Why it looks good
Genuine always-on, sub-watt, latency-bound inference in a wearable form factor. A clean technical fit for neuromorphic / analog-in-memory / SSM-ASIC designs, and a real clinical need driven by Ageing Population.
Why it fails the venture test (small / slow / concentrated / commoditised)
- Small at the layer that matters. Device market ~$9.7–10.5B (2025/26), ~22.7M units/yr, but silicon content is a few $/unit → chip-layer TAM low hundreds of $M, of which a startup wins a slice. The $10B headline is the wrong denominator.
- Slow. ~6–9% device CAGR, and DSPs are specced to last years — a multi-year qualification/design-win cadence per account. Revenue arrives slowly even on a win.
- Concentrated & partly captive. Big Five (Sonova, WS Audiology, Demant, GN, Starkey) ≈ 84% US / 90%+ global, vertically integrated into clinics and benefit managers. Buyers mostly make their own silicon (Sonova proprietary DSP/ASIC) or buy the incumbent merchant SoC (ON Semiconductor Ezairo, already a heterogeneous CPU+DSP+accelerator part). A new entrant sells into ~5 accounts, several closed.
- Commoditised from above. FDA-cleared AirPods Pro 2 as an OTC hearing aid (Sept 2024), $250 vs $1,000+, 100M+ units — value migrates to Apple’s own silicon in the CE tier you can’t sell into.
Sources
- (size, growth, concentration, Apple/OTC, ONsemi/Sonova silicon)