FAQ Friday

And the answer is, we don’t totally know because those companies aren’t always completely transparent, but we know a bunch of things. We know, for example, that there are a lot of commonalities, DC busbar-based systems and so on. There are, however, a lot of key differences. One of the key differences is we got to take a clean sheet of paper later than those folks took a clean sheet of paper.

So there are certain design decisions that we made because we were able to make them because we started with a clean sheet of paper later than they did. Classic example of that is we blind mate our compute sleds into a cabled back plane, so there is no cabling coming out of the front of our sleds.

If you look at a photo, often of a Google DC, you will see a bunch of cabling coming out the front. That’s all in the cold aisle with a DC busbar in the hot aisle. By coming to the problem a little bit later, able to take a clean sheet of paper, we were able to blind mate into that cabled back plane, and you don’t have any cabling and stuff at all.

So there are some differences like that. Another key difference is we were able to do things like eliminating the BIOS. There is no BIOS in our system. We actually do that lowest level of platform enablement ourselves. The hyperscalers don’t do that. The hyperscalers themselves, on their x86-based products, they are relying on the same BIOS that you might be relying on.

So there are things like that as well. The key differentiator though is this is a product. This is not operating in a cloistered data center. This is something that you can buy and operate in your own data center. So every design decision that we have made has been around making a product.

The software that is running on this rack runs across an air gap. It does not phone home, does not connect to Oxide at all. It runs completely contained. We need to be able to update that software, update that distributed system, and yes, it is absolutely a challenge. That’s the kind of thing that the hyperscalers don’t need to do.

They’ve got a runbook, they’ve got an SRE team that does that. We don’t have that luxury, so we’ve actually got to turn this into a real product. Last thing I would say is, because we have had plenty of folks at Oxide coming from the hyperscalers, one of the things that they point out is the actual physicality of the system is designed for the operator in mind.

The hyperscalers, they don’t always care about that. One hyperscaler in particular, their employees infamously say that the machines are made of, quote, "Razor blades and hate." There are no razor blades and no hate in the Oxide rack. It is designed purely with love for the operators that operate this thing.

So it is similar to the hyperscalers and some of the big design decisions we’ve made, different also in some important ones, and then importantly, it’s a product and something that you can buy and run in your own data center. Thanks, and see you next time.