Free tool · AI device cost calculator
Answer five plain questions and we’ll match your idea to one of four AI chip levels. Then compare two years of cloud AI bills with the one-off cost of running the AI on the device itself. No email needed to see your result.
Answer the questions on the left. Your recommended level lights up here as you go.
Your AI level, your running-cost comparison and the next steps on one page. Useful for your investor deck or your engineer.
Why this matters
Most AI device budgets go wrong in one of three places. The calculator is built around them.
Cloud AI is billed for every minute, for every user, every month. A demo feels free; a thousand users does not.
On-device AI moves the cost into the hardware: a stronger chip and more memory, paid once in every unit you build.
Each level up uses more power. A wake-word chip runs for months; a robot brain needs a big pack or a wall plug.
Assumptions
These are the chips we design with at each level, with typical prices for small production runs. Your quote uses live pricing for your exact parts.
| Level | What it does | Power | Typical price | Chip we’d typically use |
|---|---|---|---|---|
| No AI needed | A sensor or simple rule does the job | Microamps to milliwatts | $0 | A sensor and a simple microcontroller |
| 1 · Reflex chip | Wake word, gesture, unusual vibration | Microwatts to a few milliwatts | $3 | ESP32-S3 module |
| 2 · Vision microcontroller | Sees a person, pet or object a few times a second | Hundreds of milliwatts | $11 | STM32N6 (camera not included) |
| 3 · Smart camera | Real-time tracking, reading text, body pose | About 2 to 5 watts | $100 | Hailo-class AI add-on board |
| 4 · Robot brain | Several cameras, maps, talks and moves | 15 to 130 watts | $399 | NVIDIA Jetson Orin Nano 8GB module |
| Cloud voice AI | Speech in, answer out, billed per minute | — | $0.16 / min | A cloud voice AI service, billed per minute |
Not included: the rest of the parts list, the camera and memory, development, certification, app and cloud hosting, connectivity plans, and retail markup. Chip prices are for the chip or module only. This is a planning estimate, not a quote.
Distributor list prices and vendor announcements, checked November 2026: ESP32-S3 module and STM32N6 (Future Electronics); Jetson module pricing (CNX Software, 22 Jul 2026); cloud voice AI pricing (Ringly, 2026).
FAQ
It depends on how much thinking the device has to do. The chip ranges from about $3 per unit for a wake-word or gesture chip, to about $100 for an AI add-on board, to about $399 for a robot-class module. Development, the rest of the parts and certification come on top. The calculator matches your device to a level in five questions.
The cloud costs nothing up front but bills every minute, for every user, every month. On-device AI costs once per unit. At 10 minutes a day at $0.16 a minute, the cloud costs about $48 per user per month, so a small on-device chip pays for itself within weeks. At low usage the cloud can stay cheaper. Often a hybrid works best.
Match the chip to the job: Level 1, a reflex chip, for wake words and gestures; Level 2, a vision microcontroller, to recognise people or objects a few times a second; Level 3, a smart camera, for real-time video; Level 4, a robot brain, for devices that map, talk and move. Pick the lowest level that does the job: it costs less and the battery lasts longer.
A reflex chip can run for months on a small battery. A vision microcontroller usually needs a rechargeable battery. A smart camera needs a rechargeable pack or wall power, and a robot brain needs a large battery or wall power.
Yes. You see your level and your running costs without giving an email. You only enter an email if you want the PDF report.
Bring your result to a 30-minute call. We’ll tell you honestly whether your idea needs AI, which level fits, and what the next step costs.