Supply Chain 2.0: How Microsoft is powering simulations, AI agents, and physical AI

The next wave of AI innovations

Exactly one year ago, we outlined how generative AI is creating a new era of efficiency and innovation for logistics and supply chain. We mapped AI use cases across the value chain, from demand forecasting to AI-based customer service, and introduced two new reference architectures for logistics and supply chains: adaptive cloud and AI‑enhanced experiences, alongside innovations in Microsoft Dynamics 365.

Since then, technology has rapidly evolved. We are now in the agentic era of AI with agents being capable of reasoning, planning, and taking action across complex supply chain workflows. End-to-end agent hosting like in Microsoft Foundry and open protocols such as Model Context Protocol (MCP) have made it easier for AI agents to connect with each other as well as enterprise systems, tools, and data.

Additionally, there have been significant advances in 3D simulations, robotics, and embodied intelligence. Models such as NVIDIA Cosmos and the OSMO edge-to-cloud compute framework on Azure enable machines and humanoid robots to act more effectively in the physical world, resulting in broader automation across warehouses, distribution centers, and transportation. This new article picks up Microsoft’s perspective on supply chain and logistics one year after our previous blog article and explores how our own logistics teams as well as frontier customers and partners use this new wave of innovations together with Microsoft.

Microsoft supply chains: Our own “customer zero” story

Microsoft operates one of the world’s most far-reaching cloud supply chains spanning more than 70 Azure regions, over 400 datacenters, and a network of more than 600,000 km of fiber. Our datacenters are the backbone of Microsoft Azure powering everything from AI infrastructure and collaboration tools to networking and security. Microsoft also runs supply chains for Microsoft Windows and Devices with Surface hardware and PC accessories as well as Xbox consoles and gaming hardware.

All of our supply chains have undergone a fundamental transformation over the past decade, evolving from a reactive, manual environment into a rapidly emerging autonomous, agentic supply chain. In the past, our operations were dominated by Excel-based reporting, limited visibility, and siloed data. In 2018, we consolidated more than 30 systems into a single supply chain supply chain data lake on Azure, enabling predictive analytics and the first generation of cognitive supply chain capabilities. In 2022, we began experimenting with generative AI, followed by the development of an AI platform to operationalize agents at scale. Today, this foundation is accelerating to fully autonomous agents, and more than 25 AI agents and applications have been deployed. Below are three examples:

• The Demand Planning Agent drives AI‑based demand–simulations for non‑IT rack components—improving forecast accuracy and reducing manual reconciliation.
• The Multi‑Agent DC Spare‑Part Space Solver uses computer‑vision‑driven monitoring and multi‑agent reasoning to forecast spare‑part storage needs and proactively mitigates space or stockout risks.
• The CargoPilot Agent continuously analyses transport modes, routes, cost structures, carbon impact, and cycle times—providing optimized shipment recommendations that balance speed, sustainability, and efficiency.

The goal is to operate over 100 agents by the end of 2026 and equip every employee with agentic support. The impact today is already huge: AI in logistics is saving our teams hundreds of hours each month demonstrating how agentic operations are translating directly into efficiency and business value. Both in our own Microsoft supply chain transformation and Frontier customers we work with, we have seen that unifying the data estate is key. Yet, it’s what organizations do next that truly generates value with AI.

In supply chain, we believe real value gets unlocked by driving three elements:

• Enabling AI-powered supply chain simulations.
• Building agentic supply chains.
• Integrating first physical AI innovations.

Simulations: The digital twins of supply chains

As supply chains become larger, more interconnected, and more exposed to global volatility, simulating scenarios before they unfold is becoming a critical capability to reduce risk and increase resilience. Discrete event-based simulations (DES) within supply chains enable the development of a virtual risk-free model to test how a complex system reacts to interventions and variables before implementation. With Microsoft’s advanced modelling tools such as Azure Machine Learning and the new machine learning model in Microsoft Fabric with Power BI semantic models, organizations in supply chain and logistics can simulate demand patterns, shortages, or supply chain disruptions.

