How they use Nebius AI Cloud
Datadog lists Nebius, Inc. as a United States subprocessor providing AI services. Its register identifies subprocessors by their locations and the services they provide to Datadog. Nebius says Datadog signed a multiyear agreement with it for the first time in Q3 2026.
Nebius AI Cloud logs stream into Datadog Log Management through an integration covering Compute, Managed Service for Kubernetes, Applications, Managed Service for PostgreSQL, Managed Service for MLflow and Kubernetes control-plane events. Nebius says teams can investigate Nebius workloads alongside the rest of their stack in Datadog, while Nebius AI Cloud Observability remains available.
Datadog’s merged Agent code preserves a device specification observed on a Nebius H200 node running dynamic MIG. The observed resource pool was computeinstance-u00gcpath9ehe3c8xe. MIG splits one physical GPU into separately allocated compute and memory slices so separate workloads can share it.
The associated pull request reports a test with two pods using slices labelled 1g.35gb and 1g.18gb on the same GPU. Each pod’s metrics were attributed to its own MIG UUID. Validation also covered attribution after initial allocation, its removal after pod deletion, re-attribution after pod recreation, and automatic recovery after an Agent restart.
The implementation resolves Kubernetes DRA resource claims into device identities the GPU check can use for pod tagging. Allocated device names are scoped to a resource pool rather than being NVML UUIDs. Node-local CDI specifications map those names to device nodes, which the implementation resolves through NVML to MIG UUIDs.
Those UUIDs let the Agent associate GPU metrics with the containers and pods holding the allocated slices.
Nebius reports that Datadog selected it for AI cloud deployment, citing its purpose-built infrastructure, technical collaboration and support as the basis for an infrastructure partnership.
Sources
8 sourcesAgent PR #55271: node-local DRA device resolution for MIG attribution
Listed 22 Oct 2026 · GitHub · Web page
For Datadog’s Nebius H200 monitoring validation, zhuminyi reports two pods using 1g.35gb and 1g.18gb slices on the same GPU, each attributed to its own MIG UUID. The PR records resource pool computeinstance-u00gcpath9ehe3c8xe and explains how node-local CDI specifications resolve DRA allocations for GPU-metric pod tagging. Tests cover allocation, deletion, recreation and Agent restart recovery.
Open original sourceNebius Cloud Logs are now available in Datadog: Trace AI incidents across every layer
Published 15 Jun 2026 · nebius.com · Web page
Nebius says logs from core Nebius AI Cloud services stream into Datadog Log Management. The integration covers Compute, Kubernetes, Applications, PostgreSQL, MLflow and Kubernetes control-plane events, while Nebius AI Cloud Observability remains available.
Open original sourceImmutable merged NVML fixture identifying Nebius H200
Listed 22 Oct 2026 · GitHub · Web page
Datadog’s merged Agent test fixture preserves a device specification observed on a Nebius H200 node running dynamic MIG.
Open original sourceNebius Group Letter to shareholders Q3 2026
Published 12 Nov 2026 · assets.nebius.com · PDF
Nebius says Datadog selected it for AI cloud deployment, citing its purpose-built AI infrastructure, technical collaboration and support as the foundation for an infrastructure partnership.
Open original sourceCompany context
datadoghq.comDatadog provides a cloud-based monitoring and analytics platform used by developers, IT operations and security teams, as well as business users. Its products help organizations monitor infrastructure and application performance, manage logs, observe user experience and assess cloud security across cloud, hybrid and on-premises environments. The platform brings together metrics, traces, logs and security signals to help teams investigate issues and understand the health and performance of their technology systems.
Public company
Recent updates
18 updatesMinIO and Datadog partner on on-premises log analytics
MinIO announced a partnership with Datadog to offer an on-premises lakehouse certified for Datadog Bring Your Own Cloud. MinIO said AIStor for Datadog BYOC was available immediately.
MinIODatadog launches Auto-Processing for logs
Datadog announced general availability of Auto-Processing, which automatically structures logs and adapts parsing as log formats change. The feature can create parsing pipelines for custom sources without manual rule setup.
DatadogDatadog COO Adam Blitzer joins Rocket Companies’ board
Rocket Companies announced Adam Blitzer, Datadog’s chief operating officer, as an independent director on its board.
Rocket Companies