Hands-free visual AI through smart glasses and AR headsets. Enterprise field workflows lead consumer adoption — task-adjacent interfaces that disappear into the work.
Key constraints
Battery life under continuous camera, display brightness outdoors, social acceptance of face-worn cameras, unit cost at scale.
Watch for
On-device SLM benchmarks, enterprise deployment case studies, multimodal UI announcements, prescription-channel distribution.
AI models running directly on phones, tablets, and mobile-class silicon. The largest deployment surface for on-device intelligence.
Key constraints
App store policy gates for AI features, thermal/power budget on battery, fragmentation across Android OEMs.
Watch for
Apple Intelligence feature rollout, Android AICore updates, cross-platform model format convergence.
Local AI compute appliances — Jetson, Coral, Raspberry Pi AI, and industrial edge nodes. The physical backbone for private, low-latency inference.
Key constraints
Thermal envelope for sustained inference, software stack fragmentation, procurement friction vs cloud APIs.
Watch for
New silicon announcements (NVIDIA Jetson, Hailo, Intel), developer ecosystem growth, industrial gateway adoption.
Clinical-grade wearables for patient monitoring, diagnostics, and treatment adherence. The highest-stakes portable AI category.
Key constraints
FDA/CE/MDR regulatory pathway (years), clinical validation costs, hospital IT integration complexity.
Watch for
FDA 510(k) clearances, clinical trial publications, EHR integration partnerships, reimbursement code changes.
Wrist-worn AI expanding from health tracking into on-device assistants, gesture input, and personal context engines.
Key constraints
Screen size limits complex UI, battery constraints for always-on AI, health data privacy regulation.
Watch for
On-device LLM deployments (sub-1B params), new sensor integrations, watchOS/Wear OS AI SDK changes.
AI that works in the background across devices — smart home hubs, presence-aware rooms, and multi-device orchestration without explicit commands.
Key constraints
Multi-vendor interoperability, always-listening privacy tension, user mental model for invisible AI.
Watch for
Matter/Thread AI extensions, multi-device context sharing protocols, presence-sensing hardware announcements.
Consumer devices running personal AI locally — from desktop companions to hobby robots. Privacy-forward, offline-capable, personality-driven.
Key constraints
Hardware BOM cost for adequate compute, user expectation gap vs cloud LLMs, retention beyond novelty phase.
Watch for
Open-source companion OS projects, offline SLM quality thresholds, crowdfunded hardware launches.
Items that don't map cleanly to a single device category but still carry portable-AI relevance.
Finger-worn sensors for passive health tracking and subtle input. Minimalist wearables that prioritize continuous sensing over displays.
Key constraints
Extremely constrained power and compute budget, limited output modalities, niche use-case clarity vs watches.
Watch for
New sensor modalities (glucose, hydration), gesture vocabulary research, sleep/stress clinical validation.
Smartphones as the on-device AI inference platform. Silicon advances in NPUs bring local LLMs, real-time vision, and privacy-preserving assistants.
Key constraints
Thermal throttling under sustained inference, RAM ceiling for larger models, OS fragmentation for AI APIs.
Watch for
NPU TOPS announcements (Qualcomm/MediaTek/Apple), on-device model format standards, OS-level AI frameworks.
Ruggedized headsets, smart helmets, and body-worn devices for manufacturing, logistics, and field service. Workflow-first, durability required.
Key constraints
Safety certification timelines, integration with existing MES/ERP, worker acceptance and training overhead.
Watch for
ATEX/ANSI certifications, warehouse/logistics pilot results, union and workforce agreement signals.
Sub-100M parameter models running on microcontrollers and ultra-low-power silicon. The long tail of embedded AI beyond smartphones.
Key constraints
KB-level memory budgets, single-digit milliwatt power envelopes, limited tooling for non-ML engineers.
Watch for
TinyML benchmark results, MCU vendor AI SDK releases, academic distillation papers, Arduino/PlatformIO AI libs.