AI devices & consumer electronics: A2P SMS and mobile data

Devices that talk back, connected out of the box.

In figures

  • A voice assistant is a cloud client: the device captures audio, sends it for recognition and streams the answer back, so the data plan is sized from round-trip audio and sync traffic, not from local storage.
  • Children's-data rules apply to companion devices; the GDPR's Article 8 sets 16 as the default age for consent to information-society services, with member states able to lower it to 13.
  • A wearable that claims a medical purpose follows medical-device rules (in the EU, the Medical Device Regulation), not consumer-electronics rules.
  • Consumer eSIM activation follows GSMA SGP.22; devices that must be managed without a phone app follow SGP.32, which adds an IoT profile assistant and a manager that can push profiles after sale.

This page is not legal advice.

Scenarios

AI toys and companion robots

An AI toy is a microphone, a small speaker and a radio around a cloud service. The wake word runs on the device; the utterance goes up as compressed audio, typically tens of kilobits per second, the answer returns as speech, and the parent app syncs settings and usage summaries over the same link. Size the plan from conversation minutes per day plus app traffic, and decide before launch whether audio is stored, for how long and where. Children's-data rules are scoped per market with your counsel, so devices sold into several markets may need separate data-handling defaults.

Translators

A translator does four things per turn: capture, recognize, translate and speak. On-device models handle short phrases offline; longer conversations go to a cloud service, and the user waits for capture, uplink, inference and downlink to add up. A two-way conversation at a few kilobytes per second fits a narrow data profile, so the risk is latency and open sessions rather than bandwidth. Devices that cross borders need profiles that stay local per market through multi-IMSI or eSIM orchestration, and offline phrase packs should arrive over Wi-Fi before the trip.

AI glasses

AI glasses sit between a phone accessory and a standalone device. The paired version borrows the phone's connection over Bluetooth and uploads photos and short clips through it; the embedded-eSIM version needs its own profile, power budget and thermal design, because a video uplink heats the frame. A still photo is a few megabytes, a minute of 1080p video at 15 Mbps is over a hundred megabytes, and assistant queries are tiny by comparison. Cap uploads to Wi-Fi by default and require an explicit action for cellular upload, or the data plan becomes the product's largest variable cost.

Data-only wearables

Data-only wearables send small, frequent payloads: step counts, heart rate, sleep summaries and location, usually a few hundred bytes per sync, batched and retried when the link drops. Duty cycle sets the design, not throughput. A device that syncs every few minutes and downloads a firmware image twice a year uses far less data than one streaming continuous heart rate. Data-only profiles keep the device out of voice and line-registration obligations; anything that adds calling or SMS changes that classification and has to be scoped separately per market.

Firmware and model updates

Over-the-air updates are the largest single download a consumer AI device makes, and the one most likely to fail in the field. Ship deltas rather than full images, make downloads resumable, verify the image before switching partitions and stage the rollout by cohort. A model update can run to hundreds of megabytes, so decide whether it travels over cellular at all or waits for Wi-Fi. Devices on metered plans need an update budget and a fallback that keeps the previous build working when an update fails.

Who buys this

  • Out-of-box connectivity tested before the retail launch, with the retail SKU fixed.
  • One profile family across launch markets, local where the device lives.
  • Update traffic budgeted separately from daily use.

Data reality

Indicative device profiles compiled from public research. Confirm the profile for your hardware at quote time.

ProfileMB/moRATSIM
AI toy or AI voice box 300–2000 Cat-1, Cat-1 bis SGP.32, multi-IMSI
Translator, AI recorder or AI glasses 50–2000 Cat-1, Cat-4 SGP.22, SGP.32
Companion or educational robot 300–2000 Cat-1, Wi-Fi SGP.32
Kids smartwatch (with calling) 300–3000 Cat-1, Cat-4 local number, managed platform

Device admission

  • Children's devices name their markets and data handling at quote time.
  • Devices with cameras or always-on microphones describe the capture and retention design.
  • Health claims name the regulatory route before a plan is proposed.

RAT

BearerNote
LTE Cat-1 Conversation and sync default; enough for compressed speech and photo upload.
LTE Cat-1 bis Single-antenna, lower-cost modules for toys and wearables.
5G where deployed Glasses and high-throughput companions; check the power cost first.

Product bundle

FAQ

Do children's devices need anything special?

Yes. Children's-data rules apply per market and are scoped with your counsel before launch; the regulation tracker records the references. Expect different default settings per market, and design the parent app so data retention is visible and changeable.

Can wearables make calls?

Voice-capable lines fall under line-registration rules; call features are scoped separately per market. A data-only profile is the simpler route and keeps the device out of those obligations, but it also rules out voice-first features.

How do translators roam?

On profiles that stay local per market through multi-IMSI or eSIM orchestration. Tell us the countries a device will visit and we scope the profile mix; devices that stay home for their whole life take a simpler single-market profile.

Which eSIM standard applies to a device with no app?

GSMA SGP.32 covers IoT devices that need remote profile management without a phone app; SGP.22 covers consumer devices activated by QR code or app. The choice changes the module and the manufacturing test.

How should we size the data plan?

From three numbers: conversation or video minutes per day, sync traffic per day and update size per year. Those three numbers set the monthly range, and a pilot on a sample of devices confirms it before the full build.

Request a quote