Infrastructure Software Watch: AWS-Microsoft Azure Multicloud Interconnect, Kubernetes 1.37, and the Inference Utilization Shift

The CODEW Infrastructure Software Watch | September 7, 2026

Multicloud Interconnect, Kubernetes 1.37, and the Inference Utilization Shift

The CODEW Infrastructure Software Watch cover

Executive Brief

From Provisioning to Utilization: The Infrastructure Layer Becomes the Product

The biggest signal this week: infrastructure software stopped selling capacity and started selling utilization — as multicloud interconnect, Kubernetes scalability fixes, and token-metered inference orchestration converge.

Microsoft and AWS jointly productized a private high-speed interconnect between Azure and AWS, acknowledging that 84% of enterprises run significant workloads in both. Kubernetes v1.37 shipped etcd RangeStream to fix API-server memory pressure at scale. Pulumi added native HCL execution and Terraform state hosting to capitalize on IBM/HashiCorp licensing uncertainty, while Terraform 1.16.0 fixed module imports. In AI infra, Crusoe signed a $13B 5-year deal with Jane Street, Nscale sought $3.5B pre-IPO, and Megaport raised $594M for a distributed inference cloud. Observability consolidated fast with Palo Alto's Chronosphere $3.35B and Embrace acquisitions, plus Coralogix $200M. Underneath: the market is shifting from "how fast can we spin up" to "how little do we waste once running."

1. Infrastructure at a Glance

Top Developments

AWS + Azure Multicloud Interconnect: Private high-speed link combining Azure Multicloud Interconnect (preview) + AWS Interconnect - multicloud, bypassing the public internet.
Microsoft financial shift: Will report Azure sales quarterly, not just the growth rate, enabling direct AWS/Azure comparison in the AI boom.
Kubernetes v1.37: etcd RangeStream tackles hidden scalability bottleneck – memory pressure during large list ops and cache re-init.
Terraform 1.16.0: Fixes most annoying import restriction – now allows import blocks targeting resources inside modules, including for_each instances (Aug 26, 2026).
Pulumi universality push: 300+ providers, now runs HCL directly, hosts Terraform/OpenTofu state, and bridges Terraform providers via CrossCode.
AI infra surge: Crusoe $13B Jane Street deal, Nscale $3.5B pre-IPO talks, TCS HyperVault $7.4B 1GW campus in India, Megaport AI inference cloud.
Observability funding: Coralogix $200M late-stage + $115M Series E at $1B+ with Olly agentic assistant; OpenObserve $10M Series A for AI-native observability.
Data infra: Lenovo acquiring Infinidat for high-end AI storage; Databricks Neon $1B (serverless Postgres) and Arcion CDC; pgvector becoming enterprise default.
Security: CVE-2026-24513 (CVSS 8.8) ingress controller, CVE-2026-22822 (CVSS 9.3) External Secrets Operator namespace isolation break, unpatched Argo CD repo-server RCE.

2. Cloud Platform Watch

Hyperscalers Finally Productize Multicloud

Microsoft and AWS are building a native private interconnect that combines Azure Multicloud Interconnect with AWS Interconnect – multicloud. Previously, customers relied on third-party fabrics (Equinix, Megaport) or public internet VPNs.

Why it matters for infrastructure software:

Flexera 2026 State of the Cloud: 84% on AWS, 82% on Azure with significant workloads – same app, two clouds, is now the default enterprise architecture.
Infrastructure-as-Code must now model cross-cloud private networking as first-class resources, not external hacks.
Cost and compliance drivers: private link avoids egress over the internet, enables low-latency data replication for lakehouse and vector sync.
Kyndryl's launch of Cloud-Native Services for AWS, GCP, Azure reinforces managed services demand for operating multicloud, not just migrating to it.

Azure Reporting Shift + AI Zone Expansion

Microsoft will report Azure sales quarterly, plus infrastructure signals: AWS Saudi AI Zone (Dec 2026) with >$5.3B and HUMAIN partnership for 50MW by 2028, and TCS's 1GW HyperVault campus for GPU-heavy AI.

Financial transparency: Enables investors to tie Anthropic's $30B Azure commitment, $45B Nscale deal, and $1.25B/month SpaceX compute agreement to actual revenue.
Sovereign AI zones: Cloud regions are being re-architected as AI zones with dedicated power, networking, and local LLM partnerships – not generic compute.
Traffic reality check: Hyperscaler internet traffic share ~7.5% combined; extended cloud at 11.37% with CDN/regional clouds absorbing inference.

3. DevOps & Platform Engineering

Kubernetes 1.37: Fixing the Hidden Scalability Tax

v1.37 introduces etcd RangeStream to reduce API-server and etcd memory pressure during large list operations, cache init/re-init – a common cause of control-plane OOMs during upgrades at 500+ node scale. Also continues deprecation of legacy kube-dns, IPVS, cgroup v1.