Our partner paiqo offers with prognotix an AI-powered Forecasting Platform available on the Microsoft Marketplace. More than 70 algorithms enable supply chain experts to generate and optimize highly accurate demand forecasts directly within their Azure environment. Cosmo Tech offers an AI simulation platform for Advanced Supply Chain Risk Management on Azure, offering enterprise customers dynamic digital twins that simulate how disruptions and decisions impact system-wide performance. InstaDeep uses Azure in high-performance compute for AI-enabling deep reinforcement learning and predictive analytics that optimize last-mile delivery, inventory levels, and fleet utilization.

The next level of simulation combines multiple physical simulations in 3D environments and discrete event-based simulations to enable teams to build comprehensive digital twins of warehouses, distribution centers, production lines, and logistics networks. These virtual environments allow organizations to model both the physical behavior of assets and the dynamic flow of operations. By integrating these simulation methods within a digital twin and applying AI, teams can predict future outcomes, optimize performance, and prescribe actions that drive continuous operational improvements. This can help customers lower capital expenditure, shorten commissioning, and ramp up phases, as well as improve operational key performance indicators (KPIs).

Taking warehouses as an example, customers and partners can build advanced, AI-enabled 3D visualizations for four key scenarios:

• Warehouse planning (such as greenfield and brownfield).
• Warehouse monitoring (like real-time monitoring and people movement heatmaps).
• Warehouse improvement (for example trailer dwell time optimization and collision detection for safety and automation).
• Warehouse maintenance (like asset monitoring in real-time, detect quality issues, and reduce rework).

In collaboration with NVIDIA we offer access to NVIDIA libraries and frameworks including NVIDIA Omniverse™, NVIDIA Isaac Sim™, and NVIDIA Omniverse Kit App Streaming that enable developers to build applications and workflows to simulate and test things in digital twins before building or deploying anything in the real world. Applications built on these libraries and frameworks allow developers seamlessly integrate geometry data (such as 2D, 3D, and point clouds), AI capabilities (for example large language models, Volume Shadow Copy Service (VSS), and Solvers), and Internet of Things (IoT) signals across operational technology (OT) environments.

The reference architecture below illustrates how to combine cloud and edge computing using NVIDIA Omniverse Kit App Streaming to visualize warehouse operations in real-time with graphics processing unit (GPU) accelerated Kubernetes clusters natively deployed on Azure to remotely monitor, analyze, and optimize warehouse performance with greater precision and situational awareness.

Inside the physical warehouse, operational data from robotic arms, conveyors, and warehouse sensors are captured on the edge using Azure IoT Operations running on Arc-enabled Kubernetes and using MQTT broker. The architecture adopts the Universal Scene Description format (OpenUSD) to ensure that 2D, 3D, and point cloud geometry from the warehouse can be seamlessly integrated into the digital twin. Microsoft Fabric takes up the data in the cloud to provide a unified analytics foundation. Eventstream and eventhouse capture incoming telemetry as real-time streams or batch data. Microsoft OneLake acts as the governed, centralized data lake that consolidates all warehouse data. Digital twin builder transforms raw IoT signals into a contextualized virtual representation by mapping telemetry to the warehouse’s digital model. Powered by NVIDIA Omniverse, high-fidelity simulation and spatial computing occur creating a real digital twin which is streamed directly to the browser—eliminating the need for high-end local hardware. Tools such as Microsoft Copilot Studio and Microsoft Foundry enable natural language interaction. Across all stages, security is maintained through Azure Arc, ensuring consistent governance, configuration, and policy enforcement across edge and cloud.

SoftServe has proven to be an excellent delivery partner for digital twin applications. Together with Microsoft, they seamlessly integrated NVIDIA Omniverse-based environments into the beverage production simulations at Krones, enabling physical-accurate digital twins that reduced cycle times from hours to under five minutes. Similarly, at Toyota Material Handling Europe, SoftServe built a digital twin for simulating autonomous forklifts in virtual warehouse environments, enabling rapid testing, optimization, and safer deployments, helping to reduce the training times of autonomous systems by more than 30%.

TeamViewer’s augmented reality platform Frontline provides an additional simulation angle. Wearables such as smart glasses or wrist-mounted devices bring data seamlessly to frontline workers to get guidance in a hands-free manner for picking and packing as well as AI‑assisted counting. At DHL Supply Chain, TeamViewer’s solution is deployed globally to support vision picking of over 1,500 workers across 25 United States sites with fully hands‑free processes.

