A thousand of them
The BOM does collapse, roughly as expected — about $103 of hobby parts becomes about $30 of production parts. And then it stops being the interesting number, because four things that cost nothing at quantity one now cost more than all the silicon put together.
1 · What the BOM does
Modules become chips, retail becomes reel, and the hobby premium disappears. Indicative quantity-1,000 pricing:
| Line | At one | At 1,000 | What changed |
|---|---|---|---|
| ESP32-S3 module | $15 | ~$3.00 | Devkit → the bare WROOM-1 module soldered down. Still a module, deliberately — see §3. |
| NFC reader | $40 | ~$4.00 | Breakout → PN532 die and a PCB-trace antenna. The single biggest drop, and the one with a certification string attached. |
| Amplifier | $6 | ~$1.20 | Breakout → MAX98357A in a QFN |
| Speaker | $3 | ~$1.00 | — |
| Storage | $12 | ~$2.80 | Breakout → a microSD socket; card bought on a reel |
| Power (charger, buck-boost, cell) | $60 | ~$6.20 | Three breakouts → two ICs and a pouch cell |
| Controls, connectors, passives | $10 | ~$2.70 | The panel harness mostly stops existing — it becomes copper |
| PCB + assembly | $5 | ~$3.50 | Protoboard and an evening → a 4-layer board and a pick-and-place |
| A deck of 52 chipped cards | $14 | ~$6.00 | Barely drops, and becomes the second-largest line. Below. |
| Enclosure + packaging | $15 | ~$4.00 | Depends entirely on tooling — §4 |
| BOM | ~$180 | ~$34 | with the enclosure and a full deck |
Indicative, not quoted. Treat every figure as ±30% until a distributor has actually priced a line.
2 · The surprise: the cards cost more than the microcontroller
Shipping a full deck means 52 tags per box, 52,000 tags across the run. Printed PVC NFC cards land somewhere around 10–15¢ each at that volume, so a deck is roughly $6 — twice the ESP32 and second only to the whole power subsystem.
This is the one place the design and the business disagree with each other. Three ways out, and they are genuinely different products:
- Ship ten cards, sell decks. The obvious move, and it quietly turns tape into the thing it was built to not be — a razor with blades.
- Ship stickers, not cards. 25 mm NTAG213 stickers are a third the price of printed cards and the buyer supplies a deck of playing cards from a supermarket. Cheaper, more charming, entirely in keeping with the project, and it puts an assembly step on the customer.
- Ship the full deck and eat it. Honest, and it is $6.
Worth noticing that this decision was invisible at quantity one, where cards are 14¢ and nobody counts them.
3 · Certification, and the decision that already paid for itself
A radio, a lithium cell, and a four-year-old. That is three regulated things at once, and it is where garage hardware usually stops.
| What | Why it applies | Indicative |
|---|---|---|
| FCC Part 15B, SDoC | Any digital device sold in the US. Unintentional emissions from the finished product. | $1.5k–3k |
| FCC Part 15.225 | The 13.56 MHz reader is an intentional radiator in its own right. Chip-down means certifying it yourself. | $2k–4k? |
| CPC — CPSIA + ASTM F963 | It is a children's product. Lead and phthalate content, mechanical hazards, battery-compartment security, a permanent tracking label. | $0.5k–3k |
| UN 38.3 | Lithium cells cannot be shipped without a test report. Often supplied by the cell vendor — ask before you design one in. | $0–3k |
| CE / UKCA, EN 71, EN IEC 62115, RED | Only if you sell into the EU or UK. Effectively doubles the exercise. | $5k–15k |
| US only, once | ~$6k–12k | |
? The 15.225 figure is the least certain number on this page and should be quoted, not assumed.
The mirror image: going chip-down on the PN532 is the opposite trade. It saves perhaps $3 a unit and buys an intentional-radiator certification, an antenna you now have to tune, and a re-test every time the enclosure changes. At 1,000 units it is close to a wash and probably the wrong way round. At 10,000 it is obviously right.
4 · Tooling, and why you would not injection-mould at 1,000
A simple two-part moulded box is roughly $3,000–$15,000 of aluminium tooling. Across 1,000 units that is $3–15 each, on a part that costs $1.50 to mould — the tool costs more than the parts. Moulding starts paying back somewhere around 5,000–10,000 units.
So at a thousand you do what the one-off build does, industrialised: an off-the-shelf enclosure with a laser-cut front panel, or CNC-routed plywood. It costs more per unit and nothing up front, which is the right shape of risk for a first run. The enclosure routes.
5 · Things that do not exist at quantity one
- A test jig. You cannot plug in a thousand boards by hand. A pogo-pin bed that powers the board, flashes it, exercises the reader with a known card, plays a tone into a microphone, and reads the battery divider — a day of work and it defines your yield.
- A tracking label. Not optional: CPSIA requires permanent marking with manufacturer, location and batch. That is a design requirement, not paperwork — it needs a place on the moulding or the label artwork.
- Serialisation. Each box wants an identity for support, and each SD card an image. Both are production-line steps.
- Returns and support. At a 3% return rate that is thirty boxes coming back, and someone reads the emails.
- Product liability insurance. A lithium cell in a toy for a four-year-old is the sharp end of the consumer-goods risk pool. Get a quote before committing; it can change the arithmetic on its own.
6 · Where it lands
Against a Toniebox at $129.99 and a Yoto at $109.99, a $99 retail price is comfortable on paper. Through distribution, at a 2× wholesale multiple, it is thin.
7 · The problem, and it is not cost
The thing that makes tape good is that a parent records the story themselves. That is the whole point, and it means there is nothing to sell afterwards.
A Toniebox is $129.99 and every figure after it is $19.99. Over two years a modest collection runs $610–$850, and almost all of that is content. The hardware is not the business — it is customer acquisition for the business, and the margin lives in a catalogue we deliberately do not have.
So a thousand tapes is a hardware company with one transaction per customer, competing against two companies for whom the hardware is a loss leader. That is a real and specific disadvantage, and it is worth being clear-eyed about before a tool is cut. It also has a shape of answer — sell the thing whose margin you are structurally good at, which here is cards and enclosures and possibly nothing else, or do not sell it at all and publish the design so that a thousand people build one.
Which is what the repository is already for.
8 · What to do differently now, at quantity one
Very little, which is the useful finding. Three things, all cheap today and expensive to retrofit:
- Keep the ESP32-S3 as a module, forever. §3. It is already the plan; the point is not to talk yourself out of it later for $2 of BOM.
- Put test points on everything — the rails, I²C, I²S, the battery divider. On protoboard they are a pad and a label. They are what a pogo-pin jig lands on, and adding them to a laid-out board later means a respin.
- Never let the audio format become the box’s problem. The manifest names the file and ESP-ADF plays what arrives. That is what keeps a firmware update from orphaning somebody’s recordings, and at a thousand units in the field that stops being a nicety.