Why now · Three timed forces

The 18-month window when architecture decides who wins.

Apple Silicon hit cloud parity in 2025. EU AI Act Article 50 transparency lands 2 Aug 2026. Incumbents announced the war on the same six-month timeline. The category is contested at the surface; the on-device prosumer-priced layer is empty.

Apple Silicon hit cloud parity in 2025.: Apple shipped a free 3B-parameter on-device LLM to third-party developers at WWDC 2025. SpeechAnalyzer transcribes 2× faster than Whisper Large V3 Turbo on Mac. NVIDIA Parakeet TDT v3 broke 7% word error rate on the Hugging Face Open ASR Leaderboard. The hardware curve caught up. The cloud-only meeting-AI premise that defined the 2018-2024 era no longer holds. Sussur runs the speech-to-text + the LLM round-trip on the user's Mac.
EU AI Act Article 50 transparency obligations land 2 Aug 2026.: EU AI Act Article 50 transparency obligations + Annex III high-risk classification land in 51 days. Average data breach cost reached $4.88M in 2024 (IBM record high). 84% of GenAI users report concerns about data exposure (Cisco 2024). After 2 Aug 2026 every cloud-AI sales tool becomes a procurement red flag in regulated verticals. On-device is the only architecture that survives the question “where does my audio go?” with one word. Nowhere.
Incumbents announced the war.: Gong launched Mission Andromeda (their AI sales coaching platform) on 25 Feb 2026. Clari + Salesloft completed their merger on 3 Dec 2025. Granola raised $125M Series C at a $1.5B valuation on 25 Mar 2026. Cluely raised $15M Series A from a16z in June 2025. tldv pivoted to sales coaching in 2025. The category is contested at the surface, but every entrant still ships cloud STT + cloud LLM round-trip. The architecture decides who wins the regulated-vertical sell-in, and the 18-month window for an on-device prosumer-priced product is open.

Why now · Three timed forces

The 18-month window when architecture decides who wins.

Apple Silicon hit cloud parity in 2025.: Apple shipped a free 3B-parameter on-device LLM to third-party developers at WWDC 2025. SpeechAnalyzer transcribes 2× faster than Whisper Large V3 Turbo on Mac. NVIDIA Parakeet TDT v3 broke 7% word error rate on the Hugging Face Open ASR Leaderboard. The hardware curve caught up. The cloud-only meeting-AI premise that defined the 2018-2024 era no longer holds. Sussur runs the speech-to-text + the LLM round-trip on the user's Mac.
EU AI Act Article 50 transparency obligations land 2 Aug 2026.: EU AI Act Article 50 transparency obligations + Annex III high-risk classification land in 51 days. Average data breach cost reached $4.88M in 2024 (IBM record high). 84% of GenAI users report concerns about data exposure (Cisco 2024). After 2 Aug 2026 every cloud-AI sales tool becomes a procurement red flag in regulated verticals. On-device is the only architecture that survives the question “where does my audio go?” with one word. Nowhere.
Incumbents announced the war.: Gong launched Mission Andromeda (their AI sales coaching platform) on 25 Feb 2026. Clari + Salesloft completed their merger on 3 Dec 2025. Granola raised $125M Series C at a $1.5B valuation on 25 Mar 2026. Cluely raised $15M Series A from a16z in June 2025. tldv pivoted to sales coaching in 2025. The category is contested at the surface, but every entrant still ships cloud STT + cloud LLM round-trip. The architecture decides who wins the regulated-vertical sell-in, and the 18-month window for an on-device prosumer-priced product is open.