Drag to rotate · Feature orbs dim when capped or locked on your selected tier
You're under the free cap. No paywall hits you.
Why cap an on-device feature?
"On-device" rarely means the whole pipeline runs locally. Wake-word detection and capture happen on the glasses, but heavy AI — translation, scene understanding, long conversations — usually round-trips to cloud GPUs.
Cloud inference has a real marginal cost. A rough industry estimate: $0.01–$0.05 per minute of streamed multimodal AI. At 2 hours/day, that's $18–$90 per user per month — more than the hardware margin.
The playbook
- Sell hardware near cost to build an install base.
- Ship generous free AI while usage is subsidized.
- Introduce rate limits for the heaviest users first.
- Launch a subscription framed as "power user" tier.
- Move more features behind the tier over time.
The same arc played out with cloud storage, connected cars, and home cameras.
Buyer's checklist
- Which features run fully on-device? Those are hardest to paywall later.
- Is there a published rate limit, or vague "fair use" language?
- Does the maker have a subscription already? Expect features to migrate into it.
- What happens if servers shut down — does the device still work offline?
- Total 3-year cost: hardware + subscription, not sticker price.
Worked example: true 3-year cost
| Scenario | Hardware | Subscription | 3-year total |
|---|---|---|---|
| Light user (free tier is enough) | $379 | $0 | $379 |
| Daily AI user (paid tier, $20/mo) | $379 | $720 | $1,099 |
| Heavy user + accessory ecosystem | $379 + $150 | $720 | $1,249 |
The subscription can quietly cost 2–3× the glasses themselves. That's the real story behind every "why is my feature rate-limited?" headline: the device is the entry ticket, the AI is the product.