Agentic supply chains: The multi-agentic web

Agentic supply chains mark a new era of autonomous AI systems that proactively manage and optimize end-to-end supply chain operations. These agentic systems aim to continuously improve overarching KPIs like operating margin or cash conversion as well as specific KPIs such as lead time or freight cost per unit, ensuring that every agentic action contributes to measurable business impact.

Agentic supply chains are built on today’s human-driven tasks and encode the underlying decision-making logic. They include single purpose agents such as “troubleshooters” that constantly diagnose issues and propose fixes as well as “orchestrator agents” like planners or organizers that coordinate multistep workflows. These agents become functional through modern data fabrics, robust systems of record, and event-driven architectures that provide real-time information and governance.

Below is an overview of supply chain agents we have identified along the value chain through multiple customer and partner discussions.

Frontier Firms have already created value with multi-agentic systems.

• CSX Transportation has deployed a multiagent system that validates customer eligibility, routes complex requests, and supports rail operations with multistage coordination.
• Dow Chemical operates invoice analysis agents that review thousands of freight invoices each day, automatically detecting discrepancies and saving the company millions across its global shipping network.
• C.H. Robinson has rolled out a large fleet of generative AI agents including fast quoting agents that deliver tailored freight quotes and automating key steps along the shipping lifecycle.
• Blue Yonder has created an off-the-shelve Inventory Ops Agent on the Microsoft Marketplace that identifies supply–demand mismatches in real-time and recommends corrective actions such as alternate sourcing or demand swaps to keep inventory levels optimized.
• Resilinc offers an agentic supplier risk platform on Azure with pre-built AI agents (like for disruption, tariffs, and compliance) that autonomously evaluate potential impacts, initiate supplier engagement and recommend mitigation strategies.
• o9’s Digital Brain platform on Azure has been enhanced with various AI agents taking over simple tasks like getting specific data and more complex like creating full demand reviews.
• GEP recently added to their source-to-pay GEP SMART and supply chain solution GEP NEXXE (both built natively on Azure), a portfolio of AI agents that cover sourcing, negotiation, contract lifecycle, spend analysis, and market intelligence.
• Kinaxis offers its Maestro supply chain planning platform including AI agents that sense disruptions, run scenario simulations, and provide prescriptive insights through natural language.

Additionally, several delivery partners have used Microsoft tools like Microsoft Foundry and Copilot Studio to build agents for customers at high speed.

• Avanade, for example, offers concrete off-the-shelf agents including a trade compliance agent, a digital twin agent, and lead-2-cash agents.
• Softserve offers a catalyst with 3 customized agents in 30 days.
• NTT DATA is developing a simulation-driven, agentic AI-powered decision-support system for supply chain network agentic rebalancing platform.
• PwC delivers end-to-end agentic AI consulting services, powered by distinctive industry expertise.
• Capgemini is currently building an agentic end-to-end offering using the latest Microsoft IQ technology to be released at Hannover Messe in April 2026.

Microsoft Work IQ, Foundry IQ and Fabric IQ together form an intelligence layer for supply chains—from demand planning to inventory and customer service—that connects how people work, how the business operates, and what the organization knows. This gives AI agents full enterprise context so that agents can reason, simulate scenarios, and act in line with real-world constraints and KPIs such as inventory turnover to support better decisions.

Together with our strategic partner Celonis we have developed a new reference architecture leveraging Fabric IQ and the Celonis Process Intelligence Graph to transform fragmented supply chain data into agentic workflows. A collaborative stack that integrates raw data at the bottom and creates intelligent, automated actions at the top.

On the System of Record (SoR) layer, data is often siloed and does not “speak the same language,” leading to a fragmented understanding within the supply chain. Microsoft Fabric unifies this data through mirroring, streaming, or multi-cloud shortcuts with the goal to create a zero-copy connection and ensure the data is fresh and accessible without the weight of traditional extract, transform, and load (ETL) processes. Fabric IQ provides a reasoning layer that translates raw, unified data in OneLake into context-aware insights. This is the basis for Celonis’ Process Intelligence (PI) Graph which sits between data and the automation and uses process mining to map out how the supply chain actually runs—generating operational supply chain insights and suggesting improvement potentials from a process point of view. It communicates with Microsoft Fabric through Rest APIs, providing the knowledge and context that AI needs to make sense of the data. The agentic layer is divided into three functions:

• Building: Use Copilot Studio, Microsoft Foundry, and Power Automate to create custom AI agents.
• Orchestrating: Use MCP and Agent2Agent Protocol (A2A) to manage how different agents work together.
• Overseeing: Use the control plane for agents in Agent 365 (generally available in May 2026) to monitor agent activities and ensure performance as expected.

