New capabilities deepen transparency and deliver insights that keep workloads running at scale

The story of AI over the last two years has been all about scale—more GPUs, larger clusters, bigger models, and faster cycles. How to keep it all running smoothly and reliably gets less attention. Early on, a few dashboards and some scripts could do the job. But this approach falls apart fast once you’re training large models or serving production traffic across thousands of GPUs. Small issues in the networking fabric, noisy nodes, or gaps in audit visibility result in wasted compute and hours spent tracking down and fixing problems. Bottom line? Now you can keep your AI infrastructure healthy without a national lab-sized operations group.

CoreWeave  Mission Control: The industry’s first operating standard for AI at scale

CoreWeave Mission Control™ is the industry’s first operating standard for AI* at scale, and it enables AI workloads to run reliably on CoreWeave Cloud. As the operating standard for the #1 AI cloud, it reflects the same operational depth and innovation as our orchestration systems—like SUNK (Slurm on Kubernetes)—that helped CoreWeave earn SemiAnalysis’ Platinum ClusterMAX™ rating for the second consecutive evaluation. CoreWeave remains the only AI cloud provider to receive this rating. 

We continually strengthen Mission Control with new capabilities. And because CoreWeave is purpose-built for AI, we’re able to evolve the operating standard quickly. Today, we are announcing two key innovations in CoreWeave Mission Control: Telemetry Relay for greater transparency and GPU Straggler Detection for deep bottleneck analysis. These new capabilities together expand Mission Control and further optimize CoreWeave Cloud, the Essential Cloud for AI. We are also excited to announce that we are creating much easier access to performance insights with the preview launch of the CoreWeave Mission Control Agent, giving you a way to surface telemetry and remediation guidance through conversational workflows.

Why CoreWeave Mission Control matters‍

CoreWeave Mission Control runs through our entire stack, from foundational infrastructure up through observability, security, and agent workflows. It connects all of the critical layers of CoreWeave Cloud in one place—giving you real-time visibility into GPU, network, and storage behavior. It also enables you to see how systems are performing and keep them stable with secure controls for AI workloads. And it unifies identity and access controls, compliance logging, and audit history, giving you a complete, clear, and defensible record of activity across your environment.

Mission Control delivers continuous operational insight for deep knowledge of your environment. Audit and telemetry signals stream seamlessly into your SIEM on any cloud—along with health checks on GPUs, nodes, and racks. That means you always know the state of the system in real time, not just how it behaved after a failure occurs.

And Mission Control transforms every insight into action with proactive remediation paths. When something looks wrong, Mission Control proactively identifies the issue and initiates the right response—from automated recovery to routing the incident directly to CoreWeave experts who own that part of the stack. No more chasing ambiguous alerts or guessing at root causes.

The end result is that Mission Control shortens detection and repair cycles, strengthens reliability, and keeps high-throughput training and inference running consistently, from small-batch jobs to frontier scale research. It represents proven performance, enabling up to 96% goodput (the share of GPU time actually spent doing useful training work), delivering 20% higher model utilization (MFU), and saving millions of GPU hours for large-scale training programs on CoreWeave Cloud.

CoreWeave Mission Control is built on three key pillars—reliability, transparency, and insights that represent significant benefits for your AI initiatives.

Reliability keeps fleets healthy

Reliability comes first—if the fleet isn’t healthy, nothing else matters.

CoreWeave Mission Control continuously evaluates cluster health across GPU, fabric, and nodes. It proactively keeps an eye out for error signals, performance drift, and patterns, catching them before they show up and cause problems, like rising correctable ECC rates, recurring Xid errors, or sudden changes in collective execution time. When a value crosses a determined threshold, the anomaly isn’t just logged. Mission Control automatically takes quick action, taking nodes out of rotation, steering workloads, and triggering automated recovery so jobs stay on track. 

For one of our customers, a large AI lab training frontier-scale models, Mission Control’s automated recoveries and continuous node and fabric monitoring resolved issues roughly five to six times per day for every thousand nodes. That level of automation kept long training jobs running smoothly and avoided disruptions that would otherwise interrupt rapid progress and exponentially drive up costs. 

