The False Binary
Every manufacturing executive I meet asks the same question: to deploy AI (Artificial Intelligence) at the edge or in the cloud. Having worked on global MES rollouts, I can tell you that they are asking the wrong question entirely.
The false binary (edge or cloud) has led to a siloed approach. CIOs advocate for a cloud-first AI strategy to leverage the cost-effectiveness and scalability of the cloud, while plant managers demand edge computing solutions that promise real-time responsiveness.
This is also mirrored in recent reports by Fortune Business Insights, where manufacturing technology spending is split between edge and cloud AI investments. While the global edge AI market was valued at $27.01 billion in 2024, the global cloud AI market was valued at $78.36 billion.
However, based on my experience, the question is not an “either/or” scenario. The factories that will dominate the future won’t choose one; they will orchestrate a unified and intelligent system that leverages both.
Why Manufacturers Need Both
The smart factory of the future needs a hybrid approach that leverages the complementary strengths of cloud AI and edge AI.
Cloud AI excels in handling big data at scale. As recent research cites its immense computing power with processing speeds of vast datasets at 20-45 seconds for data ranging from 15-20 TB/day. The result is that data from different plants can be consolidated to identify fleet-wide trends and continuously retrain sophisticated machine learning models. In contrast, Edge AI provides the reflexes – on-the-spot intelligence – by processing data at the source, enabling safety and quality control decisions.
Forward-thinking manufacturers know that they need the scalability of cloud AI to retrain models using data from different plants to identify optimization opportunities, as reiterated by IBM research.
However, as DataBank cites, edge AI works as the nervous system of the factory floor, handling local responses and facilitating millisecond intelligent decision-making without waiting for connectivity right at the source.
“Smart manufacturers who fail to adopt the hybrid solution risk slowing down their operations (cloud only) or fragmenting them too much (edge AI).”
Use Case Examples of a Hybrid Approach
One of the clearest examples of the transformative impact of the hybrid edge and cloud AI demonstrated by the MES I’ve supported is predictive maintenance. A study by Bala et al. found that AI-based predictive maintenance can reduce costs by 30% and increase ROI by 900%.
By leveraging cloud infrastructure, the systems analyze vast amounts of equipment data to identify failure patterns and predictive indicators that human experts are likely to miss.
This intelligence is deployed to edge devices, monitoring individual equipment in real-time. Edge AI algorithms controlling the equipment adjust the operating equipment without latency to prevent failure while alerting maintenance teams.
In one case involving a global manufacturer client, the hybrid predictive maintenance systems reduced downtime by around 34%. The cloud AI systems continually optimize machine learning models for accurate predictions, while edge AI makes real-time adjustments to the equipment based on machine vibrations and significantly reduces downtime risk.
A further case example concerns the use of AI co-pilots to facilitate human-machine interaction on the factory floor through natural language processing. When a shift operator asks, “Why is line 6 running so slowly?” they need real-time answers that can only be delivered by well-orchestrated hybrid systems.
The cloud-based large language models (LLMs) understand complex manufacturing terminology and provide reasoning capabilities, while edge systems provide access to real-time local data such as equipment status and production metrics.
When shift operators query the MES in natural language, the hybrid cloud-edge AI systems deliver a conversational AI with high accuracy. The edge systems gather local data and feed the contextual information to cloud systems, which generate intelligent responses.
As observed by one client through internal benchmarking, up to 25% improvement in production outputs by leveraging AI co-pilots in MES systems used by supervisors to visualize and analyze shift data. The orchestration of the edge AI and LLMs in our cloud infrastructure ensured that highly accurate responses could be delivered promptly.
“I believe this is the way forward for the factory of the future – combining the power of the cloud with edge computing.”
Yet, a third case example was the use of the MES as a carbon tower, facilitating the reduction of emissions in real-time – a finding reiterated in a report, highlighting how AI and IoT sensors continuously monitor scope 3 emissions among manufacturing companies in real-time.
