[anon1m0us]

My initial impression is a return to on-premise data centers but with cheaper, commodity hardware and a software stack on top.

They are doing for on-prem with hotswappable computers in a cloud what hot swappable hard drives did for on-prem storage.

My prediction is that their technology will be used by existing data centers more than it will be used by enterprises wanting to return to on-prem from their current cloud operations. They might help serve internal customers in the enterprise, like developers, and intranets, but the cost of pipes will be prohibitive for companies needing lots of bandwidth to serve external customers.

[devdas]

It's a lot cheaper to get fat pipes than to pay for data transfer in the cloud at scale.

[anon1m0us]

To function at youtube scale, you need access to multiple backbones. Yes, I agree, Sprint will run an OC3, or fiber, what have you, to your office anywhere in America for $15,000 a month, but that will still be several hops from UUNET.

That Sprint line is like a farm to market road to your warehouse, which works fine when you are serving customers from an ecommerce site or sending and receiving emails or even opening files from NYC and your onprem cloud is in NYC.

It breaks down when you need higher throughput to serve millions of customers. At that point, you need to be at the crossroads of a couple interstate highways which would be analogous to a carrier hotel like at 1 Wilshire in Los Angeles.

I expect oxide.computer will open up on demand scalability to existing service providers who already rent the physical space in locations like 1 Wilshire. Most of those providers today buy hardware from HP or Dell, which is expensive and then also have to manage the software layer on top.

From what I can tell, based on the very little actual product marketing I have found, is that oxide will provide a turnkey solution from one shop. Unfortunately, at the moment, it's hard to figure out what the company actually does, because it's about the team right now.

[devdas]

If you are a cloud customer, you probably don't need to function at Youtube scale.

When you need to serve millions of customers, you get space in a datacenter with multiple telecom providers and start peering at various exchanges. You certainly don't pay per byte of data transferred, and you don't start to build your own physical plant.

[stormbeard]

That sounds exactly like what Nutanix does.

[anon1m0us]

Kind of and after some digging, I think the difference is this: https://www.nutanix.com/products/hardware-platforms#nutanix

Notice how the hardware is actually sold by IBM, Dell, etc.

The difference is oxide is building and selling the physical hardware themselves. When the customer needs more capacity, they could just order 10 new, let's call them "blade" type computers from Oxide and pop them in a rack at much lower cost than a blade from IBM.