Platform engineering takeaways:

Standardization thesis validated: Platform teams now focus on hardening infra through IDPs and standardized GitOps/CI/CD, not on Kubernetes install.
Developer productivity is now an infrastructure concern – "Kubernetes isn't the hard part anymore."
Enterprises rethinking Kubernetes: cluster sprawl, toolchain sprawl, and policy enforcement as an engineering discipline mean IDPs as foundations for self-service infra, automation, best practices, metrics (Forrester/Spotify Backstage pattern).

IaC: Pulumi Goes Universal, Terraform Fixes DX

Two moves define Infrastructure-as-Code this week:

Pulumi: Native HCL support (execute HCL via Pulumi engine), Terraform/OpenTofu state hosting in Pulumi Cloud, 300+ packages, CrossCode translation layer (Terraform, CloudFormation, ARM, K8s YAML → Pulumi languages), plus AI agent Neo.
HashiCorp Terraform 1.16.0 (Aug 26, 2026): Import blocks can now target resources declared inside a module, including for_each module instances – no more flattening modules to root just to import.

Structural shift:

IBM's HashiCorp acquisition accelerated migration interest toward open-source friendly alternatives. Pulumi explicitly targets this cohort.
HashiCorp counters with The Infrastructure Cloud (HCP) – unified ILM + SLM platform for Terraform, Packer, Vault with cross-product workflows and 30-day production proof-of-value.
2026 stack: GitHub Actions over Jenkins, Terraform/Pulumi for multi-cloud IaC, Prometheus-Grafana for monitoring, Istio for service mesh, Vault for secrets remains the consensus.

4. AI Infrastructure Software

Orchestration Layer Becomes the Control Plane

Three launches define AI infrastructure orchestration this week:

Sharon AI + Rafay Systems: 5-year agreement for standardized orchestration spanning bare-metal provisioning, lifecycle management, VMs, Kubernetes, GPU PaaS, AI dev environments, and token-metered inference services – turning raw infra into production-ready AI services.
Clarifai Compute Orchestration: Control plane for governing AI resource access, cost monitoring, and deployment to any hardware vendor across cloud/on-prem/air-gapped/SaaS for inference.
Eluvio at IBC 2026: Inline open-model video AI with 17 built-in models, fast text/vector storage, and MCP API for agentic orchestration across live sports and vertical video.

Why buyers should care:

Infrastructure upgrades for agentic AI: 83% of orgs need infra upgrades for production-grade agents that continuously reason/act across systems, requiring MLOps for continuous inference and distributed edge execution.
GPU utilization, not GPU count, is now the platform KPI.

Inference Economics Dominate Capex

Inference is now the largest driver of compute demand:

IBM + Together AI $240M: Nvidia-powered inference cluster for open-source models, Blackwell chips optimized for inference.
QumulusAI + Runpod: Expansion with 616 NVIDIA Blackwell GPUs for inference capacity.
Megaport: $594M raise to build globally distributed AI inference cloud with on-demand GPU pool ($350M GPU pool) plus CPUs, storage, connectivity.
NVIDIA + MediaTek: NVLink Fusion for custom XPUs integrated into rack-scale AI systems – "AI factories" with prevalidated paths for frontier labs.
AWS Bedrock: Observable agentic retrieval stack with seven layers of observability and continuous evaluation – governed, observable agent patterns as reference architectures.

5. Data Infrastructure

Storage and Lakehouse Converge Around AI

Data infrastructure moves this week show lakehouse + streaming + vector as unified stack:

Lenovo → Infinidat: Acquiring high-end enterprise storage for AI and gen AI workloads, synergy with Infrastructure Solutions Group targeting high-end enterprise.
Databricks → Neon ($1B) and Arcion: Serverless Postgres for AI-native apps plus CDC engine with 20+ enterprise DB connectors to natively ingest into lakehouse.
Snowflake vector push: VECTOR type + VECTOR_COSINE_SIMILARITY + Cortex Search hybrid service with TARGET_LAG auto-refresh; Photon engine competitive on BI queries.
BigQuery + Databricks Vector Search: BigQuery VECTOR_SEARCH with IVF/TREE_AH for 100M+ vectors; Databricks Vector Search with delta-sync auto-indexing, HNSW sub-200ms, Unity Catalog governance.
Postgres as default GenAI DB: Snowflake Crunchy Data $250M, Databricks Neon $1B, Supabase $100M at $5B – pgvector pattern.
Streaming: Confluent Cloud remains the backbone for real-time event pipelines feeding Databricks Delta Lake medallion + Snowflake serving layer, RAG with Pinecone/OpenSearch vector DBs.

