Executive Summary: Microsoft Azure AI Infrastructure

Microsoft Azure is the most structurally complex vendor in this assessment series because it operates three distinct authority systems simultaneously: a massive cloud infrastructure platform (Azure), a deeply integrated but newly non-exclusive frontier model partnership (OpenAI), and the largest enterprise software installed base on earth (Microsoft 365, Entra ID, Purview, Fabric). No other assessed vendor straddles all three domains. Google Cloud owns its frontier model outright. AWS partners with model providers at arm's length. Dell, HPE, and VAST operate below the model layer entirely. Microsoft is the only vendor that must coordinate authority across infrastructure, model intelligence, and enterprise application identity, and the April 2026 OpenAI restructuring has made that coordination both more flexible and more visible.

The DAPM classification for Azure reveals a paradox unique among the assessed vendors: Microsoft has more productized Layer 2C capability than any vendor except Google: Agent Governance Toolkit (open-source, sub-millisecond policy enforcement), Entra Agent ID (GA April 2026, agent identity as first-class Entra citizens), Microsoft Agent 365 (unified agent registry and control plane), Foundry Control Plane (centralized observability for agents across frameworks), yet the enterprise Cedes the most judgment to consume it. The Layer 2C surface is real. The authority delegation is also real. Both statements are true simultaneously.

The OpenAI restructuring (April 27, 2026) is the most significant borrowed judgment event in this assessment series. Azure exclusivity is gone: OpenAI models now ship on AWS Bedrock the next day. Microsoft retains a four-month first-mover window on new frontier models, IP license through 2032, ~27% equity stake, and OpenAI's commitment to $250B in Azure consumption. But the structural dependency has shifted from contractual lock-in to commercial preference. Microsoft is simultaneously scaling its own model development (MAI-1, speech/image models via Mustafa Suleyman's CoreAI division), hedging the borrowed judgment it once embraced unconditionally.

The enterprise identity story is Azure's most underappreciated differentiator. No other vendor has extended enterprise identity governance to AI agents. Entra Agent ID treats agents as identity citizens alongside humans and workloads: same Conditional Access, same lifecycle management, same risk detection. This is Layer 2C infrastructure that every other vendor will eventually need to build or integrate. Microsoft has it because it already owns the enterprise identity plane. Identity is a cross-cutting concern not fully addressed in earlier assessments in this series. The Azure assessment surfaces it as a structural dimension that should be retroactively evaluated across all vendors.

Azure's structural question is not capability. The capabilities span every layer of the 4+1 model. The question is authority composition: when the enterprise adopts Azure AI Foundry + OpenAI models + Entra Agent ID + Fabric data governance + Purview compliance, how many independent judgment systems has it inherited, and has it classified each delegation explicitly? The 4+1 model exists to make that composition visible. Azure is the vendor where the composition is most complex.

Layer-by-layer status: Layer 0A (ND-Series NVIDIA and AMD Instances, Rented by the VM; Azure Local On Premises), Layer 0B (InfiniBand RDMA for GPU Clusters on a SONiC-Switched Network), Layer 0C (Azure Managed Lustre Feeds the Accelerators), Layer 1A (Purview Classification and Lineage Over OneLake), Layer 1B (Managed Retrieval (AI Search)), Layer 1C (Managed Pipelines (Fabric/ADF)), Layer 2A (AKS + Arc Hybrid Orchestration), Layer 2B (Foundry + OpenAI Runtime), Layer 2C (Complete Intelligence-2C Plane Across Products; Gateway Architect-Wired), Layer 3 (+1) (Broadest Enterprise Ecosystem).

Assessment framework: 4+1 Layer AI Infrastructure Model. Scoring model: Decision Authority Placement Model (DAPM) — Retained, Delegated, or Ceded. Published by The CTO Advisor LLC (DBA The Advisor Bench). Author: Keith Townsend. Date assessed: October 27, 2026. Version: v1.41 - 4+1 v3: Layer 0 Sub-Layers.

Microsoft Azure AI Infrastructure

Mapped to the 4+1 Layer AI Infrastructure Model

v1.41 - 4+1 v3: Layer 0 Sub-LayersAssessed October 27, 2026Sources & revision history
ACTIVE ASSESSMENT

Summary Finding

Microsoft Azure is the most structurally complex vendor in this assessment series because it operates three distinct authority systems simultaneously: a massive cloud infrastructure platform (Azure), a deeply integrated but newly non-exclusive frontier model partnership (OpenAI), and the largest enterprise software installed base on earth (Microsoft 365, Entra ID, Purview, Fabric). No other assessed vendor straddles all three domains. Google Cloud owns its frontier model outright. AWS partners with model providers at arm's length. Dell, HPE, and VAST operate below the model layer entirely. Microsoft is the only vendor that must coordinate authority across infrastructure, model intelligence, and enterprise application identity, and the April 2026 OpenAI restructuring has made that coordination both more flexible and more visible.

The DAPM classification for Azure reveals a paradox unique among the assessed vendors: Microsoft has more productized Layer 2C capability than any vendor except Google: Agent Governance Toolkit (open-source, sub-millisecond policy enforcement), Entra Agent ID (GA April 2026, agent identity as first-class Entra citizens), Microsoft Agent 365 (unified agent registry and control plane), Foundry Control Plane (centralized observability for agents across frameworks), yet the enterprise Cedes the most judgment to consume it. The Layer 2C surface is real. The authority delegation is also real. Both statements are true simultaneously.

The OpenAI restructuring (April 27, 2026) is the most significant borrowed judgment event in this assessment series. Azure exclusivity is gone: OpenAI models now ship on AWS Bedrock the next day. Microsoft retains a four-month first-mover window on new frontier models, IP license through 2032, ~27% equity stake, and OpenAI's commitment to $250B in Azure consumption. But the structural dependency has shifted from contractual lock-in to commercial preference. Microsoft is simultaneously scaling its own model development (MAI-1, speech/image models via Mustafa Suleyman's CoreAI division), hedging the borrowed judgment it once embraced unconditionally.

The enterprise identity story is Azure's most underappreciated differentiator. No other vendor has extended enterprise identity governance to AI agents. Entra Agent ID treats agents as identity citizens alongside humans and workloads: same Conditional Access, same lifecycle management, same risk detection. This is Layer 2C infrastructure that every other vendor will eventually need to build or integrate. Microsoft has it because it already owns the enterprise identity plane. Identity is a cross-cutting concern not fully addressed in earlier assessments in this series. The Azure assessment surfaces it as a structural dimension that should be retroactively evaluated across all vendors.

Azure's structural question is not capability. The capabilities span every layer of the 4+1 model. The question is authority composition: when the enterprise adopts Azure AI Foundry + OpenAI models + Entra Agent ID + Fabric data governance + Purview compliance, how many independent judgment systems has it inherited, and has it classified each delegation explicitly? The 4+1 model exists to make that composition visible. Azure is the vendor where the composition is most complex.

Strength
Moderate
Gap
Partner
Layer 0A · ComputeComputeND-Series NVIDIA and AMD Instances, Rented by the VM; Azure Local On Premisesdecides: vendor · Ceded▼

CPUs, accelerators, servers, integrated systems, OS and hypervisor, physical plant

Vendor-Provided

Cobalt 100 (Arm CPU VMs)Retained

Azure-designed Arm CPU for general cloud workloads, rented as VMs. Arm is a multi-vendor instruction set, so binaries built here run on other Arm CPUs (Graviton, Cobalt, Axion, Ampere, Arm OEM servers) without a rebuild: Retained, the commodity x86 reading. Who designs the core doesn't place it (Arm CPU ruling, October 4, 2026).

NVIDIA GPU Instances on Azure (ND H100 v5, ND H200 v5, NCads H100 v5)Retained

ND H100 v5 and ND H200 v5 VMs with eight GPUs, and NCads H100 v5 with "up to 2 NVIDIA H100 NVL GPUs" per Microsoft Learn. Fractional GPU via Azure Kubernetes Service. Standard instances on third-party silicon: a CUDA or PyTorch workload moves to another cloud or an OEM server without a rebuild. Retained under the Layer 0 runtime rule, read from the customer's seat; the silicon owner's runtime is read on the NVIDIA row. The multi-accelerator marketplace (NVIDIA and AMD, with Maia behind Azure's own services) creates a workload-to-silicon matching problem that is itself a Layer 2C function.

NVIDIA NVL-Class Rack-Scale Capacity on Azure (ND GB200 v6 and ND GB300 v6 on GB200 and GB300 NVL72)Ceded

ND GB200 v6 and ND GB300 v6 VMs on GB200 and GB300 NVL72 racks, four Blackwell GPUs per VM and 18 VMs to a 72-GPU rack, with 800 Gbps InfiniBand per GPU on GB300 v6. Ceded on the integrated-system rule, the Lenovo NVL-class reading: a rack-scale system built around a captive NVLink fabric can't be lifted as deployed.

