No matter where it shows up across operations, AI has two jobs in manufacturing:
- Act fast and take immediate action at the machine
- Learn across operations so performance keeps getting better as time goes on
But these workloads don’t have the same requirements. Some decisions have to happen in real-time, right where the data is created. Others depend on analyzing patterns across production lines, plants and fleets.
To handle everything AI is being asked to do, most manufacturers assume they need two computing environments: the edge for speed and responsiveness, and the cloud for scale and training.
But the future of AI in manufacturing depends on using both well.
Why a cloud AI vs. edge AI mindset falls short
Cloud AI and edge AI are usually talked about as an either/or decision: You pick the one environment that best fits your workload and build around it. But that approach forces unnecessary tradeoffs.
Why cloud-only isn’t enough
A cloud-only approach can make sense when your goal is to train models or compare data across sites for a broad view of performance. But it falls apart when your application depends on immediate response at the machine. Sending every signal to the cloud can create unnecessary costs and bandwidth issues.
Even a short delay matters when the model is watching for a fault or anomaly so it can act; the data has to travel to the cloud and back before anything can happen.
Why edge-only isn’t enough
An edge-only approach solves the speed problem but creates new limitations. Edge processing is fast but isolated; it doesn’t offer visibility into what’s happening across locations. Without the cloud, models can’t learn from broad datasets, and patterns remain within sites or are missed altogether.
For instance, if a vibration pattern on a CNC machine turns out to be an early warning sign of spindle failure, other plants running similar equipment will never benefit from what was learned from that one machine.
Hybrid AI is a better path forward
The future of AI in manufacturing is local-first and cloud-enabled. Not every signal needs to leave the plant floor.
That’s why many manufacturers are moving toward a hybrid model: cloud AI + edge AI instead of cloud AI vs. edge AI. Each layer plays a central role in solving different parts of the same industrial challenges. Together, they create an immediate and scalable approach to decision-making on the plant floor, which is what industrial AI needs.
The edge handles the work that needs to happen right away, and the cloud supports broader learning, fleet-wide coordination, and long-term optimization. This lets manufacturers keep decisions close to the machine without losing sight of fleet-wide trends or long-term performance gains.
Hybrid AI in action: anomaly detection
Let’s talk about real-time anomaly detection as an example of what a local-first, cloud-enabled architecture can do.
We shared this demonstration at HANNOVER MESSE 2026 in April, along with Storm Reply and Amazon Web Services (AWS), to show how a hybrid AI architecture powered by Belden’s complete connection solutions delivers the best of both worlds. The response that followed made one thing clear: Manufacturers are eager to find ways to make cloud AI and edge AI work together.
In this setup, the edge device watches for abnormal machine behavior, filters what matters and keeps raw production data on the shop floor. This is done by running AWS IoT Greengrass v2 at the Belden edge device for data acquisition and preprocessing and using an Isolation Forest AI model for behavior scoring.
We also ran CloudRail as a containerized Greengrass component for IO-Link data normalization. This lets the system analyze motor behavior in real-time, right where the data is generated. There’s no waiting on a remote system to respond. The result is faster decisions and less time between detection and action.
Only the relevant insights and metrics moved to the cloud periodically to train models and support cross-site visibility. This ensures efficient bandwidth usage and reduces cloud ingestion and storage costs. This is done through batched MQTT uploads (every 10 minutes) to AWS S3 instead of per-sample streaming with IoT Core. AWS cloud services were used for model training/versioning; the updated models are pushed back to the edge via Greengrass.
What the numbers show
With this setup, the results were outstanding and served as a real reflection of the power of combining edge and cloud AI:
-
Inferencing time: ~63 ms edge vs. ~442 ms cloud
-
Backhaul traffic reduction: ~40% less
-
Availability during a simulated WAN outage: 99.2% hybrid vs. 5.8% only cloud
The third piece of the AI equation: the network
The future of AI in manufacturing brings together real-time intelligence at the edge, scalable intelligence in the cloud and a reliable OT network that connects them both.
A deterministic, resilient network makes sure the right data gets to the right place at the right time. It lets the edge make decisions in milliseconds and lets the cloud learn from what the edge “sees.”
Without it, hybrid AI breaks down before it can deliver value.
Belden’s complete connection solutions support AI end to end, from real-time decisions at the edge to long-term learning in the cloud. When you work with our experts, your manufacturing organization has a clear path forward for its next AI strategy: Build for the edge, build for the cloud and build the network that makes both work.