the US and China. In China, DCI remains regulated, which limits the amount of direct fiber connectivity a cloud provider can deploy. Instead, the cloud provider must comply with local laws which often require links between data centers to be provided by traditional service providers. This difference changes how a Chinese cloud provider must architect its network and also choose the type of equipment needed. Connecting through multiple service providers requires more robust interoperability at each link. In the US, cloud providers can deploy their own fiber and control the entire link, which allows for a more simplistic network.
Third, the average revenue per user (ARPU) for cloud customers in China is often lower than what we see in Western economies. This is especially true for search/social/gaming clouds in China and there hasn’t been the same uptick of ARPU in China as in the United States over the past several quarters. The result is that lower cost equipment is often deployed in order to remain profitable with a lower ARPU at the expense of the user experience (often measured in latency).
Looking closely at the buildout of Chinese cloud companies outside of China, we see that they are closing the technology gap. How Baidu, Alibaba, and Tencent (BATs) – three of China and the world’s biggest internet companies – deploy in Europe, for example, is different than in China. Server density increases and the type of equipment they deploy is different, not only from an architecture point of view, but also their selection of different vendors.
This will be an important trend to watch over the second half of 2018 as not only are the cloud providers in China growing at a more rapid pace than Western cloud companies, but the vendor relationships are also evolving with partnerships between vendors and cloud providers evolving on a regional basis.