On the top layer, with the help of Microsoft Entra ID, insights and suggested actions are shown in tools employees use, such as Microsoft Teams, Microsoft 365 Copilot, Dynamics 365, Power Apps or in the Celonis UI.

A large global pharmaceutical company is using the above architecture to unify fragmented logistics data, enabling real-time identification of temperature-critical pharmaceutical returns and designing an agentic return process that unlocks multi-million euro annual productivity gains. Uniper automated material and service needs with Celonis and Microsoft. Microsoft Copilot in Teams and Power Automate orchestrate approvals, SAP actions, and replace manual component planning with proactive, agentic workflows that ensure timely material availability.

Physical AI: From warehouse handling to last mile deliveries

Physical AI is the final evolution of supply chain intelligence, building on simulations and agentic AI and embodying that intelligence directly in the physical world. In the near future, humanoid robots and robotic systems will physically take over more and more operational tasks along supply chains and logistics: from trailer unloading and sorting, pallet handling and replenishment, to packing and labelling and autonomous last‑mile deliveries. As intelligence moves from screens into machines, supply chains and logistics may gain a new level of physical agility.

Microsoft is pushing the frontier of physical AI with it’s new Rho‑alpha robotics model that combines natural language, visual perception, and tactile feedback to make robots more adaptive and autonomous. Microsoft has launched an early access research program with selected partners to advance co‑training and domain adaptation and aims to integrate the model in Microsoft Foundry in the coming months. Already today, customers and partners may take the below robotics toolchain reference architecture to train and deploy warehouse robotics with NVIDIA Osmo on Azure.

This toolchain is an open-source, production-ready framework that integrates Azure cloud services with NVIDIA’s physical AI stack, from simulation to training and deployment. It combines Azure Machine Learning, Azure Kubernetes Services (AKS), Microsoft Fabric, Azure Arc, and NVIDIA’s robotics and AI stack. NVIDIA Isaac Sim and Isaac Lab enable high-fidelity simulation and reinforcement learning, while NVIDIA OSMO orchestrates scalable training workflows across cloud and edge environments.

Detailed information can be found here.

Hexagon Robotics has started to deploy this architecture using Azure IoT Operations as well as Fabric Real-Time Intelligence in Microsoft Fabric to provide production-ready humanoid robotic solutions. Their industrial humanoid robot, AEON, combines dexterity, locomotion, and unique spatial intelligence to tackle complex industrial use cases for warehousing and logistics such as inspection and inventory taking.

Figure AI, funded by Microsoft, enables the deployment of their humanoid robots in real-world logistics environments using Azure’s AI infrastructure. Their latest model Figure 03 can take over warehouse tasks such as sorting packages at conveyor belt speeds and help at last-mile delivery with near human-level precision.

KUKA and Microsoft jointly developed iiQWorks.Copilot, an AI-powered assistant that enables natural language robot programming and significantly simplifies automation tasks. By integrating Azure AI services, the solution allows users to design, test, and deploy robot workflows faster and more safely—cutting programming time for simple tasks by up to 80%. This has benefitted all KUKA robotics deployed in warehouses and logistics.

Wandelbots’ NOVA software layer combined with Azure cloud services unifies heterogeneous robots and brings adaptive automation to the shop floor. Wandelbots NOVA streamlines warehouse and fulfillment operations such as palletizing by simplifying robot programming, accelerating deployment, and enabling AI-powered path planning and scaling across multiple robot brands. Together, these capabilities position Wandelbots NOVA as a physical AI platform for orchestrating and scaling AI-powered automation across supply chain operations.

Get in touch with us

Contact us directly at screquests@microsoft.com or go to Microsoft for Manufacturing to explore how Microsoft technologies can transform your supply chain. Join us at Hannover Messe in April 2026 to hear directly from our industry leaders, explore cutting-edge ideas, and connect with peers.