AIOps

Nvidia Corporation unveiled Nvidia Minerva to bring end-to-end observability for AI factories.

The new AIOps platform connects infrastructure and workload telemetry to help operators increase uptime and maximize AI factory yield.

It brings together telemetry from infrastructure and AI workloads, then applies purpose-built AI models to turn that data into timely, actionable insight.

Advanced Micro Devices Incorporated introduced the AMD Helion agentic AI rack.

So people that have worked inside enterprise-level Kubernetes know that the reality is usually driven by diversification and nuance instead of just a single centralized pattern, given the diversity of opinions, experiences, preferences, which tools people have worked with.

And so when we're looking at the subject of AI ops, why does tendency design make it harder?

Oh my word, yeah, definitely.

Because now you have a lot more context and complexity that you have to feed the AI.

And that Kubernetes team is managing Fluent Bit and this log shipping, the log shipping services, and your cluster is about to tip over because even if you have auto scaling, you may probably still be clamping it.

So this is just an example.

Who has the AI ops? Is it the Kubernetes team or is it the application team? And does the application team know that they're tipping over that responsibility? These are the complexities that I kind of was starting to dive into with my teams and figure out, you know, how do you get all the information to the tooling? One plus one, though, to Datadog MCP.

It's really good.

Yeah, absolutely.

And another key update they came out with was the partnering with NVIDIA to launch Minerva, NVIDIA's new AIOps platform on AI Cloud.

Now, the benefits of this is that the new tool watches the whole data center at once, learns what normal looks like, and then warns where there's potential failures before they actually happen and bundles hundreds of alarms into one alert pointing to the likely cause.

Now, you might be asking why this matters.

It grew into a much stronger platform.

And they built AI in it as well, AIOps.

But it was not the same approach, right? And there was a lot of strength in the on-premise version.

So...

Our duty is also to cover the cost.

At GTC Berlin, NVIDIA unveiled Minerva, an AIOps platform designed for what the company is calling AI factories.

Minerva consolidates GPU telemetry, SuperNIC data, switch and cable information, and job-level workload data into a dynamic knowledge graph, running purpose-built models over billions of data points per hour.

The goal is to group cascading alert storms into single incidents with a likely root cause, dramatically reducing the operational burden on teams managing large-scale GPU clusters.

And for example, there will be agents that are focused on change management, right? So understanding what's happening in a particular change cycle, what the new feature that's being added or added.

parameter that's being changed and how that's going to impact the network and then putting active tests in place and using AI agents to understand what's changing, design the tests that are going to be run and ultimately implement those and report back to both the AIOps system via an A2A interface or MCP interface, and then also to the engineers that are overseeing the network so they have that visibility.

Once that change is made, moving into performance monitoring, there's a lot of telemetry that's going to be coming from this network.

And we see kind of two areas of being very important.

[chuckles] Yeah, I think, um, you know, Yash, what's not necessarily clear or for folks that may be not familiar with your platform is how are we building that trust? You know, the, the thought of, and you had mentioned it, that we don't wanna release rogue agents on the network to do configurations.

So how are we self-driving in a way that's practical for an engineer who's maybe new to the ecosystem? How do we help them understand the competitive differentiation that you guys have, and, and what does it actually do from a day two perspective? You know, you mentioned things around AIOps and reducing alert fatigue.

I-- obviously, the old running joke is the first thing you do is you create the SNMP alert, a-and then the second thing you do is you create the Outlook rule to delete it, um, because you just get overwhelmed with noise, right? And so how are we actually applying the technology that you guys have created to help the network engineer in the field to do the day two? Your business cases were great, but tactically, what does that look like for them?

Um, I know, Michael, you might be hearing a lot about today's development on the agents, and people are talking about native MCP, and your question is spot on why, how the, um, Mist platform or central platform is so unique in the industry and why they have to trust the Mist platform.

if your data is not in order.

And what's interesting is I was around for the first version of AIOps about nine or 10 years ago.

Observability, APM, we called it then, Application Performance Monitoring Tools, based on machine learning algorithms rather than neural networks or generative AI.

you know, we're trying to apply analytics based on automation.

or AI as a service like Cloud, for example, where Anthropix is running it, or Amazon Bedrock.

That's where people tend to start is with a hosted service.

And then as far as more self-hosted AI, internal applications, things that are not about the workforce or business process, I'd say that that same domain of IT service management, IT operations observability, it has caught on pretty quickly in part because many of them already lived through the AIOps trend and got a lot of those knowledge base documentation in order.

And so they've been able to make a bit of a faster transition to that.