When automation isn’t enough, incidents route straight to CoreWeave experts who work with your team, instead of leaving you to guess what’s happening in isolation. That level of trusted reliability means fewer surprises for your team, faster recovery when issues do occur, and fewer wasted cycles on jobs disrupted by underlying infrastructure issues.

Transparency means you always know what’s happening in your environment

Your environment shouldn’t be opaque. You need metal-to-token visibility that shows you exactly what’s happening so you can investigate it, take action, and explain it to your teams, your security partners, and to auditors. Transparency lets you control the data, tracing what happened and when—and your teams can take quick, effective action with complete confidence.

Transparency is crucial for managing a cluster,but it’s also the cornerstone of operating a secure solution. Mission Control integrates CoreWeave’s security foundation together with CoreWeave Observe™, our extensive observability stack. Now identity and access controls, compliance logging, and audit history sit alongside the metrics that describe how AI GPUs, networks, and storage are behaving. And all of these insights can be delivered into your SIEM and monitoring systems, so your security and SRE teams can work with the tools they already know.

CoreWeave Telemetry Relay, a new Mission Control capability now entering general availability, extending transparency even further. It forwards audit and observability signals—including Kubernetes audit logs, API and console activity, and hardware and fabric metrics—into your SIEM or monitoring tools with reliable, predictable delivery. With Telemetry Relay, you won’t have to build and maintain a one-off, custom export pipeline. 

Mission Control provides Grafana dashboards and metrics out of the box for GPUs, clusters, network, and storage with no extra setup. Integration with Weights & Biases means CoreWeave Cloud can stream more than one million data points per second into training and experiment tracking, giving your teams a granular, real-time view of how models and infrastructure behave together instead of forcing them to stitch together separate tools.

Insights move you from “something’s wrong” to “problem solved”

Once your infrastructure is reliable and the signals are visible, the next step involves getting to answers quickly. That’s where having the right insights matters most.

Maximizing performance inside large distributed jobs remains one of the most critical (and expensive) issues at scale. While node failures are tough on a job, gray failures are extremely painful. A single slow or unhealthy rank in a collective can pull down throughput for an entire run and are very difficult to detect. From the outside, these issues often look like a job is just moving more slowly than it should. The truth is much more complex—and far more disruptive—than it appears from the outside.

To address this problem, we are announcing the preview launch of GPU Straggler Detection. It uses signals from NCCL to identify the specific rank, GPU, and node that are out of line with the rest of the job. In Grafana, that outlier clearly stands out. Alert patterns take you directly to the hardware you need to investigate, and you can then correlate that with your training runs in Weights & Biases.

To see GPU Straggler Detection in action, watch the demo video below. It shows how easy it is to ask our new CoreWeave Mission Control Agent in Slack why a job is running slowly. The agent traces the straggler through our observability stack and highlights the bottleneck GPU and node. Then it replies in the thread with a short explanation and the relevant Grafana view. This simple example shows how insight from CoreWeave Mission Control optimizes performance and improves throughput for your training runs.

The result is a much shorter path to a specific action like swapping a node, moving a workload, or adjusting a configuration. In our internal benchmarks, this automated path is about 3x faster than traditional manual investigation.

CoreWeave enhances Mission Control to help you run AI at scale

At CoreWeave, we are relentlessly committed to bringing new innovations to CoreWeave Mission Control, continually expanding its capabilities to raise the bar on operational efficiency. Along with Telemetry Relay (entering general availability) and GPU Straggler Detection (in preview), we also announced the new CoreWeave Mission Control Agent (in preview), which makes it much easier and faster to get to critical insights. Finally, to make all of this information more actionable before the jobs even start, we offer CoreWeave Mission Control Reviews, where we map your environment to this operating standard so that we can collaboratively build a clear activation plan together. 

As we all know, running AI at scale continues to get more complex. CoreWeave Mission Control ensures that your infrastructure is ready for the escalating challenges—and amazing opportunities—that lie ahead.

• Request access to the CoreWeave Mission Control Agent
  
• Explore CoreWeave Mission Control
  
• Find out why CoreWeave is the Essential Cloud for AI
• Talk to one of our AI experts about enabling Telemetry Relay or joining the GPU Straggler Detection preview