AWS CloudOps sharpens multi-cloud support for AI and resilience
AWS used re:Invent 2025 to send a clear signal: multi-cloud and resilience are no longer optional. AI apps and agents are multiplying traffic, telemetry, and blast radius. Outages and regulations are forcing plans to span providers.
The updates focus on three things leaders care about: simpler cross-cloud networking, less toil in Kubernetes, and observability that scales with AI. The subtext is simple-meet customers where they run AI, even outside AWS.
What AWS announced
- AWS Interconnect - multi-cloud (preview): Built with Google on a new open source API spec. Provisions dedicated, on-demand bandwidth and cross-cloud connectivity from your console or API.
- AWS Interconnect - last mile (gated preview): Automates Direct Connect to customer sites (private data centers, campuses). First ISP partner: Lumen; more partners required.
- EKS Capabilities (GA): Managed Argo CD for GitOps, AWS Controllers for Kubernetes (ACK) to manage AWS resources via Kubernetes, and Kube Resource Orchestrator (KRO) for templating.
- CloudWatch and AIOps updates: GenAI observability (latency, token usage, errors), Application Signals auto-grouping, CloudWatch Investigations with automated "Five Whys" reports, MCP servers, and a GitHub Action for code-to-telemetry correlation.
- Data and security: NLP for data pipelines, cross-account/region log collection, CloudWatch Database Insights across RDS/Aurora and accounts/regions, and aggregated CloudTrail events.
Why this matters for managers
- Resilience: Multi-cloud is now a compliance and risk strategy, especially with new EU requirements for digital operational resilience.
- AI scale: Agentic systems balloon telemetry and network needs. Tooling must keep up without adding people.
- Operating model: More abstraction (and agents) means fewer manual runbooks, more policy and guardrails.
- Vendor posture: Even AWS is acknowledging cross-cloud workloads. That's a shift you can use in negotiations.
Networking: Interconnect reduces cross-cloud friction
Historically, cross-cloud networking meant long lead times, hardware, and many teams. Interconnect aims to cut that to minutes via APIs. For teams mixing providers to source the "right" model per task or region, this removes hand-built tunnels and error-prone routes.
Leaders echoed two benefits: speed and resiliency. One executive called cloud-to-cloud links a "giant project" that rarely feels reusable; another pointed to EU resilience expectations as the real driver for clean, reliable, low-latency interconnects.
Two watch-outs: the ISP ecosystem for last mile needs to mature, and total cost will decide mainstream adoption. As one analyst put it, the proof arrives as a monthly bill.
Kubernetes: EKS Capabilities move ops "up the stack"
EKS Capabilities lean on familiar open source but keep you in the AWS lane. Managed Argo CD simplifies GitOps. ACK manages AWS services via Kubernetes. KRO templatizes common platform resources. The pitch: ship apps, not platforms.
Pricing matters. Managed Argo CD is listed at $0.02771 per hour plus $0.00136 per app per hour. With 100 apps, that's roughly $130/month. At higher app counts and multiple environments, costs can climb fast. Some leaders will prefer building their own; others will trade cash for speed and standardization.
Longer term, expect AI agents as the control layer. AWS introduced an EKS Model Context Protocol server and a DevOps agent-hinting at an interface where agents recommend the metrics to scrape and the actions to take, rather than engineers combing dashboards.
Observability and AIOps: built for AI signals
- GenAI observability: Latency, token usage, and error tracking compatible with agent frameworks such as LangChain, LangGraph, and CrewAI-works across clouds.
- Application Signals: Auto-builds app topology and now auto-groups resources by application without instrumentation.
- Investigations: Automated incident reports with the "Five Whys," standardizing learning and accountability.
- MCP servers: For CloudWatch and Application Signals, enabling agent-driven workflows.
- GitHub Action: Correlates CloudWatch telemetry with source code so teams can see what changed when incidents occur.
Data and storage: AI agents create data gravity
Always-on agents generate huge volumes of telemetry, security events, and traces. The updates aim to contain sprawl: NLP for pipelines, cross-account/region log collection, database insights across RDS/Aurora and accounts/regions, and aggregated CloudTrail events.
The goal many leaders want-a single source of truth for AI ops-depends on faster connections and simpler data movement. Waiting months for circuits blocks AI initiatives; APIs shorten that path.
Budget and ROI: where the money goes
- Connectivity: Interconnect could offset weeks of networking effort. Model likely varies by bandwidth, paths, and partners. Track egress, metered bandwidth, and service premiums.
- Kubernetes: Managed features reduce toil but add steady costs. Benchmark against your internal platform team's fully loaded cost to run and support alternatives like Crossplane or OpenShift.
- Observability: AI signals multiply ingestion. Set rate limits, retention tiers, and sampling policies before data bills set them for you.
Executive checklist
- Run a pilot of AWS Interconnect between two clouds that host your highest-value AI workload. Measure latency, failover time, and provisioning speed.
- Update outage and failover playbooks to include cross-cloud routing tests and ISP dependencies for last mile.
- Standardize on an EKS operating model. Decide: managed Argo CD and ACK, or a DIY stack with existing licenses and talent.
- Instrument generative AI apps now. Track latency, tokens, and error budgets. Tie alerts to customer impact, not just infrastructure noise.
- Centralize logs across accounts and regions with clear retention tiers. Enforce access policies at the org level.
- Trial an AI ops agent for one service path (build → deploy → rollback). Define guardrails and human approval steps.
- Validate compliance requirements such as EU digital operational resilience and confirm your multi-cloud design meets them.
- Set cost guardrails: budget alerts, per-team quotas, and weekly reviews of top drivers (bandwidth, storage, ingest).
Useful links
Upskilling your team
If you're building a skills plan for AI operations across roles, here's a practical catalog to map training by function: AI courses by job.
Bottom line
AWS is meeting customers where AI actually runs-across providers, accounts, and regions. The wins are faster network setup, less Kubernetes maintenance, and observability that speaks the language of AI. The trade-off is cost. Pilot, measure, and commit where the math works.
Your membership also unlocks:
