AI Promises of Fixing the World

Yeah, ok, maybe the title is very dooms day-ish and I think by now we know that Artificial Intelligence (AI) can be very useful in many situations. It can solve a lot of problems very quickly, give us good ideas, generate some fun images (some even used on this site), write code, quickly summarize, and the list goes on, however, it has also introduced some real problems for customers.

One those problems affecting the entire industry is massive price increases for server components, notably with memory, and component shortages causing long lead times ... really long. The industry has never seen anything like this. The unpredictability of the market is (and I hate to use this word) unprecedented. The cause, well ... AI. It didn't exactly solve all our problems (at least in the short-ish term). Prices going up 100% or more on any given week, lead times for components can shift from 30 days to +180 days in the blink of an eye, and it doesn't appear that the problem will be relieved anytime soon.

The sad reality in all of this is that this is not only affecting Cisco but every single OEM and causing policy changes across the board. One should really not point the finger at any one server vendor for these issues. They're all in the same boat and there's really not much anyone can do, kind of.

I won't get into the details of how or why we got here other than to say ... AI. The short of it is that ASML (Advanced Semiconductor Materials Lithography) is unable to keep up with the market and AI demands and more specifically, for their newer, $400M machines, which allows component manufacturers to create the high bandwidth memory required for AI. As a matter of fact, if you're looking to understand more how this process works, there's an absolutely awesome video that was created by the Veritasium channel on YouTube about ASML machines (not to be confused with ASMR) ... check it out here.

Why Cisco Matters with Nutanix

So why the hell am I saying all this? First, it's important to provide just a bit of context, even if it's the tip of the iceberg, but understanding the problem really brings the offering from Cisco with Nutanix to a new level.

I've written about the hardware reusability program/feature for Cisco UCS and Nutanix (Part 1, Part 2) before this even became an industry problem, and now, it's as relevant as ever. I highly recommend reading those two articles but the gist of it is that an existing Cisco UCS customer is able to retrofit (if needed) their qualified servers and run the Nutanix Cloud Platform software without needing to buy a dedicated appliance (or compute-only node), and yes, this is fully supported by both organizations. This image from Part 2 kind of brings the point home:

Awful "Perfect" Storm

It's a like an awful "perfect" storm has hit the industry; Broadcom, component shortages, and massive price increases, but Cisco and Nutanix are stepping up and answering the bell. The two tech companies continue to offer the hardware reusability feature for HCI and took it even further when FlashStack with Nutanix was released introducing retrofit support for B200 M5 and M6 servers. Supported components can be found in the Cisco spec sheet and Cisco is always looking to improve on this. Why are the B200's being qualified a big deal? I was fortunate enough to attend Nutanix's Global Tech Summit (GTS) in Orlando a few weeks ago where Cisco was up on stage with Everpure (formally Pure Storage) and mentioned there are over 500,000 B200 M5/M6 servers out there, ripe for the picking (conversion).

The opportunity is pretty big to not only be able to move off VMware or introduce a dual vendor strategy, but also do it without having to purchase new hardware for both UCS Compute and Everpure storage arrays (aka Brownfield deployments). There are a lot of FlashStack customers out there that have been itching for a solution to their Broadcom problem, Cisco, Everpure, and Nutanix answered.

This program now officially also includes UCS X-Series X210c M6 for compute-only nodes as well as the latest generation of Cisco servers, M8! That's right - Cisco and Nutanix have qualified the X210c M6 as a compute-only node, meaning it can be used in a FlashStack with Nutanix solution (CO+external storage) or in an HCI cluster that needs to scale with compute-only resources (CO + HCI). It is important to point out, however, that the X210x M6 is not supported as an HCI node (with local storage inside the node itself).

Additionally, Cisco has also qualified their M8 platforms with the compute-only node program. While this doesn't necessarily fall directly in line with retrofitting existing servers as M8s are fairly new, it still very much plays into the overall flexibility of the solution; FlashStack with Nutanix (CO + external storage) and scaling compute only resources in an HCI cluster (CO + HCI) ... or science forbid, yet another industry shift that requires the customer to shift their strategy again, scary, but possible.

Imagine a customer being able to:

• Have a solution to move off Broadcom (or introduce multiple hypervisors) with an enterprise ready hypervisor (Nutanix AHV)
• Have the choice to go with HCI or CI (FlashStack with Nutanix) with UCS, more or less, on demand, giving them a cloud like architecture and design
• Relieve pressure off budgets by being able to reuse hardware
  • Cisco UCS
    • C-Series M6/M7/M8
    • X-Series M6/M7/M8
    • B-Series M5/M6
  • Everpure storage arrays
    • //X
    • //XL
    • //C
• Keep their projects running by avoiding long industry lead times by repurposing their existing inventory

And all that without even starting to discuss the operational and management simplicities that the whole solution brings, like direct Intersight integration.

That's a win if I've ever seen one. Let's keep the fires roaring!