Azure Local: Hyperconverged Cluster Compute and Hypervisor (On-Prem)Ceded

Hyper-converged, customer-owned cluster running Azure services on-premises. Windows/Linux VMs, AKS containers, Azure Virtual Desktop. Connected and fully disconnected modes (Feb 2026). Validated OEM hardware from Dell, HPE, Lenovo, and others. Azure Arc extends management, governance, and security. Foundry Local for on-prem AI inference. $10/core/month + optional services. Sovereign Private Cloud (Azure Local + Microsoft 365 Local) for air-gapped environments.

Azure Boost (Virtualization Offload)Ceded

Microsoft's offload system, which "offloads server virtualization processes traditionally performed by the hypervisor and host OS" onto purpose-built software and hardware, including local, remote and cached disk storage. Azure Boost runs under the current ND, NC and general-purpose VMs. Ceded with the platform, the Nitro and Titanium reading.

AMD Instinct MI300X Instances on Azure (ND MI300X v5)Retained

AMD Instinct MI300X via ND MI300X v5, eight GPUs per VM, which Microsoft says "was designed for high-end Deep Learning training" and scale-out generative AI. A standard instance on third-party silicon: a ROCm or PyTorch workload moves to another cloud or an OEM server without a rebuild, Retained; AMD's runtime is read on the AMD row.

NVIDIA-Provided

NVIDIA GPU Silicon

ND-series H100 and H200, and GB200 and GB300 on ND GB200 v6 and GB300 v6. Microsoft manages the NVIDIA integration and instance types. The CUDA dependency is read on NVIDIA's row.

◆ Gap Analysis

Azure rents NVIDIA H100 and H200 on its ND-series VMs and AMD Instinct MI300X on ND MI300X v5. Its ND GB200 v6 and GB300 v6 VMs run on GB200 and GB300 NVL72 rack-scale systems. Cobalt 100 is Azure's own Arm CPU, and Azure Boost offloads virtualization from the host. The buyer gets per-workload silicon choice with no procurement cycle. That choice recurs with every workload, and matching workload to silicon is itself a Layer 2C function. The architect's concern is the runtime and the rack. A CUDA, ROCm or PyTorch workload on standard ND instances keeps its exit; the NVL72 racks are integrated systems on a captive NVLink fabric. Azure Local runs Azure services on validated servers the enterprise buys from Dell, HPE, Lenovo and other OEMs, but those services run only on Microsoft's hypervisor and Storage Spaces Direct. Calibration: AWS and GCP read strong on the same rented shape; Dell and HPE strong on hardware the enterprise owns.

◆ Borrowed Judgment

Rented, so the runtime call is Azure's. Azure decides which host a VM lands on and when maintenance moves it, and the enterprise picks a VM size and reverses none of it: vendor decides, visible, not overridable, Ceded. Portability splits: the ND instances and Cobalt read Retained; the NVL72 racks, Azure Boost and the Azure Local stack read Ceded.

◆ Working Notes

Named, not scored (October 5, 2026): Maia 100 runs Azure OpenAI Service and Copilot and isn't rented to the enterprise, and under the vendor-shape ruling silicon behind a service earns nothing at Layer 0. Maia 200 remains in development. Vera Rubin NVL72 on Azure is announced, with no Rubin size on Microsoft Learn (read October 22, 2026), so it's named, not scored. Not a chip (October 22, 2026): the validated servers under Azure Local are sold on Dell, HPE and Lenovo paper, not Microsoft's, and are read on those rows.

Layer 0B · FabricNetwork FabricInfiniBand RDMA for GPU Clusters on a SONiC-Switched Networkdecides: vendor · Ceded▼

Interconnects, switching and routing, network virtualization

Vendor-Provided

ND-Series InfiniBand (RDMA on a Non-Blocking Fat Tree; 800 Gbps per GPU on ND GB300 v6, Up to 100,000 GPUs)Ceded

The RDMA network that ND-series and HPC VMs reach through a secondary interface: a full fat-tree design, non-blocking with bisectional symmetry, with RDMA only over InfiniBand and never over the Ethernet network. ND GB300 v6 gives each GPU 800 Gbps, to up to 100,000 GPUs. Azure runs the fabric and makes its routing and congestion calls. Ceded, the EFA and GCP RoCE GPU network reading.

Accelerated Networking (SR-IOV Front-End NICs on NVIDIA ConnectX or MANA)Ceded

The VM front end: Accelerated Networking enables SR-IOV, so traffic skips the host and the virtual switch while the network policy is enforced in hardware, on NVIDIA ConnectX or the Microsoft Azure Network Adapter (MANA). On ND GB300 v6 it's the one 160 Gbps front-end NIC, apart from the scale-up and scale-out NICs. Ceded: the policy offload and the adapter are Azure's to run.

ND GB300 v6 (GB300 NVL72): NVLink FabricCeded

The captive NVLink fabric inside the GB300 NVL72 racks ND GB300 v6 runs on, 18 VMs and 72 GPUs to a rack with 130 TB/s of intrarack NVLink bandwidth. Ceded with the system on the integrated-system rule.

Azure Virtual Network (Subnets, User-Defined Route Tables, Network Security Groups, Peering, BGP via VPN Gateway and ExpressRoute)Ceded

Private networks for Azure resources: subnets spanning a region's availability zones, custom route tables per subnet that override Azure's default routes, network and application security groups, VNet peering, and BGP route propagation from on-premises.

NVIDIA-Provided

NVIDIA InfiniBand, ConnectX and BlueField

InfiniBand carries the GPU cluster interconnect, with ConnectX and BlueField NICs.

NVLink in Vera Rubin NVL72

The rack's NVLink domain is NVIDIA's, captive inside the integrated system.

◆ Gap Analysis

Azure joins its ND-series GPU clusters on InfiniBand with RDMA for distributed training, and Accelerated Networking gives VMs hardware-level SR-IOV. Underneath runs SONiC, the network OS Microsoft created and open-sourced through the Open Compute Project (OCP), now on switches from Broadcom, NVIDIA, Intel and others. Virtual Network wraps it in subnets, route tables and network security groups the enterprise writes. The buyer gets a training fabric without buying a switch. The architect's concern is who runs it. SONiC being open source is a portability fact about switches, not an authority fact about Azure. Microsoft runs it. The NVL72 racks add a captive NVLink fabric inside the rack. Calibration: AWS's EFA and GCP's RoCE GPU network read strong on the same shape; HPE's Slingshot and NVIDIA's Spectrum-X strong as fabrics the enterprise buys.

◆ Borrowed Judgment

Ceded where the grade rests. The route tables and security groups are the enterprise's code, but the grade rests on the InfiniBand fabric, which makes the routing and congestion calls with no per-decision override: vendor decides, visible, not overridable, Ceded. Portability: the InfiniBand fabric and Accelerated Networking are Azure's to run, Virtual Network is configured through Azure Resource Manager only, and the NVLink fabric goes with the rack, Ceded.

◆ Working Notes

Named, not scored (October 22, 2026): Vera Rubin NVL72 on Azure; Microsoft Learn's sizes overview (updated September 25, 2026) lists no Rubin size. SONiC is open source Microsoft created and gave away through the Open Compute Project. Azure doesn't sell it, so it reads in the gap prose as a portability fact about switches, not as a chip. The rack architecture, power, cooling and region counts that sat in the old chip are 0A facts.

Layer 0C · StorageStorage SubstrateAzure Managed Lustre Feeds the Acceleratorsdecides: vendor · Delegated▼

Arrays, storage servers and software, front-end protocols: where the data sits

Vendor-Provided

NFS 3.0 on Azure Blob Storage + ADLS Gen2 (Hierarchical Namespace Accounts)Delegated

NFS 3.0 mounts of Blob containers on accounts with hierarchical namespace enabled, a protocol independent file servers implement: the mount configuration changes to another NFS server, and the data migrates. Delegated. The storage service behind it, with its tiers, immutability, versioning and lifecycle rules, is the Blob Storage chip, Ceded.

Azure Blob Storage + ADLS Gen2, through the Azure Storage REST API, CLI, PowerShell and client librariesCeded

Object and data lake storage. Hierarchical namespace for analytics workloads. Hot/Cool/Cold/Archive tiers. Immutable blobs, versioning, lifecycle management. Accessed through the Azure Storage REST API, CLI, PowerShell and client libraries, with SFTP and NFS 3.0 mounts; no S3-compatible API is documented. The storage substrate for OneLake, Fabric, and AI training data pipelines. The listed capabilities vary by account configuration (versioning isn't available on hierarchical-namespace accounts). The service, its tiers and lifecycle management, and the REST API and client libraries are Microsoft's: Ceded. NFS 3.0 at the surface is its own chip.

Azure Local: Hyperconverged Storage (On-Prem)Ceded

The storage half of the hyperconverged Azure Local cluster, on validated OEM hardware. A hyperconverged storage OS is 0C, because a component sits where its decision is made.

