The Cheapest Industrial IoT Board Is the Most Expensive One You'll Ever Buy
I'll go ahead and say it: the cheapest industrial IoT board is often the most expensive thing you can install in a plant. That is not a slogan. It's what I've learned after eight years of specifying, buying, and occasionally rescuing edge hardware for factory clients.
In my role coordinating hardware selection for industrial IoT solutions, I've handled over 70 rush orders. I'm the person who gets called at 4:00 PM on a Friday when a gateway dies on a production line. After enough of those calls, I stopped asking 'What is the board price?' and started asking 'What is the total cost if this fails?'
As of January 2025, I still get at least one panicked call per quarter about an obsolete carrier board. Usually it's a product that was fine on paper but couldn't survive the plant floor.
When I first started in this field, I assumed a low-cost single-board computer was a smart starting point for any automation project. The specs were fine. The price was hard to argue with. Then three things happened: a supplier couldn't commit to a delivery window, the Ethernet controller dropped off the bus under load, and a customer asked if the device had been tested at 60°C ambient. It hadn't.
The trigger event in March 2024 changed how I think about industrial alternatives to Raspberry Pi. A gas-fired heating line at a food processing plant had a failed controller. The original furnace control board carrier was obsolete, and the OEM quoted six weeks. The client called at 3:42 PM on a Thursday. If the line stayed down through the weekend, the plant would face a $50,000 penalty.
What saved us? A spare RK3576 SOM we had in stock, a partially finished carrier design from a previous project, and a decision to stop treating the module as the product.
Computer on Module (COM) Changes the Procurement Question
A computer on module (COM) is the brain. It carries the CPU, memory, and core I/O in a standardized footprint. The brain doesn't work in the field without a carrier. The carrier brings out the interfaces, conditions power, drives relays, reads sensors, and survives the noise of a factory floor.
People often forget that part. The most expensive lesson in industrial IoT is that the system is the carrier, the enclosure, the power supply, and the software image, not just the chip.
According to PICMG, the COM Express specification defines a standardized pinout for modules and carrier boards. That is exactly why you can change the module and keep the carrier. It turns a hardware refresh from a redesign into a swap.
In April 2023, we had a project stuck for six weeks because a consumer-grade SBC had a 20-week backlog. The distributor quote was $95 per board, but the cost of waiting was far larger. After that, we standardized on an RK3576 SOM for new designs. The module itself was maybe $42 more than the Raspberry Pi board we had started with, though I might be misremembering the exact comparison after volume discounts. The vendor gave us a written availability commitment, a real industrial temperature range, and a carrier design that made sense in a DIN-rail enclosure.
What I mean is that a module with a supply guarantee is worth more than a module with a lower price. If I can't buy the same board next quarter, then every future deployment is a gamble. And in industrial IoT, gambling is the most expensive line item on the budget.
Edge Computing in Cloud Computing Is a System, Not a Dashboard
Edge computing in cloud computing isn't a separate universe. It's the front end of the cloud. If the edge device is down, the cloud is blind. You can have a beautiful analytics dashboard, but if the gateway doesn't survive the environment, the dashboard is showing you data from a dead machine.
Why does this matter? Because a cheap board that goes offline twice a month doesn't just lose data. It erodes trust in the entire industrial IoT program. The plant manager starts ignoring alerts. The maintenance crew stops using the mobile app. The project dies quietly.
The RK3576 SOM helped us in this regard. Its integrated NPU gave us enough on-device intelligence to filter data before sending it upstream. That meant we could run basic anomaly detection right at the edge and only push meaningful events to the cloud. It also meant the system still worked during internet outages, which was not something my clients expected.
Put another way: edge computing in cloud computing is a division of labor, not a choice between two worlds. The edge handles the fast, critical decisions. The cloud handles the long, analytical ones. If you starve the edge, the whole system starves.
