What Does AI Hardware Mean? A Buyer Definition Worth Testing
Most buyers already know how to source connected hardware: a radio module, a cloud account, an app, and a compliance file that travels with the shipment. AI smart hardware adds a layer that is easy to miss in a quotation. The device now runs on-device inference, which means a model, a compute budget and a data path sit inside the product itself. That changes what you are buying, what you must document, and which claims you can safely put on the box. This article sets out a working AI smart hardware definition for procurement purposes and shows where the risk sits.
Where connected hardware ends and AI smart hardware begins
A connected device can be fully specified by its radio, its sensor set, its power budget and its cloud dependency. An edge AI device adds a fourth variable: the inference workload. Ask whether inference runs locally, in the cloud, or both, and what happens when the network is unavailable. A device that stops working without a connection is a different product from one that keeps functioning offline, and the two carry different support costs.
For procurement, the practical test is simple. If the supplier cannot state the model version, the memory footprint and the update path, you are buying connected hardware with an AI label. The AI device line is a useful reference for how those questions are normally answered in a specification sheet.
The claims question: what the hardware can and cannot say
Claims are where AI hardware projects fail commercially, not technically. A wellness metric is guidance, not a diagnosis. A translation feature produces reference output, not a certified translation. A companion robot is not a monitoring or emergency-response device, and it does not detect falls or contact anyone on your behalf. If your marketing copy implies any of those things, you own the consequences in your market.
Write the claim wording into the purchase order and tie it to a hardware revision. When firmware changes, the claim should be re-checked rather than assumed. This is the single cheapest risk control available to an importer.
Documents to request per model and hardware revision
Compliance is scoped to a build, not to a brand. Ask for the file that matches the exact revision you are shipping, and treat any gap as a reason to pause. Certification is a test scope booked with an accredited laboratory for each hardware revision, so a certificate from a previous revision does not automatically cover the next one.
| Item | What it covers | Buyer action |
|---|---|---|
| CE RED | EN 300 328, EN 301 489, EN 62368-1 | Match the document to the shipped revision |
| FCC Part 15 B/C | US emissions and radio | Confirm the module and antenna as built |
| UKCA | Great Britain market | Request on order, do not assume |
| RoHS 3 / REACH SVHC | Substance restrictions and screening | Keep for customs and retail audits |
| UN38.3 + SDS | Lithium battery transport | Required before air freight booking |
| EN 1811 | Nickel release on skin-contact parts | Relevant for rings and wearables |
| Toy scope | EN 71, EN 62115, Directive 2009/48/EC | Decided by the age grading on the packaging |
Commercial terms that differ from standard connected hardware
Tooling and firmware work dominate the timeline on AI products, so plan the calendar before the price. The OEM/ODM services page sets out how a project moves from specification to build, and OEM versus ODM explains which decisions stay with you.
- MOQ is 500 units as standard, with logo at 500 and custom packaging at 1,000.
- Firmware or app white-label starts at 2,000 units; AI glasses at 300; ring, education and companion models at 1,000.
- Service robots are quoted per project from 50 units.
- Samples ship in 3-5 days from stock, or 7-15 days configured, with sample cost credited against mass production.
- Production runs 25-35 days, education 30-40 days, and ODM tooling 12-20 weeks.
- Terms are EXW, FOB Shenzhen, or DDP by air or sea; warranty is 12 months with 2 % free replacement parts shipped with the order.
Two further points decide whether a project is viable. Google Play and GMS are not licensed by default, so an Android-based device may need a different application route. And EU or UK importer duties such as WEEE, BattG, DEEE/Triman, EORI and VAT sit with you as the importer, not with the factory. Build those costs into the landed price before you compare quotations.
How to vet the supplier before the first tooling payment
Ask for the QC structure, not just a quality promise. A documented process with IQC, IPQC, FQC and OQC gates at AQL 2.5 major and 4.0 minor tells you how defects are caught, and the factory page shows how that runs in practice. For service robots, expect an EMC and low-voltage scope, a class 1 laser statement for the LiDAR, and a written risk assessment. Where a supplier cannot produce these, the correct answer is no.
It also helps to compare categories side by side. The product catalogue and the AI learning devices range show how the same compliance logic applies across very different builds, and the supplier vetting guide lists the questions that separate a real manufacturer from a trading desk. More background sits in the journal and the wider buyer resources library.
The next step is not a quotation. Write a one-page specification that fixes the inference location, the claim wording, the age grading where relevant, and the document list per hardware revision, then send it to two or three suppliers and compare their answers line by line. You can send us the specification and we will tell you which parts we can build, which parts need a different approach, and where the compliance scope has to be booked before tooling starts.
Questions buyers ask
What is the difference between AI smart hardware and connected hardware?
Connected hardware only needs a radio link and a cloud service to work. AI smart hardware performs inference locally on the device, so the model, the compute and the data handling are part of the product you are buying, not just the app behind it.
Do I need extra compliance documents for an edge AI device?
The radio, safety and battery documents are the same as any connected product. What changes is that you should also ask for the model version, the data flow and any claim wording to be fixed per hardware revision, because a firmware change can alter what the device does.
Related: more on ai smart hardware · buyer guides · customisation tiers