Azure Managed Disks (Premium SSD, Premium SSD v2, Ultra, Standard SSD; LRS or ZRS)Ceded

Block disks for Azure VMs. LRS writes three copies in one data center; ZRS replicates synchronously across three availability zones (Premium SSD and Standard SSD only). Shared ZRS disks support clustered applications across zones.

Azure Managed Lustre (Durable File Systems on LRS Premium SSD; Blob Storage HSM Integration)Ceded

Fully managed Lustre file system for HPC and AI training, OSTs on LRS Premium SSD managed disks, Blob Storage integration through Lustre HSM for import and export, and an AKS CSI driver. The Lustre client is its own chip. This is the service Microsoft runs: the tiers, the Blob HSM links and the CSI driver, which don't lift to another Lustre. Ceded.

Azure NetApp Files (NFSv3, NFSv4.1, SMB, Dual-Protocol Volumes; Five Service Levels; Cross-Region and Cross-Zone Replication)Ceded

Microsoft's first-party file service on a bare-metal fleet powered by NetApp ONTAP, billed through Azure: volumes in capacity pools, snapshots and vaulting, cloning, cool access to an Azure Storage account, and an S3-based object REST API. Microsoft sells it and runs it, and it decides at runtime. The opinions accumulate in ONTAP, so it reads Ceded to NetApp on portability, as ONTAP does through any seller; NFS and SMB at the surface are their own chip.

Lustre Client Protocol on Azure Managed LustreDelegated

Mounted with the open Lustre client, which DDN, HPE and the other clouds' managed Lustre services share: the client and its tooling stay usable against another Lustre, while service topology, stored layouts, backups and integrations are revalidated or recreated. Delegated. The managed service is the Azure Managed Lustre chip, Ceded.

Front-End Protocols on Azure NetApp Files (NFSv3, NFSv4.1, SMB, S3-Compatible Object REST API)Delegated

NFSv3, NFSv4.1 and SMB, and the S3-compatible object REST API, which Microsoft documents with six operations (ListBucket, ListObjects, GetObject, PutObject, DeleteObject, HeadObject) over the same files, which independent implementations share: hosts and applications re-point to another server or target speaking the same protocols, and the data and provider-specific policies migrate with it. Delegated. Ceded behind them: ONTAP's data management is NetApp's, and the managed service's control plane is Microsoft's.

NVIDIA-Provided

Assessment pending

◆ Gap Analysis

Azure covers the substrate end to end. Blob Storage and ADLS Gen2 are the object and data lake foundation under OneLake and Fabric, and Managed Disks handle block. Azure NetApp Files serves NFS and SMB as a first-party service on NetApp ONTAP. Azure Managed Lustre is the parallel file system for training. Microsoft publishes its MLPerf Storage v3.0 results, at about 96 to 97% of nominal bandwidth across five tiers. The buyer gets AI-throughput storage sized to the workload. The cell rests on Managed Lustre, and that's where the enterprise keeps a handle: clients set a file's layout with lfs setstripe. Underneath, Blob durability placement takes no per-object override, and Azure Local's Storage Spaces Direct runs only under Azure Local's stack. Calibration: AWS and GCP read strong on managed Lustre too; DDN and WEKA strong on the same function sold as software or appliances.

◆ Borrowed Judgment

Delegated where the grade rests. Managed Lustre places data across its storage targets, and the enterprise can override a specific file's layout: vendor decides, visible, overridable, Delegated. Portability: the Lustre client is a multi-vendor standard, Delegated; Azure NetApp Files reads Ceded to NetApp, as ONTAP does through any seller, whatever its NFS and SMB surface; Blob splits by interface, Delegated through NFS 3.0 and Ceded through its native REST API and client libraries; Managed Disks and Azure Local's storage read Ceded.

Layer 1A · DataData Foundation & GovernancePurview Classification and Lineage Over OneLakedecides: vendor · Delegated▼

The data itself, organized and governed: the Governance Catalog that Layer 2C queries

Vendor-Provided

Microsoft Fabric / OneLakeCeded

Unified SaaS data platform: data engineering, warehousing, real-time analytics, data science, Power BI, all on a single data lake (OneLake). Zero-copy access patterns: data remains in place while being reused across experiences. FabCon/SQLCon 2026 positioned Fabric as 'the operating system for enterprise data' and the central control plane for OneLake data management. EXPOSE TO FABRIC T-SQL extension virtualizes database objects in OneLake without moving data. The Delta Lake surface in front of it is its own chip.

Microsoft Purview (Governance)Ceded

Unified data governance across clouds, data types, and the full data estate. Built into Fabric. Automated sensitivity labeling on OneLake assets. Data Loss Prevention policies detect and restrict sensitive data uploads. Audit logs capture all Fabric user activities including AI interactions. Classification, lineage, compliance (GDPR, HIPAA, PCI DSS). Sensitivity labels, access policies, and compliance controls extend to data shared across tenants via OneLake data sharing. Purview extends beyond Azure into Microsoft 365 (SharePoint, Teams, Exchange), on-premises SQL Server, and multi-cloud environments.

Delta Lake Tables in Microsoft Fabric OneLakeDelegated

Delta Lake has independent implementations: code and definitions written to it move to another compatible implementation within the features both support, the data moves by export, replication or rewrite, and the provider's control-plane configuration and operational state don't lift. Delegated. Compatible engines read or register the Delta tables within Fabric's documented Delta feature matrix (Warehouse tables are read-only from other engines); Fabric pipelines, workspace objects, policies and T-SQL extensions don't lift. The service behind it is its own chip, Ceded.

NVIDIA-Provided

Assessment pending

◆ Gap Analysis

Azure's 1A is among the most expansive on the instrument. Fabric's OneLake is the single logical lake across the tenant, and Purview classifies, labels and traces lineage across the estate. Purview reaches past Azure into Microsoft 365, on-premises SQL Server and other clouds, so the catalog a reasoning plane would query already describes data that never touched Azure. The buyer gets one governance plane over the whole data estate. The architect's concern is the query path. Governance as compliance reporting and governance as runtime policy input are different functions, and it's unclear whether Purview's metadata is reachable programmatically for real-time placement decisions. Blob Storage and ADLS Gen2 hold the bytes, a 0C function. The grade rested on Purview, so it doesn't move.

◆ Borrowed Judgment

Purview's Data Map classifies automatically while it scans, inferring from the data, and each classification can be edited afterward; auto-labeling won't replace a manually applied higher-priority label: vendor decides, visible, overridable, Delegated, the Snowflake and Databricks classification reading. Portability: Fabric, OneLake and Purview are Microsoft's, Ceded.

Layer 1B · RetrievalContext Management & RetrievalManaged Retrieval (AI Search)decides: vendor · Delegated▼

Low-latency retrieval for RAG — vector/hybrid search, context windows

Vendor-Provided

Azure AI SearchCeded

Enterprise retrieval engine combining full-text search (BM25/Lucene), vector search (HNSW), hybrid search with Reciprocal Rank Fusion (RRF), and transformer-based semantic ranker, all in a single managed platform. Integrated vectorization with built-in chunking and embedding. Agentic retrieval (preview): LLM-assisted query planning, multi-source access, structured responses optimized for agent consumption.

Enterprise + Web Knowledge Grounding (Azure AI Search, Grounding With Bing Search, File Search Over Blob Storage)Ceded

Foundry Agent Service agents ground on an Azure AI Search index, on Grounding with Bing Search, and on file search over the Azure AI Search and Azure Blob Storage resources connected at setup, all unbadged on the Foundry tool pages. The SharePoint tool ("Use SharePoint tool with the agent API (preview)"), the Fabric data agent, and Grounding with Bing Custom Search are marked preview and sit in the notes, so the Microsoft 365 corpus isn't a scored agent context source today. Bing adds web-scale grounding beside the enterprise's own indexes.

NVIDIA-Provided

Assessment pending

◆ Gap Analysis

Azure AI Search is one of the strongest Layer 1B offerings in this assessment. The hybrid search + semantic ranker + agentic retrieval combination addresses the retrieval quality problem comprehensively. The structural contrast with Google's Knowledge Catalog is the key 1B finding. Google's approach is model-integrated: Gemini enriches raw data on arrival, extracting business semantics and building a context graph that didn't exist before. Microsoft's approach is corpus-integrated: the M365 corpus already contains business context because humans created it in business workflows. A SharePoint document about Q3 revenue already carries the business semantics that Knowledge Catalog would need Gemini to extract from a raw CSV. Microsoft doesn't need a model to derive meaning because the data was born semantic. Both approaches have trade-offs. Google's model-derived semantics are consistent and exhaustive: every data asset gets enriched. Microsoft's human-created semantics are richer but inconsistent: the quality depends on how well the enterprise organizes its M365 content. Google's approach works on any data. Microsoft's advantage depends on the enterprise already having its knowledge in M365. The agentic retrieval mode (preview) blurs the boundary between retrieval (1B) and reasoning (2C): the retrieval engine uses an LLM to decompose complex queries into sub-queries. Same cross-layer blurring observed in other vendors' products. Bing grounding introduces a unique borrowed judgment: the enterprise inherits Bing's web index, coverage, biases, and ranking decisions as part of agent context. No other vendor has this dependency because no other vendor owns a web search engine.

