Edge vs cloud on the line

Artificial intelligence in factories is splitting into two jobs: fast decisions stay on site, while heavier analysis moves to the cloud. (docs.aws.amazon.com) In plain terms, “edge” means software running inside the plant, close to cameras, robots, and control systems. Siemens says its Industrial Edge apps run on shop-floor devices and are managed centrally, while Microsoft says Azure IoT Edge lets companies run artificial intelligence and business logic directly on local devices. (docs.eu1.edge.siemens.cloud) (azure.microsoft.com) That setup is built for timing. Amazon Web Services says some manufacturing execution system functions “require low latency and cannot tolerate intermittent connectivity to the cloud,” which is why those services are better suited to run on premises. (docs.aws.amazon.com) The cloud still handles the bigger lift. Google Cloud’s Manufacturing Connect splits its product into factory software deployed within the plant and a cloud application in the customer’s Google Cloud tenant for managing edge instances, while Microsoft’s 2025 manufacturing release added factory data tools in Microsoft Fabric and Azure Artificial Intelligence for analysis and decision support. (docs.cloud.google.com) (learn.microsoft.com) Vendors are now building products around that division instead of treating edge and cloud as rivals. Siemens said its Industrial Edge platform integrates with Microsoft Azure IoT Operations so manufacturers can move data and workloads between production lines and cloud systems for artificial intelligence, digital twins, and simulation. (press.siemens.com) Google is making the same case with different plumbing. Its Manufacturing Connect documentation says customers can choose decentralized deployments close to individual lines or more centralized setups in a factory data center, then manage those edge systems from the cloud. (docs.cloud.google.com) The argument is not only about speed. International Business Machines defines data sovereignty as data being subject to the laws and governance structures of the country where it is generated, and says edge computing can help keep processing near the source even as companies debate how much control cloud providers retain. (ibm.com) Factory operators also have a cost problem. Edge systems can cut bandwidth and keep lines running during network outages, but they add on-site hardware and maintenance; cloud systems pool computing power, but sending every camera feed and machine signal off site can get expensive and adds delay. (aws.amazon.com) (docs.aws.amazon.com) That is why the market is converging on hybrid designs. Cisco said in February 2026 that intelligent manufacturing needs immediate data processing, security, and centralized cloud management in one platform, and Qualcomm and International Business Machines said in February 2025 that they were expanding generative artificial intelligence systems “from edge to cloud” for privacy, reliability, and efficiency. (cisco.com) (qualcomm.com) The practical rule emerging on factory floors is simple: if a model has to stop a machine in milliseconds, it runs near the machine; if it has to compare months of production across plants, it usually runs in the cloud. (docs.aws.amazon.com) (learn.microsoft.com)