Data Center as Infrastructure Software

Physical infrastructure financing now moves with software economics:

Keppel DC REIT $1.2B: Buying 90% effective interests in two Tokyo data centers – 88.62% REIT, 1.38% sponsor Keppel.
Southeast Asia $11.5B funding: Data center companies raised across 19 rounds as of Aug 25, 2026 – Singapore dominates.
ByteDance $29.6B loan: For AI push, competing with local hyperscalers on AI data centers and global hyperscalers on multimodal models.
SLB $4B Kelvion cooling hardware acquisition to double revenue per GW of delivered data center capacity.

6. Observability & Operations

From Monitoring to Autonomous Operations

Observability enters new phase as cost, AI, and scale collide:

Coralogix: $200M late-stage raise + $115M Series E at $1B+ valuation launching Olly agentic assistant – claims RDAF moving beyond traditional monitoring to observability pipelines and AIOps.
OpenObserve: v1.0 RC (Sept 3, 2026), $10M Series A led by Nexus with Dell Technologies Capital – single Rust binary for logs/metrics/traces/session replay, plus MCP support and LLM observability, vision for Observability 3.0 with AI SRE and anomaly detection.
Dabble → Grafana Cloud: Migrated from self-hosted Grafana to Grafana Cloud to cut incident response and control costs.
Middleware: Positioned as an OpenTelemetry-native backend vs Datadog, Grafana, New Relic, Honeycomb.

Buyer guidance:

Choose between managed depth (Datadog/New Relic), open-source flexibility (Grafana/SigNoz/OpenObserve), or OTel purity.
For AI stacks, LLM observability and agent telemetry (prompts, retrievals, evaluations) are now must-haves.

AIOps Consolidation Accelerates

Large platform vendors embedding observability alongside security and AI operations:

LogicMonitor → Unomaly: Advance AIOps by embedding advanced analytics and automation into core IT monitoring – no separate AIOps platform needed.
Dynatrace → DevCycle: Continues push beyond observability into direct automation via Keptn (deployment hooks since 2021) and Davis AI generating K8s deployment resources and Git PRs in response to issues.
Palo Alto: Embrace (July) + Chronosphere $3.35B acquisitions show security vendors buying observability to own the full telemetry + response loop.

7. Security & Resilience

Platform-Layer CVEs Dominate Risk

Infrastructure security flaws this week target the control plane itself:

Ingress controllers – CVE-2026-24512 / 24514 / 24513 (CVSS 8.8 high): Unsafe parsing of user-supplied ingress config / HTTP request data, exposing controller to unsafe config handling and control-plane risk.
External Secrets Operator – CVE-2026-22822 (CVSS 9.3 critical): Insecure Secret Retrieval breaks namespace isolation – fundamental to K8s security – allows cross-namespace secret access.
Argo CD repo-server: Unpatched unauthenticated RCE when internal port is reachable, reported Jan 2025, still unpatched as of Sept 2026 – no CVE assigned. GitOps supply chain risk for platform teams.
AI infra in CISA KEV: 3 of 7 KEV additions target AI/ML infrastructure – LiteLLM AI gateway/proxy (CVE-2026-59822 pattern) first batch where AI components are nearly half.

Resilience notes:

CVE program funding secured – CISA and MITRE renegotiated the contract, eliminating the 2025 expiration panic, ensuring 26-year-old CVE program continuity.
Platform teams should enumerate operator versions: kubectl get deployment -A -l app.kubernetes.io/name=logging-operator -o wide and similar for ingress/secret operators.

Agentic Identity as Infrastructure Security

CrowdStrike Falcon Guardian and Agentic Identity Provider (complement to infra watch) shows security moving from "protect users" to "protect agents and their identities" with cryptographically verifiable agent identities – directly impacts infrastructure policy enforcement for AI workloads accessing K8s, cloud APIs, and data systems.

Boomi Agent Control Plane model – vendor-neutral layer between agents and SAP/Oracle/Salesforce/Workday with identity, rate limits, token caps – is an infrastructure pattern for all AI agent access.
For platform engineering: integrate agent IdP with Vault, OPA/Gatekeeper, and service mesh for runtime enforcement.