◆ Borrowed Judgment

Ceded with high integration value. Azure AI Search is Microsoft IP. The semantic ranker is a Microsoft model. Agentic retrieval uses Microsoft's LLM for query decomposition. The enterprise Cedes retrieval intelligence to Microsoft but gains the integrated M365 corpus and Bing web index as context. The M365 corpus as borrowed judgment: the enterprise's own data is the context source, but Microsoft controls how it's indexed, chunked, embedded, and served to agents. The enterprise created the content; Microsoft controls the retrieval path. Decision authority under the override rule (October 5, 2026): semantic ranking is off by default and set per query (queryType=semantic or semanticQuery), results return @search.rerankerScore, and agentic retrieval returns its query plan in the activity array. Vendor decides, visible, overridable, Delegated, the Snowflake and Databricks 1B reading.

◆ Working Notes

The 'retrieval as reasoning' evolution (agentic retrieval mode) is worth tracking as a 4+1 model observation. If the retrieval engine uses LLM reasoning to plan queries, where does Layer 1B end and Layer 2C begin? The product boundary (Azure AI Search) doesn't align with the architectural boundary (retrieval vs. reasoning). The Google Knowledge Catalog contrast deserves tracking as both approaches mature. If Google's Gemini-derived semantics prove more reliable than human-created M365 content for agent grounding, the model-integrated approach wins. If enterprise-specific context (internal jargon, organizational knowledge, relationship context) proves more valuable than machine-extracted semantics, the corpus-integrated approach wins. The answer likely varies by use case. GA evidence table (October 24, 2026), per named item: the Foundry SharePoint tool ("Use SharePoint tool with the agent API (preview)"), the Fabric data agent, and Grounding with Bing Custom Search are marked preview and come out of the grounding chip, so the Microsoft 365 corpus discussed above is direction, not a scored agent context source, until the SharePoint tool is affirmed GA.

Layer 1C · PipelinesData Movement & PipelinesManaged Pipelines (Fabric/ADF)decides: code · Retained▼

Move/transform data — ETL/ELT, lineage, cost-aware movement, KV cache tiering

Vendor-Provided

Azure Data Factory + Fabric Data PipelinesCeded

Cloud-based ETL/ELT orchestration available as a standalone Azure service (Data Factory) and within the unified Fabric experience (Data Factory in Fabric). 200+ connectors in Fabric. Visual pipeline builder. Gartner Leader for Data Integration Tools (5 consecutive years). Fabric version adds Dataflows Gen2 for self-service data preparation with Power Query (no-code), Medallion architecture (Bronze → Silver → Gold) with Delta Lake, ACID transactions, schema enforcement, time-travel queries, Data Activator for event-driven actions, and Copilot for natural-language pipeline authoring. Scheduled and event-based triggers.

Azure DatabricksCeded

Apache Spark-based analytics platform. Delta Lake as the transactional storage layer. Data engineering, data science, ML. Azure-optimized but Databricks-owned IP, and under the channel-substitution rule that decides the chip: Databricks is a proprietary single-implementation platform whose pipeline and workspace opinions are Databricks-captive through any cloud. Multi-cloud availability is a choice of invoice, not an exit. Ceded-to-Databricks, matching the Databricks row. The Spark API surface in front of it is its own chip.

Spark API on Azure DatabricksDelegated

The Spark API has independent implementations: code and definitions written to it move to another compatible implementation within the features both support, the data moves by export, replication or rewrite, and the provider's control-plane configuration and operational state don't lift. Delegated. The service behind it is its own chip, Ceded.

NVIDIA-Provided

Assessment pending

◆ Gap Analysis

Azure's Layer 1C is comprehensive and mature. Data Factory's 5-year Gartner leadership position reflects enterprise-grade data movement capability. The Fabric integration collapses what were previously separate services (Data Factory, Synapse, Power BI) into a unified pipeline-to-analytics experience. Azure's Layer 1C advantage: Fabric's unified data platform means the pipeline, storage, analytics, and governance are the same system. Data moves through experiences within OneLake rather than between services. This architectural philosophy parallels vertically integrated approaches in other assessed platforms but at a different layer of the stack. The Databricks dependency is worth noting: many enterprise Azure customers use Databricks rather than native Fabric pipelines for complex data engineering. This creates a component Ceded to Databricks, with the Spark API as its Delegated door: the enterprise's data pipeline intelligence is Databricks' IP, running on Azure's infrastructure. No KV cache tiering story is evident on Azure. Dell has validated NVIDIA CMX (19x TTFT improvement), HPE has Alletra X10000 KV cache support, VAST collocates cache and compute in CNode-X. This gap matters as inference workloads scale and KV cache management becomes a data movement problem. Decision authority: Data Factory and Fabric pipelines carry the grade and run what the enterprise defined. Schema drift and auto-create table are opt-in settings, and Copy activity's name-matched column mapping is a fixed documented default an explicit mapping replaces, so neither reads as the engine deciding. Pipeline runs are monitored visually. Code decides, visible, overridable, Retained, the judgment test and its defaults clause (October 4 and 5, 2026).

◆ Borrowed Judgment

Ceded for native services. Delegated for Databricks. Copilot for Data Factory adds a dimension: natural-language pipeline authoring means the enterprise inherits Microsoft's LLM's understanding of data engineering patterns. Whether Copilot-authored pipelines match the quality of expert-authored ones is an open question with borrowed judgment implications.

◆ Working Notes

The Fabric positioning as 'operating system for enterprise data' (FabCon/SQLCon 2026) makes Layer 1C the layer where Microsoft's data platform ambitions are most visible. If Fabric succeeds as the unified data control plane, it collapses 1A (storage) + 1B (retrieval) + 1C (movement) into a single authority boundary, which simplifies DAPM classification but concentrates authority.

Layer 2A · OrchestrationInfrastructure OrchestrationAKS + Arc Hybrid Orchestrationdecides: vendor · Delegated▼

GPU scheduling, quotas, RBAC, fair-share scheduling, utilization optimization

Vendor-Provided

Azure Kubernetes Service (AKS)Delegated

Managed Kubernetes with GPU-aware scheduling. Dynamic Resource Allocation (DRA) GA at KubeCon Europe 2026: fine-grained GPU resource allocation enabling multiple AI workloads to share GPU resources, reducing idle GPU time from typical 30-40% rates. AKS-managed GPU node pools automate NVIDIA driver, device plugin, and DCGM metrics. MIG, MPS, and time-slicing for GPU sharing. Kueue for fair queuing and priority. Cilium mTLS encryption in public preview: sidecarless pod-to-pod security eliminating sidecar proxy overhead for AI workloads (built with Isovalent/Cisco). AI Runway (preview): unified inference API positioning Kubernetes as the AI infrastructure operating system, with cross-cloud GPU scheduling previewed. Microsoft contributed DRA to upstream Kubernetes: these GPU scheduling primitives are available to all Kubernetes users, not just AKS. AKS is a managed service behind the standard Kubernetes API: manifests lift to another conformant cluster, so Delegated (operation delegated to Azure; the standard interface keeps the opinions portable). Not scored within this chip under the GA evidence table: Cilium mTLS encryption ("Cilium mTLS encryption with Advanced Cluster Networking Services is currently in PREVIEW.").

Azure Arc (Hybrid Orchestration)Delegated

Extends Azure management to any infrastructure: on-prem, edge, multi-cloud. Projects external resources into Azure Resource Manager. Unified RBAC, policy, monitoring. AKS enabled by Azure Arc: AI inference on hybrid Kubernetes clusters with centralized governance. Connected and disconnected operation modes. The only hyperscaler management plane designed to orchestrate across non-native infrastructure: AWS manages AWS; Google manages GCP and GDC; Azure Arc manages anything.

NVIDIA-Provided

NVIDIA GPU Operator + DRA

NVIDIA GPU Operator manages GPU drivers and device plugins on AKS. DRA GA provides Kubernetes-native GPU scheduling primitives. Microsoft contributed DRA to upstream Kubernetes.

