Azure Weekly Update - 2nd October 2026

Technological concepts & product updates (Azure Weekly Update – Oct 2, 2026)

Agent “sandbox” for isolated execution

• New video/product(s): Azure Container Apps “sandboxes” (ACA) introduced as a research-backed concept.
• Key idea: even if you don’t use Azure Container Apps, the sandbox tech lets you run agents in an isolated environment at the kernel level, with:
  • a separate outbound proxy for controlled communication
  • isolation intended to prevent “agents breaking out of sandboxes”
• Practical positioning: Microsoft internal agent hosting (e.g., Foundry hosting agents) uses similar technology; now it’s available for customers.
• Use case: build your own orchestrator, but still run agents safely—this sandbox becomes the execution layer behind the scenes.

---

Decommissioning / retirement guidance (migration planning)

• Azure IoT Central decommissioning: scheduled for September 2029
  • Guidance: assess current IoT capabilities and migrate to more scalable options such as:
    • Azure IoT Hub + DPS for device connection and provisioning
  • Mentioned alternative: Fabric real-time analytics capabilities.
• Microsoft HPC Pack: retired around next year (relative timeline in the update)
  • Suggested alternative: Azure Batch (managed HPC + parallel scheduling; not a one-to-one replacement).
• SQL / compute SKU support endings
  • D & E V3 support ends November 2029 → upgrade to newer models.
• Confidential compute series
  • DC series decommission: Oct 31, 2029
• Other SKUs scheduled to be out soon
  • NVv3/V4 due Sept 2026 → “give them up as soon as possible”
• Container Apps Function hosting V1
  • Decommissioning: in 2027
  • Need: migrate to V2 (native Azure Container Apps) using a migration assistant
• Azure Arc for SCVMM
  • Decommissioning: Sept 2029
  • Plan: use Arc-enabled servers instead

---

Compute & infrastructure features

• Automatic zoning for VMSS (preview)
  • Select desired zones; VMSS auto-selects placement based on SKU availability/capacity and placement requirements
  • Balances new deployments (does not rebalance existing resources)
• New VM SKUs GA
  • LASv5 and O-indexed variant
  • Storage-optimized, based on 5th Gen AMD EPYC
  • Emphasis: very large local NVMe and high internal bandwidth:
    • ~720 GB local NVMe per virtual processor (O variant) vs ~240 GB (regular)
    • ~200 GB/s network bandwidth
  • Workload guidance: caching, search, distributed file systems, big data, databases—use local storage with replication for reliability
• Azure Linux / SQL deployment improvements
  • SQL deployment on Azure Linux VMs updated in preview:
    • shift from custom images to using a deployment script with standard images + install/config workflow
• Azure Extended Zones
  • Luxembourg added as a public extended zone for ultra-low latency and strict data placement needs
• Azure Local (public)
  • Edge option for running SQL Server via Azure Arc
    • supports management-driven local SQL instance placement (also in standalone mode with local control plane)

---

Storage updates

• Azure Elastic SAN + Azure Backup integration (GA)
  • Backup/restore integration now supports Elastic SAN
  • Separates storage protection from Elastic SAN itself
  • Offers up to 450 restore points
  • Pricing described as: full initial snapshot size + incremental change costs later
  • Motivated by protection against accidental/intentionally deletion
• Azure NetApp Files
  • SMB volumes support intra-realm Kerberos authentication (preview)
    • reduces need for legacy/direct-attached AD domain controllers for auth
  • Cool access GA
    • QoS for Premium/Ultra tiers with bandwidth adjustments based on Cool tier data amount
  • Large volume mode GA
    • up to 2 PB volumes
    • throughput up to ~80 GiB/s
    • targeted at high-end compute: ultra-high throughput/low latency with isolation from noisy neighbors
• NetApp Files feature note
  • “Cool tier” moves inactive data to cheaper storage while maintaining adaptive throughput behavior.

