Datacenters are set to standardize on the larger, 21-inch rack format by 2030, according to Omdia, as hyperscalers and server makers fully embrace it, leaving enterprises to the existing 19-inch standard.
The analyst biz forecasts that the larger rack format, popularized by the Open Compute Project (OCP), will make up over 70 percent of kit shipped by the end of the decade, as it is increasingly adopted by vendors such as Dell and HPE that have been riding the AI infrastructure wave.
The 19-inch rack frame predates the computer industry, originating as a housing for relay circuits for the rail industry, according to some sources, before being adopted by telephone companies for exchanges, then picked up later by the IT industry as a handy, ready-made standard.
But moves to supersede it have been afoot for some time. As The Register wrote way back in 2012, Facebook (now Meta) decided the current flavor was cramping its style and started up the OCP to get the ball rolling on a new standard.
These days, all the largest cloud service providers including Microsoft, Amazon, Meta, Google, ByteDance, Huawei, and Oracle have shifted to using 21-inch racks as their datacenter standard, Omdia says in an analysis note.
In fact, Huawei made the switch some time ago, announcing that all its datacenters would be built around 21-inch rack infrastructure back in 2019.
The advantages of the larger format include better airflow for more optimal thermal management and the ability to fit larger fan arrays, while there is also more space for high-capacity power distribution and piping to enable direct-to-chip liquid cooling for AI-optimized infrastructure.
It is also more future-proof because kit designed for the older rack format can be mounted into a 21-inch frame, but the reverse is not possible.
Omdia expects that rack-scale server designs will become the de facto standard for cloud service providers, whereby an entire rack of IT equipment can be wheeled pre-assembled into the datacenter. Nvidia’s DGX GB200 NVL72 is already there.
Dell and HPE are also said to be embracing the OCP’s modular server architecture known as the Data Center Modular Hardware System (DC-MHS). In these specifications, each component is a module that can be swapped out, enabling upgrades while maintaining the same I/O.
Omdia has been seeing AI servers account for a larger and larger chunk of overall server spend for a while now – 66 percent last year – as the hyperscalers and specialized AI cloud service providers splash out on kit to feed the AI frenzy. With these big spenders favoring the 21-inch OCP rack, it was only a matter of time before it won out. ®