◆ Gap Analysis

Azure's Layer 2A is mature and comprehensive. AKS is the most widely deployed managed Kubernetes service for AI workloads on Azure, and the KubeCon 2026 DRA GA + AI Runway announcements extend its capabilities specifically for AI scheduling. Azure Arc is the hybrid orchestration differentiator: it extends Azure's management plane to any infrastructure, including competitor hardware. An enterprise running Azure Arc on Dell, HPE, and Lenovo hardware gets a unified orchestration plane across OEM boundaries, managed by Microsoft rather than Broadcom (VMware) or the OEM itself. The AI Runway + cross-cloud GPU scheduling preview is the most aggressive multi-cloud orchestration claim in this assessment. If Azure can schedule workloads across its own GPUs, AWS GPUs, and GCP GPUs based on availability and pricing, it becomes the first cross-hyperscaler Layer 2A orchestration plane. This is aspirational (the preview was just announced) but architecturally significant. DRA contribution to upstream Kubernetes means these GPU scheduling primitives are available to all Kubernetes users, not just AKS. Microsoft is building the open-source foundation that competitors also benefit from, a strategic choice that prioritizes ecosystem leadership over proprietary advantage. Brendan Burns' authorship (Kubernetes co-creator, Microsoft employee) gives Azure unique credibility in shaping Kubernetes' evolution. Capability and deployment are separate readings. The grade states what an architect can deploy on this vendor's paper; it does not assert that a given customer consumes it. Azure Red Hat OpenShift is the concrete case, a first-party jointly engineered service in which the customer consumes Azure for Layer 0 and capacity while Red Hat owns the Kubernetes control plane, leaving AKS unconsumed. Arc's hybrid reach does not change that reading: it extends where Azure's orchestration can govern, not whether a given customer has chosen it.

◆ Borrowed Judgment

Ceded for cloud workloads. Delegated for Azure Arc-managed on-prem infrastructure. Microsoft's Kubernetes contributions (DRA, AI Runway) are open-source: the enterprise can run them on any Kubernetes. But operational maturity (managed upgrades, monitoring, scaling) is Azure-specific. The Kubernetes interface keeps the enterprise's opinions portable to any K8s (Delegated: managed K8s service; the enterprise could switch without rebuilding); Azure-specific operational maturity is convenience, not capture.

◆ Working Notes

Microsoft's KubeCon 2026 framing of 'Kubernetes as the AI Infrastructure OS' is a Layer 2A statement, not a Layer 2C statement. The distinction matters: Kubernetes schedules and manages resources. A Reasoning Plane governs them with policy-driven intelligence. Same distinction applies to VMware's framing of VCF as 'the permanent abstraction layer.' Out of the Azure Kubernetes Service chip under the GA evidence table (October 24, 2026): Cilium mTLS encryption ("Cilium mTLS encryption with Advanced Cluster Networking Services is currently in PREVIEW.").

Layer 2B · RuntimeApplication Runtime & ExecutionFoundry + OpenAI Runtimedecides: model · Delegated▼

Model serving, agent execution, inference APIs, distributed inference

Vendor-Provided

Model Serving via the OpenAI-Compatible Interface (Azure OpenAI in Foundry)Delegated

Frontier OpenAI models (GPT-5.5, o-series: four-month first-mover window per the April 2026 restructuring) plus catalog models served through the OpenAI API shape: the same client libraries with a base-URL swap and deployment names. A managed service behind the multi-vendor standard interface: integration opinions lift to any compatible endpoint, so the serving interface is Delegated. The remaining switching cost is behavioral recalibration, graduated by how deeply prompts are tuned to one model. OpenAI models are now also on AWS Bedrock: the model is no longer an Azure differentiator; the platform integration is.

Customer-Created and Managed Tools (Function Calling / Client-Side)Retained

GA through Azure OpenAI in Foundry (function calling on the OpenAI-compatible interface). The enterprise owns and operates the business logic, APIs, commands and deterministic validators invoked through tool calls: the seam where control passes from instructing the model to executing code outside it. The opinions are the enterprise's and run against a multi-vendor interface, so they lift to another provider with an adapter and re-evaluation, not a rebuild. What the enterprise cedes at this seam is the decision to act: the model chooses whether, when and with which arguments to call, and that judgment is inherited under the serving component above. Added instrument-wide on September 2, 2026 (M4): the seam exists on every runtime that hands control to portable customer code, and a real Retained position is scored, not narrated.

Microsoft Foundry Platform Surface (Agent Service + Control Plane)Ceded

The platform beyond the standard interface: Foundry Agent Service (fully managed agent runtime: no-code prompt agents and code-based agents, hosting, scaling, identity, observability), OpenResponses/Activity/Invocations and A2A protocols for agent distribution through M365 Copilot, Teams, and the Entra Agent Registry, Foundry Control Plane centralizing observability for agents across frameworks (including agents not running on the platform), batch evaluations, and the 11,000+ model catalog integration. Opinions built against this surface run nowhere else: the beyond-the-interface capture, Ceded. Not scored within this chip under the GA evidence table: Entra Agent Registry, Foundry Control Plane ("select Entra ID > Agents > Agent registry (Preview)").

Microsoft Agent Framework (Open-Source)Retained

Production framework for building multi-agent systems. Orchestration patterns: sequential, concurrent, handoff, group chat (Magentic One). OpenAPI integration, A2A protocol, MCP support. Local development → Azure deployment with observability and compliance. Open-source (MIT license). KPMG Clara AI built on Agent Framework.

NVIDIA-Provided

NVIDIA Nemotron + NIM on Foundry

NVIDIA models available through Foundry model catalog. Vera Rubin support via Azure Local + Foundry Local for on-prem inference.

NVIDIA NeMo Data Designer

Integration through Foundry for model training and fine-tuning. Same NVIDIA training dependency seen across multiple assessed vendors.

◆ Gap Analysis

Azure's Layer 2B is one of the broadest: 11,000+ models, managed agent runtime, open-source agent framework, cross-framework observability. The breadth creates complexity: the enterprise must choose between Foundry Agent Service (Ceded, managed), Agent Framework on AKS (Retained, self-hosted), Azure OpenAI direct (Ceded, model-specific), and fully self-hosted options. The April 2026 Custom Agent Monitoring is architecturally significant: Foundry extends governance to agents it doesn't host. This is a Layer 2B/2C crossover: the runtime observability reaches beyond the runtime boundary. The OpenAI restructuring creates a unique Layer 2B dynamic: OpenAI models are Azure's flagship capability AND are now available on AWS Bedrock. The model is commodity; the platform around the model is the lock-in. The Agent Framework being open-source (MIT) means the enterprise Retains the code and can run it anywhere. But the operational envelope (Foundry hosting, Entra identity, Purview compliance) is Azure-specific. Code portability vs. operational portability: same distinction identified in the Google Cloud assessment with ADK.

◆ Borrowed Judgment

Multi-layered. OpenAI models: borrowed alignment, training data, and safety decisions, the most significant model-provider borrowed judgment in this assessment because OpenAI is simultaneously a partner, a competitor (ChatGPT Enterprise vs. Microsoft 365 Copilot), and a platform (OpenAI API vs. Azure OpenAI Service). Microsoft's own models (MAI-1, CoreAI): borrowed judgment shifts to Microsoft's model team. Open-source models: community-borrowed judgment. The enterprise using Azure OpenAI Service inherits three judgment systems simultaneously: Microsoft's platform decisions (Foundry defaults, content filtering), OpenAI's model decisions (alignment, capabilities, safety), and NVIDIA's silicon decisions (GPU scheduling, driver behavior). This is the most complex borrowed judgment stack in the assessment.

◆ Working Notes

The April 2026 restructuring eliminated Azure exclusivity for OpenAI models. GPT-5.5 appeared on AWS Bedrock the next day. This validates the 4+1 model's prediction that model-layer lock-in is transient while platform-layer lock-in (2A/2B/2C) is durable. Microsoft's response is right: invest in Foundry, Entra Agent ID, and governance infrastructure that doesn't move with the model. The CoreAI division under Mustafa Suleyman signals Microsoft is building its own model capability to reduce OpenAI dependency. The borrowed judgment composition changes as Microsoft's own models mature. Today it's Microsoft platform + OpenAI models. Tomorrow it could be Microsoft platform + Microsoft models, a vertically integrated model similar to Google's. Out of the Microsoft Foundry Platform Surface chip under the GA evidence table (October 24, 2026): Entra Agent Registry, Foundry Control Plane ("select Entra ID > Agents > Agent registry (Preview)").

Layer 2C · ReasoningAgentic Infrastructure — The Reasoning PlaneComplete Intelligence-2C Plane Across Products; Gateway Architect-Wireddecides: model · Delegated▼

Policy-driven placement and resource coordination — the Autonomy Layer

Vendor-Provided

Microsoft Entra Agent ID (GA May 1, 2026)Ceded

Agents as first-class Entra identities: authentication, authorization, lifecycle, and protection. Conditional Access templates for agents, ID Protection risk detection, access packages and sponsor lifecycle workflows. Copilot Studio creates an Agent ID for every new agent (since May 2026, no opt-out) and attaches its connectors as API permissions Conditional Access can target. Federation patterns bring Amazon Bedrock and n8n agents under Agent ID.

