Skip to main content

Observability Has a Physical Layer Now: What AI Infrastructure Teaches IT Operations

Rupesh Mainali
Reliability Engine

The next observability gap may not be in the code.

It may be under the rack.

That sounds strange until you think about how AI incidents actually feel in the middle of an investigation. Latency moves around. A few accelerator nodes behave differently from their peers. A model-serving tier has a rough hour and then settles down. Logs are not empty, but they are not decisive. Traces show where the pain is showing up, not always why it started.

The application dashboard may be accurate. It may also be stopping at the wrong boundary.

AI systems depend on software, but they also depend on a dense physical stack: racks, power paths, thermal margin, maintenance activity and, in many environments, liquid cooling. Those physical dependencies can change slowly before they look like a software incident.

Application Signals Are Not the Whole Story

Application teams still need the usual evidence: service health, latency, traces, logs, node behavior and workload timing.

None of that becomes less important because the infrastructure stack is changing. It becomes more useful when the physical dependencies around the workload are visible too.

But AI workloads can make the physical context harder to ignore. A group of servers may share a rack, power path or cooling loop. A maintenance event may affect one slice of capacity. A thermal trend may be visible to facilities before the application team sees a clear pattern. A set of nodes may look logically unrelated in software while being physically related in the data hall.

During an incident, those facts are either close at hand or discovered slowly.

More Dashboards Are Not the Answer

The answer is not to push every facilities tag into a software dashboard. That is how teams get noisy dashboards that nobody trusts.

The better question is: which physical facts explain reliability risk?

Which racks support this workload?

Which shared dependencies serve those racks?

Did maintenance happen before the performance change?

Are the affected nodes physically related, or only logically related?

Is the thermal condition sudden, or is it the visible end of a trend that has been moving for weeks?

That is not facilities data for its own sake. It is reliability context.

Liquid Cooling Is a Useful Example

Liquid cooling shows the pattern clearly because some early signals can be quiet.

A loop can show acceptable temperatures while its condition is becoming less predictable. Filters can load. Particles can collect. Chemistry can drift. A loop can be opened during service and then return to normal operation with a new piece of history attached to it.

No operations team needs to become a coolant chemistry team. That is not the point.

The point is that if a cooling-loop dependency can explain workload behavior, the incident team should not have to find that dependency by accident. If a cluster problem follows service work on the same loop, that relationship should be visible. If a filter alarm affects the same racks as a latency complaint, someone should be able to connect those facts quickly.

Observability Needs a Better Map

This is partly a data-context issue, but mostly an operating model issue.

Software, infrastructure and facilities teams often use different systems, different language and different escalation paths. That separation worked better when workloads were less concentrated and physical margin was larger.

AI changes the cost of that separation.

The practical step is dependency mapping. Map workloads to racks. Map racks to shared physical dependencies. Bring maintenance events, alarms and slow-moving coolant or thermal trends into incident review when they can explain service behavior.

The goal is to make physical reliability context available before a small cooling issue becomes a larger operational mystery.

That starts by treating the rack, the loop and the maintenance history as part of the reliability picture.

Rupesh Mainali is a Senior Member of Technical Staff at Reliability Engine

The Latest

While organizations want to take control of their telemetry, building telemetry pipelines from scratch can be a very daunting, complicated task, even when leveraging open-source standards like OpenTelemetry. It requires specialized knowledge across distributed systems, data engineering, and security. This fragmented approach across systems causes higher operational costs; it puts a strain on resources and reduces efficiency as teams have to work with different interfaces and processes ...

For decades, enterprise networks were designed around a simple assumption: work happened inside the office. Applications lived in centralized data centers, employees connected through internal infrastructure, and security focused on protecting the perimeter that surrounded everything ... But the way organizations operate today bears little resemblance to that environment. Cloud platforms host critical applications, employees connect from homes and airports as often as they do from offices, and partners collaborate through shared systems that exist far beyond corporate walls. In short, the corporate network no longer resembles the environment it was designed to protect ...

As an analyst who researches how IT organizations design, build, and operate their networks, I find that network data is a constant source of pain. Network teams struggle with data quality, fragmentation, authority, access, and trust. And these issues undermine everything they try to do. Here are the numbers: Only 45% of network teams are completely confident in the accuracy of their network source of truth, which documents the intent of their network ...