8. M&A & Funding

Deal / Move	Parties	Value	Strategic Rationale
AI cloud contract	Crusoe → Jane Street	$13B (5-year)	GPU clusters for AI training/inference; finance as anchor tenant
Pre-IPO raise	Nscale	$3.5B sought	UK neocloud scale for $45B Anthropic-scale deals
Inference cloud raise	Megaport	$594M	Globally distributed AI inference cloud + $350M GPU pool
Data center acquisition	Keppel DC REIT → Tokyo DCs	$1.2B	90% interests in two Tokyo facilities
Storage acquisition	Lenovo → Infinidat	Undisclosed	High-end enterprise storage for AI workloads
Observability funding	Coralogix	$200M + $115M Series E	$1B+ valuation, agentic assistant Olly
Observability funding	OpenObserve	$10M Series A	AI-native Observability 3.0, LLM monitoring
Observability acquisitions	Palo Alto → Embrace, Chronosphere	$3.35B (Chronosphere)	Security + observability convergence
AIOps acquisition	LogicMonitor → Unomaly	Undisclosed	Embed AIOps into monitoring platform
Database acquisition	Databricks → Neon / Arcion	$1B (Neon)	Serverless Postgres + CDC for lakehouse

Pattern:

Inference infrastructure and observability + AIOps are the hottest funding themes – both tied to utilization economics.
Neoclouds are raising at infrastructure multiples but securing AI lab-scale contracts previously reserved for hyperscalers.
Storage consolidation driven by AI density requirements – high-end arrays + data center power/cooling as an integrated bet.

9. The Infrastructure Shift

From Orchestration to Utilization

For five years, infrastructure software sold orchestration – Kubernetes for containers, Terraform for clouds, Databricks for data. The bottleneck was provisioning: how fast can we spin up?

This week's news shows the bottleneck flipped to utilization:

Rafay's KPI is GPU utilization, not GPU provisioning – turning raw infra into token-metered services.
Kubernetes 1.37 fixes memory waste during list ops, not new workload types – reducing the tax of running at scale.
Pulumi's HCL support is about utilizing existing Terraform state, not greenfield language choice.
AWS-Azure private link is about utilizing capacity already paid for across two clouds.
Observability 3.0 is about autonomous action (AI SRE, Olly) to reduce MTTR waste, not just dashboards.

Training was about max cluster size. Inference is about tokens-per-dollar, memory per concurrent request, continuous operation at the edge. That's why $13B deals are now 5-year capacity contracts, not capex, and why every observability vendor is adding LLM evals and agent telemetry. The infrastructure layer beneath the app layer is no longer measured by how fast you can spin up, but by how little you waste once it's running.

Strategic Takeaway

Infrastructure software is becoming the enterprise utilization control plane – the layer that governs cost, security, and performance across multicloud, Kubernetes, data, and AI inference.

For Enterprises: Audit interconnect costs now – native AWS-Azure link will undercut third-party fabrics 30-50% at GA. Prioritize patching for CVE-2026-22822 and CVE-2026-24513 in shared platform components. Adopt token-metered inference abstraction vs raw GPU allocation.
For Developers / Platform Teams: Prioritize K8s 1.37 for large clusters; freeze new IaC modules until the Pulumi HCL vs HCP decision; integrate agent IdP (CrowdStrike/Boomi pattern) with Vault, OPA/Gatekeeper, service mesh for AI workload governance.
For Investors: Money rotating from training clusters to inference fabrics and from monitoring to AIOps automation. Coralogix $1B+, Nscale $3.5B pre-IPO, Crusoe $13B contract show inference clouds commanding SaaS multiples. Storage and data center REITs are AI beta plays with near-term cash flow.

This complements Enterprise Software Watch: Enterprise covers the applications businesses use; Infrastructure covers the platforms that build, run, connect, secure, and scale those applications. This week both converge on control planes – Enterprise on agent control planes, Infrastructure on utilization control planes.

11. What to Watch Next Week

IBC 2026 (Sept 11-14, Amsterdam): Eluvio open-model video AI, MCP for media infrastructure – watch for Nvidia + cloud inference partnerships and agentic orchestration standards.
Kubernetes 1.37 GA rollout: EKS, GKE, AKS timing for RangeStream-enabled managed etcd and deprecation notices for kube-dns/IPVS.
Pulumi HCL GA timeline: Q1 2026 private beta → GA signals will trigger Terraform Cloud migration evaluations.
Nscale $3.5B close: Whether it prices as infrastructure or AI model proxy, and implications for Crusoe/CoreWeave comps.
Argo CD patch: If maintainers finally assign a CVE and ship a fix for the repo-server RCE, expect forced upgrades across GitOps fleets.
Azure revenue breakout Q1 FY27: First quarterly Azure sales number will reset cloud market share narratives vs AWS.
OpenObserve 1.0 GA: Community adoption vs Grafana Cloud/Datadog for AI-native observability with LLM monitoring.

The CODEW Stat

Flexera 2026 State of the Cloud: 84% of enterprise respondents run significant workloads on AWS, 82% on Azure – the statistical foundation for AWS-Azure native multicloud interconnect. Meanwhile, Crusoe's $13B Jane Street deal marks the first non-lab anchor tenant at hyperscaler scale for a neocloud.

The CODEW