Microsoft Agent 365 (GA May 1, 2026)Ceded

Tenant-wide agent registry and control plane. Copilot Studio agents register automatically; administrators block, reassign, or delete agents, with actions logged in Purview. Copilot Studio agents emit OpenTelemetry invocation, tool, and output spans to Agent 365 automatically. AWS and Google Cloud registry sync is preview.

Copilot Studio Multi-Agent Orchestration (GA)Ceded

A main agent delegates to child agents, connected Copilot Studio agents, and any Agent2Agent (A2A) agent (A2A GA April 2026), including agents on other frameworks, clouds, or on-premises. By default a model routes by each agent's description; topic redirects, child-agent conditions and priorities, and agent flows that call agents are deterministic overrides. The orchestration definitions are Copilot Studio objects.

Azure API Management AI Gateway Capabilities (GA)Ceded

Capabilities of the existing API Management gateway in every tier, not the preview AI Gateway tier: governs model APIs (OpenAI, Anthropic, Vertex schemas), MCP servers, and A2A agent APIs with authentication, token and rate limits, content safety, and logging. In an agent's path only where the architect routes traffic through it; Foundry's built-in gateway integration is preview. Policies are API Management XML, a single-vendor surface.

NVIDIA-Provided

No NVIDIA Layer 2C Dependency

All Layer 2C components are Microsoft IP or open-source. NVIDIA does not provide or control governance, policy, agent identity, or reasoning in the Azure stack.

◆ Gap Analysis

Microsoft ships a complete Intelligence-2C plane, generally available, assembled from five products rather than one. Read against the five legs, with the joins between them: Identity (GA May 1, 2026): Entra Agent ID makes agents first-class identities with Conditional Access, ID Protection, and lifecycle governance. Since May 2026 Copilot Studio creates an Agent ID for every new agent with no opt-out, and attaches each connector the agent uses as an API permission that Conditional Access can target. Agents built before May 2026 still run on app registrations until Microsoft migrates them. Registry (GA May 1, 2026): Agent 365 is the tenant-wide inventory. Copilot Studio agents register automatically on creation, update, and deletion, and administrators can block, reassign, or delete them from the registry. Registry sync for agents on AWS Bedrock and Google Cloud is public preview. Orchestration (GA): Copilot Studio's multi-agent orchestration lets a main agent delegate to child agents, to other Copilot Studio agents, and to any agent that speaks the open Agent2Agent (A2A) protocol, including agents on other frameworks, other clouds, or on-premises. By default a model routes by each agent's description; explicit topic redirects, conditions and priorities on child agents, and agent flows that call agents are overrides the enterprise can set. Connections to Foundry agents, Fabric data agents, and Agents SDK agents are preview. Gateway (GA): Azure API Management's AI gateway capabilities apply to every API Management tier and govern model APIs, MCP servers, and A2A agent APIs with authentication, rate and token limits, and logging. These are capabilities of the existing API Management gateway, not the preview AI Gateway tier, and Foundry's built-in gateway integration, which would route Foundry traffic for the architect, is also preview. This is the weakest join: Copilot Studio can call an API Management-hosted MCP server, and an A2A connection accepts any endpoint URL, so the gateway sits in an agent's path only where the architect routes traffic through it. Copilot Studio's connector calls don't pass through API Management; Power Platform connector and data policies govern them, revalidated by the connector runtime at call time. Observability (GA): Copilot Studio agents emit OpenTelemetry spans for agent invocations, tool executions, and outputs to Agent 365 automatically, visible in the Microsoft 365 admin center, Defender, and Purview, for authenticated sessions. Foundry evaluations are GA, and Foundry tracing is GA for prompt and hosted agents (preview for workflow and external agents). That is the complete plane the 2C criterion asks for, which reads strong, calibrated to Google Cloud and Databricks. The difference is construction. Google's plane is one product, Agent Platform, which wires its legs together. Microsoft sells the legs as Entra, Agent 365, Copilot Studio, API Management, and Foundry. Identity, registry, and observability join automatically for Copilot Studio agents; the gateway path, and the joins for agents built elsewhere, are the architect's to build. The enterprise owns those seams. A plane the architect wires from GA parts on the vendor's paper counts as complete (ruled October 5, 2026), the same reading the instrument gives the enterprise's own agent loop at 2B. Infrastructure Layer 2C: no service answers 'given data residency, cost, latency, and compliance, should this workload run on NVIDIA or AMD in which region?' Arc and AKS scheduling are building blocks, and there's no documented GA product for live policy-driven inference placement. Like live placement everywhere on the instrument, this is a universal finding, not a charge against Microsoft.

◆ Borrowed Judgment

The orchestrator is a model. Copilot Studio's generative orchestration picks which agent handles a request from each agent's description. The enterprise can see that choice (activity maps, Agent 365 telemetry) and override it: conditions and priorities on child agents, explicit redirects from topics, agent flows that call agents deterministically, and registry blocks. Conditional Access, connector and data policies, and API Management policies are gates the enterprise writes, not the runtime decider; the risk signals they can condition on (ID Protection, Content Safety) are Microsoft's. The deciding half is the model's orchestration, so the layer reads model-decides, visible, overridable: Delegated, beside Snowflake, Databricks, and Nutanix. The Entra Agent ID dependency is worth flagging: extending agent identity into Entra ties agent governance to Microsoft's identity plane. An enterprise running agents on AWS that are governed by Entra Agent ID has a cross-cloud identity dependency on Microsoft, the way Active Directory became the enterprise identity standard. Infrastructure 2C: not yet built. Arc and AKS scheduling are building blocks that have not been composed into a policy-driven placement engine.

◆ Working Notes

Not scored (preview, no GA date): the Agent Governance Toolkit (MIT-licensed; its repository reads 'Public Preview -- production-quality public preview releases. May have breaking changes before GA'), Foundry's AI Gateway integration and its MCP tool governance, Foundry workflows (preview, retiring December 1, 2026, with new development directed to Microsoft Agent Framework), Copilot Studio connections to Foundry, Fabric, and Agents SDK agents, Agent 365 registry sync with AWS and Google Cloud, and Foundry monitoring. None carries a GA date, so none is a watch-list item. Microsoft Agent Framework 1.0 (stable April 3, 2026, with sequential, concurrent, handoff, group chat, and Magentic-One orchestration) is open-source code the enterprise runs and is scored at 2B; the orchestration leg here rests on Copilot Studio, a deployable product. Surfaces taken: the Copilot Studio 'Add other agents' overview carries a page-level preview banner because it covers the preview Foundry, Fabric, and SDK connections; the child-agent, connected-agent, and A2A pages carry none, and Copilot Studio's 'What's new' lists multi-agent orchestration (November 2025) and A2A (April 2026) as released, so the specific pages govern. Foundry's GA overview lists 'Manage | AI Gateway | Preview', which governs over the Foundry gateway configuration page's missing label. Copilot Studio is scored at Layer 3 for building agents and credited here only for its orchestration function, the function an architect shops for at 2C. Microsoft's Cloud Adoption Framework for agent governance (April 2026) lays out the composition: data governance and compliance (Purview), agent observability (Agent 365, Defender, Log Analytics), agent security (Defender, Content Safety, RBAC, Sentinel), and agent development (Agent Framework, Foundry SDK, MCP, A2A). The sibling finding applies here as well: you cannot prompt your way to deterministic output. The controls credited in this cell answer whether an action is legal (identity, Conditional Access, connector and data policies, quotas, rate limits), and none of them validates whether the outcome was right under a graduated escalation policy. That deterministic outcome-validator is missing instrument-wide and is noted as universal.

Layer 3 (+1) · ApplicationsAI Application Layer — The Value PlaneBroadest Enterprise Ecosystemdecides: model · Delegated▼

AI-powered business capabilities — business logic, workflow automation

Vendor-Provided

ISV + Partner EcosystemDelegated

Azure Marketplace with thousands of AI applications. SI partnerships (Accenture, Deloitte, KPMG, Infosys). ISV integrations across every industry. The Microsoft partner network is the largest in enterprise technology.

GitHub Copilot (Agentic Developer Platform)Ceded

GitHub Copilot has evolved from code completion to a full agentic development platform with three surfaces: IDE agent mode (VS Code, Visual Studio 2026 with cloud agent integration), Copilot CLI (GA for all paid subscribers: Plan mode, Autopilot mode, dynamic agent delegation), and cloud agent (autonomous coding agent that researches repos, creates plans, makes code changes, opens PRs without developer in the loop via GitHub Actions). Multi-model: Claude, GPT, and OpenAI Codex models selectable per task. GitHub Copilot SDK enables building custom agents using Copilot's orchestration runtime: planning, tool invocation, streaming, MCP server integration. Foundry Local integration enables fully local, air-gapped agentic coding with data sovereignty. Microsoft Agent Framework supports Copilot SDK as agent backend. Visual Studio 2026 adds Debugger Agent that validates fixes against real runtime behavior. The most pervasive developer AI surface by installed base, integrated into the largest code hosting platform (GitHub) and the most widely used enterprise IDE ecosystem (VS Code + Visual Studio). Not scored within this chip under the GA evidence table: the cloud coding agent and the Debugger Agent (previously in public preview, no GA affirmation found).