The 2026 Global Data Center Survey from Uptime Institute reveals an industry navigating workforce constraints, escalating outage expenses, even as rising costs remain the top concern for management teams ...

The next observability gap may not be in the code. It may be under the rack. That sounds strange until you think about how AI incidents actually feel in the middle of an investigation ... The application dashboard may be accurate. It may also be stopping at the wrong boundary. AI systems depend on software, but they also depend on a dense physical stack: racks, power paths, thermal margin, maintenance activity and, in many environments, liquid cooling. Those physical dependencies can change slowly before they look like a software incident ...

Certificate expiration is the rare outage you can see coming. Every TLS certificate carries the date it stops working, so the moment it will begin breaking connections is knowable in advance. That's what makes an expired certificate such a frustrating way to lose a service. What's changing now is how often that date comes around ...

Enterprises operate different combinations of workloads across cloud, hybrid and multicloud environments. For business-critical workloads, teams need to consider monitoring and observability early so they can detect health issues, investigate failures, and understand operational impact. Organizations place workloads on cloud platforms based on a combination of technical requirements, economics, existing dependencies, organizational standards, and business priorities. Their monitoring priorities therefore depend on what they operate and where those systems run. Those priorities will not look the same for every organization ...

Top-performing businesses prioritize data-driven decision making, enabling leaders to move from intuition and gut feel towards evidence-based judgment. But that judgment is only sound when the data underpinning decisions is accurate. With incident management, data accuracy is particularly important. Long-term revenue, customer trust, and operational stability depend on high-quality data that enables teams to quickly identify and address the root cause of major incidents. Against this backdrop, governance becomes a critical endeavor to ensure the right data drives the right action ...

In MEAN TIME TO INSIGHT Episode 26, Shamus McGillicuddy, VP of Research, Network Infrastructure and Operations, at EMA discusses network compliance ... 

Most production autonomous agents do not run in a vacuum. They run inside cloud infrastructure: virtual machines, containers, pods, managed clusters or private servers. That is where most operations teams start monitoring. Is the VM alive? Is the container running? Did the pod restart? Is memory stable? Is CPU too high? Did the health check pass? Those signals are useful. They tell you whether the shell around the agent is alive. They do not tell you whether the agent inside is actually operational ...

Observability Has a Physical Layer Now: What AI Infrastructure Teaches IT Operations

Rupesh Mainali
Reliability Engine

The next observability gap may not be in the code.

It may be under the rack.

That sounds strange until you think about how AI incidents actually feel in the middle of an investigation. Latency moves around. A few accelerator nodes behave differently from their peers. A model-serving tier has a rough hour and then settles down. Logs are not empty, but they are not decisive. Traces show where the pain is showing up, not always why it started.

The application dashboard may be accurate. It may also be stopping at the wrong boundary.

AI systems depend on software, but they also depend on a dense physical stack: racks, power paths, thermal margin, maintenance activity and, in many environments, liquid cooling. Those physical dependencies can change slowly before they look like a software incident.

Application Signals Are Not the Whole Story

Application teams still need the usual evidence: service health, latency, traces, logs, node behavior and workload timing.

None of that becomes less important because the infrastructure stack is changing. It becomes more useful when the physical dependencies around the workload are visible too.

But AI workloads can make the physical context harder to ignore. A group of servers may share a rack, power path or cooling loop. A maintenance event may affect one slice of capacity. A thermal trend may be visible to facilities before the application team sees a clear pattern. A set of nodes may look logically unrelated in software while being physically related in the data hall.

During an incident, those facts are either close at hand or discovered slowly.

More Dashboards Are Not the Answer

The answer is not to push every facilities tag into a software dashboard. That is how teams get noisy dashboards that nobody trusts.

The better question is: which physical facts explain reliability risk?

Which racks support this workload?

Which shared dependencies serve those racks?

Did maintenance happen before the performance change?

Are the affected nodes physically related, or only logically related?

Is the thermal condition sudden, or is it the visible end of a trend that has been moving for weeks?

That is not facilities data for its own sake. It is reliability context.

Liquid Cooling Is a Useful Example

Liquid cooling shows the pattern clearly because some early signals can be quiet.

A loop can show acceptable temperatures while its condition is becoming less predictable. Filters can load. Particles can collect. Chemistry can drift. A loop can be opened during service and then return to normal operation with a new piece of history attached to it.

No operations team needs to become a coolant chemistry team. That is not the point.