A Furnace Control Board Carrier: Where 'Details Are Details'
Back to the March 2024 call. The failed component was a furnace control board carrier. Not a motherboard, not a PLC—a carrier designed to hold a processor module and connect it to gas valves, flame sensors, fans, and safety interlocks. The original carrier was an old design, built around a module that had been discontinued. The OEM wanted to replace the entire furnace controller assembly. That quote was around $8,400 per unit, with a six-week lead time.
Our fix cost less and took less time. We took the RK3576 SOM, adapted the carrier prototype we had been working on, added the correct relay outputs and a watchdog circuit, and flashed a Yocto image with a Modbus TCP stack. The client had a working unit early Monday morning. So glad I had stocked a spare module rather than ordering one at the last minute. Almost didn't. And that would have meant the $50,000 penalty.
Do I recommend every factory build its own furnace control board carrier? No. That was a unique situation. But it shows the value of thinking in modules instead of appliances. The module is the long-term investment. The carrier is the application-specific part. If you own the carrier design, a hardware failure no longer requires an OEM lightning strike.
Price the Whole System, Not Just the Board
My rule after all these rush orders: calculate total cost of ownership over at least five years, including engineering time, procurement risk, replacement cost, and the cost of downtime. Let me give you an example from our internal project data.
In Q3 2024, we compared two designs for a 50-unit deployment of monitoring gateways. Design A used a consumer SBC and an aftermarket I/O shield. Design B used an RK3576 SOM and a carrier we had already qualified for another client. Per-unit BOM was $395 for Design A and $470 for Design B—a gap of only $3,750 over 50 units. The real gap was schedule. Design A's lead time was 14 weeks; Design B was 4 weeks. That 10-week difference delayed a plant project and cost an estimated $18,000 in lost production. Then Design A's vendor released a firmware update that broke our boot chain, and the client's security review flagged the bootloader. After rework and a second review, Design A wasn't the budget option. It was the expensive option.
That $75 per-board advantage turned into an $18,000 schedule problem before we even added the rework hours. Put another way: the cheap board won on paper, lost on the invoice.
One more thing: support. A consumer SBC vendor supports one product line for a few years, then moves on. A COM vendor has an incentive to keep modules and carriers compatible across generations. That means I can update the module without redesigning the carrier, which is something you cannot do with a closed board. In 2025, that's not a nice-to-have. It's a budget line.
'But what about time-to-prototype?' You might be thinking that. Fair. If you're building three bench units to validate a concept, a Raspberry Pi is a legitimate tool. I still use one in the lab. But a lab sample and a deployed industrial IoT solution are different products with different BOMs. The lab device can be hand-assembled and tethered to a bench power supply. The field device has to be manufactured, certified, and supported.
At least, that's been my experience across deployments ranging from 20 to 200 nodes. That said, if your volume is five units and a $10,000 failure won't sink you, a good COTS industrial PC might be the pragmatic call. The COM approach isn't a religion.
The most frustrating part of this work? Not the hardware. It's the assumption that any SBC can be turned into an industrial solution with a DIN rail clip and a prayer. You'd think that after a few high-profile supply chain crises, the lesson would stick. But I still get calls from engineers who picked a board because it cost $75 and now they're trying to source 400 units with no supply guarantee.
Industrial IoT Solutions Are System Decisions
The question shouldn't be 'Is this the cheapest board?' It should be 'Will this still be running in three years, with the operating system patched, the carrier intact, and the data flowing?'
Never quote a board without quoting the system around it.
An industrial alternative to Raspberry Pi isn't necessarily a different brand or a bigger budget. It's a different class of product. If the product is available, supported, and paired with a carrier that's actually designed for the application, it's not a premium. It's the cheap option.
I'm not saying go spend money on ruggedized everything. I'm saying price the failure, not just the board. Because in my role coordinating rush orders, I've watched too many 'savings' turn into fire drills. And every fire drill costs more than the money you saved.