NVIDIA-Provided

NVIDIA NIM + Blueprints on Foundry

NVIDIA models and application patterns available through Foundry. Same blueprints available across other assessed vendors, non-differentiating.

◆ Gap Analysis

Azure's Layer 3 is structurally unique because Microsoft owns both the infrastructure platform AND the largest enterprise application estate in market. GitHub Copilot (developer AI), Microsoft 365 Copilot (knowledge worker AI, 3.3% paid adoption as of early 2026), Power Platform AI (business process automation), and Dynamics 365 Copilot (CRM/ERP AI) are Microsoft application products, not Azure services, but they consume Azure AI infrastructure at Layers 0–2C. This creates a dynamic no other vendor has: the enterprise's Layer 3 application decision often drives the Layer 0 infrastructure decision rather than vice versa. On-prem vendors sell infrastructure first; Azure sells applications first. This application estate context does not appear as assessed components because these are Microsoft products, not Azure services. The parallel exists in the Google Cloud assessment where Gemini's consumer properties are acknowledged as context without being assessed as GCP Layer 3 components. The GitHub Copilot vs. Microsoft 365 Copilot adoption contrast has implications beyond Azure. GitHub Copilot succeeds (high adoption, measurable productivity impact in a specific workflow). M365 Copilot struggles (3.3% conversion on the largest enterprise installed base). This suggests Layer 3 AI applications succeed when targeting specific professional workflows rather than augmenting general knowledge work, an observation relevant to every vendor's Layer 3 strategy. Copilot Studio is the Azure-native Layer 3 capability: low-code/no-code agent building with distribution through M365 and Teams. This is the bridge between the Microsoft application estate and the Azure AI platform: agents built in Copilot Studio consume Foundry models, are governed by Entra Agent ID, and are distributed through the M365 surface. GitHub Copilot's evolution from code completion to autonomous cloud agent represents the most significant Layer 3 shift in the Azure/Microsoft ecosystem. The cloud agent (coding autonomously in GitHub Actions, opening PRs without developer presence) creates a new category of AI-generated code flowing through enterprise repositories. The DAPM question: when Copilot's cloud agent writes production code autonomously, who owns the engineering judgment? The developer who assigned the issue, the model that generated the code, or GitHub's agent orchestration logic? The GitHub Copilot SDK is strategically important: it exposes Copilot's production agent runtime as a programmable API, enabling enterprises to build custom agents on top of GitHub's orchestration engine. Combined with Foundry Local for air-gapped on-device inference, this creates a Retained-authority path for enterprises that need agentic development without cloud dependency, the only assessed cloud vendor offering a fully local agentic developer tool. Three-cloud comparison of agentic developer surfaces: • AWS Kiro: Spec-driven, methodology-opinionated. Enforces structured requirements → design → implementation. Replaces Q Developer. Deep AWS integration (pricing, Well-Architected, Bedrock). Most prescriptive. • Google Antigravity 2.0: Agent orchestration platform. Multi-agent parallel execution, scheduled background tasks. Desktop + CLI + SDK. Replaces Gemini CLI. Gemini-native. Most ambitious multi-agent vision. • GitHub Copilot: IDE-embedded + CLI + cloud agent. Multi-model (Claude, GPT, Codex). GitHub-native (repos, issues, PRs, Actions). Copilot SDK for custom agents. Foundry Local for air-gapped. Most pervasive installed base. All three are Ceded or Delegated: the enterprise adopts the vendor's opinions about how AI-assisted development should work. The competitive differentiation is in the development philosophy: AWS imposes discipline (specs first), Google enables parallelism (multiple agents), Microsoft/GitHub enables delegation (assign and forget).

◆ Borrowed Judgment

The most complex borrowed judgment landscape in this assessment. The enterprise using Azure AI inherits judgment from: Microsoft's platform decisions (Foundry defaults, content filtering), OpenAI's model decisions (alignment, capabilities, safety), NVIDIA's silicon decisions (GPU scheduling, driver behavior), ISV application decisions, and Microsoft's enterprise application decisions (Copilot integration points, Power Platform automation patterns). Model choice is a relationship, not a Layer 3 grade: the Foundry catalog lets the enterprise procure OpenAI, Meta, Mistral and Cohere models on Microsoft's paper, and it reads at 2B with the Foundry platform. The unique risk: Microsoft's Layer 3 applications are also the enterprise's productivity tools. If Copilot in Word makes a poor suggestion, it affects the document. If Copilot in Dynamics 365 makes a poor recommendation, it affects the sales pipeline. The blast radius of borrowed judgment at Layer 3 is larger for Microsoft than for other vendors because the applications are mission-critical business tools, not standalone AI applications. Decision authority (October 5, 2026): Microsoft 365 Copilot carries the grade. Operations that change data prompt the user, show the data to be sent, and offer allow or decline (plugin confirmation prompts docs). Model decides, visible, overridable, Delegated.

◆ Working Notes

The Microsoft 365 Copilot adoption data is the enterprise AI reality check this assessment series benefits from. At 3.3% conversion on the largest installed base in enterprise software, the question is whether the industry's Layer 3 ambitions are ahead of enterprise readiness, and whether that readiness gap affects infrastructure investment decisions at Layers 0–2. GitHub Copilot's success vs. Microsoft 365 Copilot's adoption challenge has implications for every vendor's Layer 3 strategy: AI applications may succeed faster when they target specific professional workflows (coding, design, data engineering) than when they target general knowledge work. The GitHub Copilot SDK's availability as a programmable agent runtime backend via Microsoft Agent Framework creates a developer platform play that spans Layers 2B and 3. Enterprises building custom agents on the Copilot SDK inherit GitHub's orchestration opinions (planning, tool invocation, context management) as borrowed judgment. The SDK is the distribution mechanism for Microsoft's agent architecture opinions into enterprise codebases. Foundry Local + Copilot SDK for air-gapped agentic development is a unique capability in this assessment. No other cloud vendor provides a fully local, data-sovereign agentic developer tool. AWS Kiro requires Bedrock connectivity. Google Antigravity requires Gemini API. GitHub Copilot with Foundry Local runs entirely on-device. This matters for defense, financial services, and government enterprises where source code cannot leave the local environment. The agentic developer tools market is consolidating around three philosophies: structured discipline (Kiro), parallel orchestration (Antigravity), and delegation-first autonomy (Copilot). Red Hat's OpenShift Dev Spaces supporting Kiro, Copilot, Claude CLI, and others from a single governed runtime suggests the enterprise will run multiple agentic developer tools simultaneously, governed by the platform layer (2A) rather than choosing a single tool at Layer 3. Named, not scored at Layer 3 (October 19, 2026: a builder whose only output is software the enterprise runs earns no Layer 3 grade): Copilot Studio (a low-code agent builder; its orchestration is scored at 2C). Named, not scored at Layer 3 (October 22, 2026): the Microsoft Foundry Model Catalog. Model catalogs earn no Layer 3 chip; the catalog reads at 2B, inside the Foundry platform chip. Out of the GitHub Copilot chip under the GA evidence table (October 24, 2026): the cloud coding agent and the Debugger Agent (previously in public preview, no GA affirmation found).