The point is that if a cooling-loop dependency can explain workload behavior, the incident team should not have to find that dependency by accident. If a cluster problem follows service work on the same loop, that relationship should be visible. If a filter alarm affects the same racks as a latency complaint, someone should be able to connect those facts quickly.

Observability Needs a Better Map

This is partly a data-context issue, but mostly an operating model issue.

Software, infrastructure and facilities teams often use different systems, different language and different escalation paths. That separation worked better when workloads were less concentrated and physical margin was larger.

AI changes the cost of that separation.

The practical step is dependency mapping. Map workloads to racks. Map racks to shared physical dependencies. Bring maintenance events, alarms and slow-moving coolant or thermal trends into incident review when they can explain service behavior.

The goal is to make physical reliability context available before a small cooling issue becomes a larger operational mystery.

That starts by treating the rack, the loop and the maintenance history as part of the reliability picture.

Rupesh Mainali is a Senior Member of Technical Staff at Reliability Engine

The Latest

While organizations want to take control of their telemetry, building telemetry pipelines from scratch can be a very daunting, complicated task, even when leveraging open-source standards like OpenTelemetry. It requires specialized knowledge across distributed systems, data engineering, and security. This fragmented approach across systems causes higher operational costs; it puts a strain on resources and reduces efficiency as teams have to work with different interfaces and processes ...

For decades, enterprise networks were designed around a simple assumption: work happened inside the office. Applications lived in centralized data centers, employees connected through internal infrastructure, and security focused on protecting the perimeter that surrounded everything ... But the way organizations operate today bears little resemblance to that environment. Cloud platforms host critical applications, employees connect from homes and airports as often as they do from offices, and partners collaborate through shared systems that exist far beyond corporate walls. In short, the corporate network no longer resembles the environment it was designed to protect ...

As an analyst who researches how IT organizations design, build, and operate their networks, I find that network data is a constant source of pain. Network teams struggle with data quality, fragmentation, authority, access, and trust. And these issues undermine everything they try to do. Here are the numbers: Only 45% of network teams are completely confident in the accuracy of their network source of truth, which documents the intent of their network ...

The 2026 Global Data Center Survey from Uptime Institute reveals an industry navigating workforce constraints, escalating outage expenses, even as rising costs remain the top concern for management teams ...

The next observability gap may not be in the code. It may be under the rack. That sounds strange until you think about how AI incidents actually feel in the middle of an investigation ... The application dashboard may be accurate. It may also be stopping at the wrong boundary. AI systems depend on software, but they also depend on a dense physical stack: racks, power paths, thermal margin, maintenance activity and, in many environments, liquid cooling. Those physical dependencies can change slowly before they look like a software incident ...

Certificate expiration is the rare outage you can see coming. Every TLS certificate carries the date it stops working, so the moment it will begin breaking connections is knowable in advance. That's what makes an expired certificate such a frustrating way to lose a service. What's changing now is how often that date comes around ...

Enterprises operate different combinations of workloads across cloud, hybrid and multicloud environments. For business-critical workloads, teams need to consider monitoring and observability early so they can detect health issues, investigate failures, and understand operational impact. Organizations place workloads on cloud platforms based on a combination of technical requirements, economics, existing dependencies, organizational standards, and business priorities. Their monitoring priorities therefore depend on what they operate and where those systems run. Those priorities will not look the same for every organization ...

Top-performing businesses prioritize data-driven decision making, enabling leaders to move from intuition and gut feel towards evidence-based judgment. But that judgment is only sound when the data underpinning decisions is accurate. With incident management, data accuracy is particularly important. Long-term revenue, customer trust, and operational stability depend on high-quality data that enables teams to quickly identify and address the root cause of major incidents. Against this backdrop, governance becomes a critical endeavor to ensure the right data drives the right action ...

In MEAN TIME TO INSIGHT Episode 26, Shamus McGillicuddy, VP of Research, Network Infrastructure and Operations, at EMA discusses network compliance ... 

Most production autonomous agents do not run in a vacuum. They run inside cloud infrastructure: virtual machines, containers, pods, managed clusters or private servers. That is where most operations teams start monitoring. Is the VM alive? Is the container running? Did the pod restart? Is memory stable? Is CPU too high? Did the health check pass? Those signals are useful. They tell you whether the shell around the agent is alive. They do not tell you whether the agent inside is actually operational ...