---

Database product updates (major section)

Azure Database for PostgreSQL (“Horizon DB”)

• PostgreSQL 18 support: version 18 in preview
• Long-Term Storage V2: in preview
  • Changes storage approach:
    • from logical backups (pg_dump/pg_restore) to physical snapshots based on managed disks
    • deeper integration with Azure Backup vaults
  • Benefits highlighted:
    • much larger databases (example given: ~64 TB on Premium SSD v2; less on v1 vs older ~1 TB)
    • daily backups instead of weekly
    • incremental after the first full backup
    • restore directly to the server (previously restored to storage then imported)
    • retention still 10 years
• Client-managed keys across tenants (GA)
  • Customer keys stored in another tenant’s Key Vault, used by PostgreSQL in a different tenant
  • Especially useful for ISVs needing customer-controlled encryption for a PostgreSQL “trusted state layer”
• Ultra Disk support
  • Now available with PostgreSQL flexible server
  • Target: lowest latency, highest I/O and throughput

Azure Database for MySQL

• Read endpoints (feature evolution)
  • With up to 10 replicas per primary, a read endpoint distributes load across available replicas
  • Improves scalability and fault tolerance by routing to available replicas

Azure SQL / Azure SQL Managed Instance

• Azure SQL MI GA features
  • Flexible memory
    • separate memory sizing “slider” without changing vCores
    • available for General Purpose and Business Critical on Premium series
  • Business Critical MI transaction log write throughput increased (GA)
• Portal / operational improvements
  • Mentions: new Hyperscale Premium larger SKUs, automatic index compaction, and SQL query editor for schema comparison
  • Managed performance monitoring preview for:
    • Azure SQL Database and SQL Server on Azure VMs
    • goal: pipeline/dashboarding for performance issues and trends
• Azure SQL Dev Hub (preview)
  • Developer-focused starting point for building SQL Database work
  • Positioned for SQL + AI/agents in different languages
• Azure SQL vector capabilities
  • Vector capabilities GA
    • built-in vector index creation + vector search using Microsoft Research’s DiskANN
    • embeddings convert semantic meaning into multidimensional vectors
    • retrieval via nearest-neighbor (semantic similarity, not keyword matching)
• Hyperscale serverless auto pause/resume (preview)
  • Automatic pause and resume to reduce compute costs
  • Billing stops when paused; resumes on most DB interactions (auth, queries, service, data access)
• SQL Agent “skills” (preview)
  • SQL Agent skills help AI agents execute SQL-related tasks
  • Includes skills for planning migrations to Azure SQL Database and Azure SQL MI
• Azure Arc for SQL Server migration (preview)
  • Unified portal for:
    • discovery, assessment, migration of Arc-enabled SQL Server instances
    • migration into Azure SQL Database / Azure Hyperscale
  • Performance monitoring improvements for Arc-enabled SQL Server
  • Region availability expanded (noted: Germany West-Central and Italy North for sovereignty)
• SQL Formatter GA
  • SQL Formatter (VS Code extension part of MS SQL extension)
  • Focus: consistent formatting (alignment, case/operator styles, multiline lists, unions)

---

Data platform / AI enabling concepts: Microsoft Fabric + OneLake

• Fabric Database Hub (announced earlier)
  • Central management for multiple database types (SQL Server, Azure VM/Arc-enabled, Azure SQL, PostgreSQL, Cosmos DB, Fabric SQL instances)
  • Supports discovery, creation, management, monitoring
  • Includes Copilot analytical tools for “what to pay attention to” and “what to do next”
• Data virtualization via OneLake
  • Motivation: AI can’t easily integrate with many scattered silos
  • Approach: present data through a single virtualization layer, leaving data in place or mirroring it
  • OneLake enables semantic models for enterprise entities
  • New/expanded sources noted:
    • SharePoint lists and Dremio mirroring earlier
    • Oracle + SAP Data Sphere mirroring now GA
    • Azure Databricks can read OneLake via Unity Catalog
    • Databricks can interact with Snowflake and other capabilities

---

Copilot / models availability

• GPT 6.1 Seoul (“Sonic 55”) made available in Copilot experiences:
  • Copilot chat, Word, Excel, PowerPoint, Code Work, Copilot Studio, Foundry, etc.
• Positioning:
  • GPT 6.1 Seoul: advances agent coding, computer usage, and professional work
  • Sonic 55: more intelligence with fewer tokens, “clear boundaries” for coding
• Operational guidance emphasized:
  • need evaluation systems, security measures, and structured organization for trustworthy AI outputs
  • ensure the system is fed appropriate data (performance, business status, knowledge bases, Internet) to adapt safely to new model versions

---

Main speakers / sources (as implied by subtitles)

• Speaker: the host delivering “Azure Weekly Update” (name not stated in subtitles).
• Primary product teams/technology sources referenced:
  • Microsoft teams working on Azure Container Apps / ACA sandbox technology
  • Microsoft Research (DiskANN algorithm for vector search)
  • Azure Arc framework (for SQL Server migration/management)