For another client, the MES platforms combined edge and cloud AI to track real-time emissions from individual machines. In this case, edge AI accessed real-time data on consumption and emissions from IIoT sensors at individual devices.
This data was fed to the cloud LLMs for optimization based on global market prices and production schedules. The output from the cloud, comprising updated efficiency parameters, was sent to the edge devices, allowing the system to make data-driven decisions aligned with ESG goals.
A well-orchestrated edge AI and cloud AI system within our MES allowed the client to enable significant emissions reduction aligned with ESG goals.
Common Pitfalls
The success of hybrid cloud-edge AI systems is challenged by several common pitfalls. At the outset, I believe that the main issue is poor internet connectivity, leading to latency and delays, even as small as milliseconds.
I have seen first-hand the high costs of a cloud-only approach. A single network disruption can grind manufacturing lines to a halt, costing companies tens of thousands of dollars. This is a significant problem that cannot be ignored in mission-critical environments.
Network connectivity is a significant problem and also affects edge AI systems, which create intelligent islands that can’t learn from each other or benefit from centralized LLMs in the cloud.
I believe that unreliable connectivity should not be allowed to occur within factory floors using hybrid cloud-edge AI systems because it renders them inoperable. What is needed is designing systems that degrade gracefully when connectivity fails, while continuously improving when it is available.
A second mistake is the siloed approach. We have cloud-purists who, although champions for its scalability, ignore the fact that their systems can be crippled by network latency. On the other hand, opting for an edge-only solution leaves intelligence fragmented, thus making it difficult to aggregate insights across a global fleet.
What is in fact happening is that smart manufacturing companies opting for the siloed approach are missing out on opportunities for cross-plant optimization and centralized learning.
We need to leverage pragmatic hybrid intelligence at the factory, and considering a modular deployment approach will decouple these functions and ensure orchestration between the systems.
The Path Forward: Modular Hybrid AI
The path forward to the factory of the future, one powered by a hybrid edge-cloud AI paradigm, requires that we build a modular AI architecture into a composable stack. There is an urgent need to consider these as interchangeable modules in a larger orchestration framework.
At the granular level, this means using protocols such as MQTT and OPC UA to facilitate seamless communication and ISA-95 compliance to ensure integration with the existing manufacturing systems.
This will transform the cloud into the training ground and the edge into the execution environment. We can then use containers and microservices to facilitate composability, enabling manufacturers to deploy and scale AI across their fleet with significant speed and efficiency.
Such an approach will enable manufacturers to focus on the business use case rather than being constrained by infrastructure. Further, using federated learning frameworks such as Tensor Flow Federated (TFF), FedML, and OpenML, smart manufacturers can engage in collaborative machine learning without sharing sensitive data.
The modular approach is the future and may represent one of the most scalable pathways to factory AI maturity. By separating inference from training pipelines, we ensure our systems are resilient and continue learning. Adding modularity in the deployment will also ensure prototypes are generated rapidly without the overhaul of the entire infrastructure.
Conclusion
With the rapid evolution of smart manufacturing catalyzed by Industry 5.0, there is an increasing demand to orchestrate systems that optimize the well-being of humans. Yet for too long, the false binary has influenced leaders to opt for cloud or edge AI deployment. Despite this, fast-thinking smart manufacturers know that the future and Industry 5.0 demand a symbiotic relationship between machines and humans.
To successfully deploy hybrid AI in smart factories, we need to sensitize manufacturers that cloud and edge are not rivals but complementary partners that deliver the comprehensive intelligence needed for business success. Only hybrid AI architectures can ensure manufacturers compete effectively in the market even as sustainability requirements tighten.
The factory of tomorrow is not defined by a single technology but by a seamless union of them. It’s a system where the cloud’s intelligence and the agility of edge computing work together. It’s time we stop choosing and start combining.
(Photo by Growtika on Unsplash)