Sources & revision history · v1.41 - 4+1 v3: Layer 0 Sub-Layers

Build 2025, GTC 2026, KubeCon Europe 2026, FabCon/SQLCon 2026, Ignite 2024, Microsoft/OpenAI restructured agreement (April 2026), Entra Agent ID GA (April 2026), Agent Governance Toolkit (April 2026), Foundry Agent Service, analyst coverage. v1.2 (instrument reconciliation): 2A AKS Retained→Delegated: managed K8s behind a standard interface is Delegated, not Retained. Azure Blob remains Ceded (single-vendor API). v1.3 (July 12, 2026, /reconcile): statusLabels moved to capability vocabulary (authority lives in the DAPM chips); Azure Databricks Delegated→Ceded under the channel-substitution rule (proprietary platform Ceded-to-owner through any paper, matching the Databricks row). Inference-interface reconciliation (July 23, 2026): the completions-interface ruling (OpenAI v1.0) and its Messages-API extension (Anthropic v1.0) applied: model access consumed through a genuine multi-vendor standard inference interface is Delegated; proprietary surfaces beyond the interface remain Ceded. 2B Foundry+Azure OpenAI split: serving via the OpenAI-compatible interface Delegated; the Foundry platform surface Ceded. /reconcile (September 1, 2026): shared 2C findings stated explicitly (outcome-validation finding). /reconcile (September 1, 2026): the availability-not-occupancy reading applied at 2A: the grade states deployable capability on this vendor's paper and no longer reads as though the vendor's layer is automatically the customer's. /reconcile (October 5, 2026): Layer 0 runtime rule (ruled September 28 and October 4, 2026). GPU accelerators split: ND-series NVIDIA and ND MI300X v5 instances Ceded to Retained from the customer's seat, Vera Rubin NVL72 Ceded on the integrated-system rule. Grade unchanged. /reconcile (October 5, 2026, same pass): Maia removed from scored components (silicon behind a service, vendor-shape ruling); Cobalt split out and Ceded to Retained on the Arm CPU ruling. /reconcile (October 5, 2026): 1C authority column (rulings of September 15 and October 4 and 5, 2026). 1C authority vendor / Ceded to code / Retained: schema drift and auto-create table are opt-in settings; Copy activity's name-matched mapping is a fixed documented default an explicit mapping overrides, configuration under the defaults clause; pipeline monitoring is documented. Grade unchanged. /reconcile (October 5, 2026): authority visible set from the cell's own text where it settles it: layer1a true. Direction, grades, and components unchanged. /reconcile (October 5, 2026): authority re-read against product docs under the judgment test, override rule, and Layer 3 ruling: layer1a to vendor / Delegated; layer3 to model / Delegated. Grades and components unchanged. /reconcile (October 5, 2026): authority read against product docs: layer1b visible true; layer1b Ceded to Delegated on per-request retrieval overrides. Grades and components unchanged. Layer 2C re-read (October 5, 2026): public-source check found the Agent Governance Toolkit in public preview (its repository) and the Foundry AI Gateway in preview (Foundry GA overview), so neither can carry a leg. Strong holds on the complete-plane criterion on GA products: Entra Agent ID (GA May 1, 2026), Agent 365 (GA May 1, 2026), Copilot Studio multi-agent orchestration with A2A (GA), API Management AI gateway capabilities (GA, architect-routed), and Agent 365 observability for Copilot Studio agents (GA). Authority code / Retained to model / visible / overridable / Delegated: the orchestrator is a model, with deterministic overrides. Agent Governance Toolkit moved to notes. Ruled with it: a complete plane the architect wires from GA parts on the vendor's paper counts (building blocks count). ChatGPT peer check, two passes (reviews/azure-2c-chatgpt.json, reviews/azure-2c-chatgpt-v2.json), all findings upheld or folded. v1.16 (Layer 0 sub-layers, October 16, 2026): Layer 0 split into 0A Compute, 0B Network Fabric and 0C Storage Substrate by runtime decision (Keith, October 16, 2026); storage left 1A for 0C, and 1A, now Data Foundation & Governance, is graded on the catalog that classifies the data. 0A strong, 0B strong, 0C strong. v1.17 (October 19, 2026, /reconcile) Layer 3: Copilot Studio (a low-code agent builder; its orchestration is scored at 2C) moved out of the chip list under the Layer 3 exclusion for software-only builders (Keith, October 19, 2026). Grade and authority unchanged. (Same pass) Plain-language pass: methodology rule numbers in the prose and audit notes replaced with the reasoning they stood for; no grade, chip or authority reading changed. v1.18 (October 19, 2026, /reconcile voice pass): em-dashes in the prose replaced with periods, colons, commas or parentheses; wording, grades, chips and authority readings unchanged. v1.19 (October 21, 2026, /reconcile): Azure NetApp Files Delegated to Ceded. The earlier reading rested on its NFS and SMB surface; ONTAP on a cloud's paper is Ceded to NetApp on portability, with the cloud deciding at runtime (Keith, October 21, 2026), matching NetApp's own ONTAP chip and Lenovo's ONTAP arrays. Grade and cell authority unchanged; the cell rests on Managed Lustre. v1.20 (October 22, 2026, /reconcile): Vera Rubin NVL72 on Azure (no Rubin size on Microsoft Learn's sizes overview, updated September 25, 2026) moves to named, unscored notes at 0B and 0A. In its place goes ND GB300 v6 on GB300 NVL72 (size page updated October 7, 2026: 18 VMs and 72 GPUs to a rack, 130 TB/s of intrarack NVLink, 800 Gbps InfiniBand per GPU). The 0B calibration line now cites GCP's RoCE GPU network in place of Virgo, after GCP v1.26. Found by the 0B fabric research. Grades and authority unchanged. v1.21 (October 22, 2026, /reconcile): 0B chip 0 split into the ND-series InfiniBand fabric (Microsoft Learn: full fat-tree, non-blocking, RDMA only over InfiniBand; 800 Gbps per GPU on ND GB300 v6, to up to 100,000 GPUs) and Accelerated Networking (SR-IOV front-end NICs on ConnectX or MANA), both Ceded. SONiC, which Microsoft gives away rather than sells, moves to a note. Found by the 0B fabric research. Grade and authority unchanged; the cell rests on the InfiniBand fabric. v1.22 (October 22, 2026, /reconcile): the 0A NVL-class chip cites Lenovo's NVL-class racks as its precedent in place of Dell's XE9812, which Dell v2.22 moved to a named, unscored note. Grade and authority unchanged. v1.23 (October 22, 2026, /reconcile): the Foundry Model Catalog comes off Layer 3 under the model-catalog ruling (catalogs belong to 2B, where the Foundry platform chip already carries it), and the relationship point, procuring third-party models on Microsoft's paper, moves to the Layer 3 DAPM prose. Grade and authority unchanged. v1.24 (October 22, 2026, /reconcile): Blackwell moves from the standard ND chip, where Microsoft Learn lists no B200 or B300 series, to the NVL-class chip, which adds ND GB200 v6; the undocumented "1M+ NVIDIA GPUs" comes out; MI300X v5 is described as Microsoft's docs describe it, a training-class VM; the NVIDIA column drops Vera Rubin, which the notes already hold as unscored. Azure Boost replaces the validated OEM server chip, which Microsoft doesn't sell. Found by the 0A compute research. Grade and authority unchanged. v1.25 (October 23, 2026, /reconcile): 0A: Separate VM series, each with its own Microsoft Learn page. Found by the Layer 0 data pass after the sniff read. Grade and cell authority unchanged. v1.26 (October 23, 2026, storage door pass): NFS 3.0 on Blob, the Lustre client and Azure NetApp Files' protocols are doors; the managed services read Ceded. Grade unchanged. v1.27 (October 23, 2026, Layer 1 calibration): Azure Databricks drops "(Partnership)": Microsoft documents and bills it as an Azure product. Grade unchanged. v1.28 (October 23, 2026, Layer 1 managed-services pass): Microsoft Fabric / OneLake gets a door; Azure Databricks gets a door. Grade unchanged. v1.29 (October 24, 2026, /reconcile): GA evidence recorded on 1 of 2 Layer 1B chips under the GA evidence table. Grade unchanged. v1.30 (October 24, 2026, /reconcile): Layer 1B under the GA evidence table version 2: GA evidence on every chip, preview items and unaffirmed chips to notes. Grade unchanged. v1.31 (October 24, 2026, /reconcile): GA evidence recorded on 3 of 3 Layer 1A chips under the GA evidence table version 2. Grade unchanged. v1.32 (October 24, 2026, /reconcile): GA evidence recorded on 3 of 3 Layer 1C chips under the GA evidence table version 2. Grade unchanged. v1.33 (October 24, 2026, /reconcile): GA evidence recorded on 4 of 4 Layer 0B chips under the GA evidence table version 2. Grade unchanged. v1.34 (October 24, 2026, /reconcile): GA evidence recorded on 6 of 6 Layer 0A chips under the GA evidence table version 2. Grade unchanged. v1.35 (October 24, 2026, /reconcile): GA evidence recorded on 8 of 8 Layer 0C chips under the GA evidence table version 2. Grade unchanged. v1.36 (October 24, 2026, /reconcile): GA evidence recorded on 1 of 2 Layer 2A chips under the GA evidence table version 2. Grade unchanged. v1.37 (October 24, 2026, /reconcile): GA evidence recorded on 3 of 4 Layer 2B chips under the GA evidence table version 2. Grade unchanged. v1.38 (October 24, 2026, /reconcile): GA evidence recorded on 1 of 2 Layer 3 chips under the GA evidence table version 2. Grade unchanged. v1.39 (October 24, 2026, /reconcile): GA evidence recorded on 4 of 4 Layer 2C chips under the GA evidence table version 2. Grade unchanged. v1.40 (October 24, 2026, /reconcile): GA evidence table version 2.1 applied: 3 chips recorded or narrowed. Grade unchanged. v1.41 (October 27, 2026): each component now carries a durable offering id; nothing else changed.

4+1 Layer AI Infrastructure Model · Vendor Assessment Series · The CTO Advisor LLC (DBA The Advisor Bench) · thectoadvisor.com