How to Buy System on Modules and AI Edge Hardware Without Getting Ripped Off: A Cost Controller's FAQ

I've spent the last six years managing procurement for embedded and IoT hardware at a company that builds connected control systems. Our annual budget for boards, modules, and assembly runs around $500,000. In that time I've reviewed quotes from dozens of SBC manufacturers, evaluated a handful of PCB assembly contract manufacturers, and tracked every failure back to a dollar figure.

If you're looking into system on modules, single board computers, or AI edge devices, you don't need another generic tech explainer. You need to know where the money goes and why the cheapest option often isn't. Here are the questions I get asked by our own engineers – and the answers I've learned the hard way.

• What's the real difference between a system on module (SOM) and a single board computer (SBC)?
• Why do similar-looking SBCs have such different prices?
• When should I use a PCB assembly contract manufacturer?
• Do IoT transportation and telecom applications really need overpriced boards?
• What hidden costs come with AI edge devices?
• Should I pay more for a SOM with a longer product lifecycle?
• What's the most overrated cost-saving tactic in hardware procurement?

What's the real difference between a system on module (SOM) and a single board computer (SBC)?

A SOM is basically a computer on a carrier board – it's meant to be embedded into your own PCB. An SBC is a complete, standalone board you can plug into with standard connectors. From a cost perspective, the difference is about how much engineering and integration you're willing to pay for. A SOM might cost $80–$150 more than a comparable SBC chipset, but if it saves you several weeks of layout work, the total project cost can actually be lower. I've seen teams spend $3,000 on custom PCB design to save $50 on a module – that's false economy.

Why do similar-looking SBCs have such different prices?

Let me give you a real example. In Q1 2025, I compared two SBCs with the same processor and memory from different manufacturers. One was $89, the other $129. The cheap one had lower-quality power supply filtering, no conformal coating option, and used a smaller connector that broke in testing. We ordered 50, had 7 failures in the first year, and spent $260 in replacement shipping plus 12 hours of tech time. Adjusted cost: $154 per board, not $89. The other board we've had one failure in 40 units. That's how a $40 difference becomes a $65-per-unit swing in total cost.

When should I use a PCB assembly contract manufacturer?

If your volumes are over a few hundred boards a year, you should seriously consider a PCB assembly contract manufacturer. The reason isn't just labor cost – it's test equipment and reliability. We moved from hand-soldering to a contract manufacturer in 2023 because they offered airborne and inline testing that our shop couldn't. Their per-board price was 12% higher, but our failure rate dropped from 4% to 0.8%. For a run of 500 boards that's 16 fewer rework cycles at $40 each – $640 saved in rework. That more than covered the premium.

Do IoT transportation and telecom applications really need overpriced boards?

This is where I see the biggest hidden costs, honestly. In transportation and telecom, the operating environment is brutal: wide temperature swings, vibration, power surges. A board that's perfectly fine in an office will die on a bus or in a roadside cabinet. A $75 commercial-grade SBC might need external insulation, heating, or redundant power supplies to survive – adding $30–$50 in components per unit. Meanwhile, a $110 industrial-grade SOM with an extended temperature rating does the job without any add-ons. The 'cheap' solution ends up costing more, plus you own the field-failure risk. To be fair, if you're building lab prototypes only, the commercial board is fine.

What hidden costs come with AI edge devices?

AI edge devices are the worst for quote comparison because the core chip is only part of the story. The hidden costs are memory, thermal management, and certification. A $200 AI module might need 8GB RAM to run the actual workload, but the vendor's datasheet glosses over that. Then add a heatsink or fan, plus FCC/CE testing if you're using a new wireless chip. I've seen projects blow $15,000 on certification because they picked a board with an unapproved RF section. When evaluating AI edge devices, request the full BOM and ask about the thermal design – not just the NPU specs.

Should I pay more for a SOM with a longer product lifecycle?

Yes – and this has saved our budget more than any other single decision. In mid-2024, we chose a SOM with a seven-year lifecycle guarantee over a cheaper one with no commitment. The pricier module added $12 per unit, but it meant we didn't have to redesign a control board when the chip went end-of-life in early 2025. A redesign runs us about $8,000 in engineering and 10 weeks. On a 500-unit project, that's $16 per unit saved – not counting the delay. The conventional wisdom is that you buy for today's specs. My experience with 200+ orders says buy for tomorrow's supply risk.

What's the most overrated cost-saving tactic in hardware procurement?

Getting multiple quotes and automatically taking the lowest one. It's not that you shouldn't compare – it's that you shouldn't compare just price. In 2022, I switched to a cheaper PCB assembler that saved $2,200 on an order. They shipped boards that failed at a 3% rate in the field, and the warranty and rework costs came to $3,400. The next time, I built a total cost spreadsheet: unit price, test costs, supplier failure rates, yield losses, expedite fees, and hidden minimums. After comparing six vendors over two weeks using that spreadsheet, we picked one that wasn't the cheapest but had a 99.5% historical yield. Since then, our cost overruns on hardware have dropped by about 40%.