Often we hear about massive data centers (like the ones that run the world’s largest social network), but do we ever hear about how those servers got there, hosting all of our information? Traditionally, companies have contacted original equipment manufacturers (OEMs) like Dell and HP to purchase servers that will best fit their needs. These companies then send the order to original design manufacturers (ODMs) that are primarily based in China and Taiwan (due to lower labor and parts costs). ODMs make the parts from scratch and send them to OEMs (e.g. Dell and HP) to assemble and maybe add a few more pieces of specialized hardware (to keep their edge over companies wanting servers directly from ODMs).
Customization used to be a four-letter word for companies like Dell and HP because it would create a wrinkle the assembly line of production and shipping (which equals a cost increase for them). Now the tides have changed and customization is the name of the server game.
Companies like Google cut out the middle man and began contracting with ODMs to build custom servers to lower their own costs (i.e. not buying servers with features and hardware they don’t need) and to lower power consumption in their data centers. However, Google still considers their design of these servers their secret sauce and guard it carefully, even though ODMs have the resources and personnel to design server specifications if that’s what a customer orders.
Facebook had the same idea to use custom servers in their data centers. However, they recognized that hiring the talent to design a server specific to their needs would cost them money, and they began using ODMs for designing and manufacturing their servers. Once the servers are completed, they’re sent to a testing and assembly location, specifically Hyve Solutions in California, and from there, delivered to Facebook data centers. When Facebook saw the success of their simplified server design, they made the blueprint public in a Facebook founded and run open source venture called Open Compute Project.
Hyve’s level of involvement in Facebook’s server construction is rather minimal, but for its other customers they do much more. Hyve is the newest division of Synnex, a company that’s been buying and selling computers, chips, hard drives, memory, and all kind of hardware for thirty years. The secret to Hyve’s new found success comes directly from Synnex. Without Synnex’s working relationships with manufacturers in Asia, Hyve wouldn’t be able to get custom parts for such low prices. And with Hyve’s goal to supply the world’s “large-scale Internet companies,” they need those low prices for the bulk of servers they deliver.
Companies like Hyve, which act as a small scale OEM replacement, have been around for awhile. SGI and ZT Systems (to name only two) have worked with the likes of Amazon and Microsoft to deliver custom server solutions for their data centers. Hyve’s novelty is that it has blurred the lines between OEM, ODM, and simple delivery systems further than anyone else in the industry. It might seem like a haphazard approach to gain a strong foothold in manufacturing and assembly of servers, but so far, Hyve has had nothing but success. Ordering servers for companies based on the Facebook designs from the Open Compute Project was Hyve’s first homerun, but now companies are coming to them for custom servers that aren’t based on Facebook’s open-sourced designs.
Again, Hyve isn’t necessarily breaking new ground on the computer manufacturing and assembling industry, but the trend of companies looking towards customization instead of big name OEMs is an interesting one. While Google is confident in taking their servers into their own hands, companies like Hyve seems to be supplying that extra step for data centers that are looking at options other than the big time server providers for the first time.