Intelligence at the edge

For a decade, phones and computers improved incrementally and replacement cycles lengthened. On-device AI is the first capability in years that genuinely requires new silicon — and that makes it the catalyst for an upgrade cycle across an enormous installed base.

Historical context: consumer electronics grew through distinct capability waves — the smartphone, the tablet, wearables — each followed by a plateau where devices got faster but not meaningfully more useful. Plateaus stretch replacement cycles and compress the whole supply chain.

The structural driver is that useful on-device AI needs dedicated neural processing and much more memory than older devices contain. Users who want assistants that work offline, privately, and instantly must buy new hardware; software updates cannot deliver it.

The technology layer spans neural processing units in phones and laptops, memory capacity increases, model compression techniques that fit capable models on devices, thermal management in thin enclosures, and the operating-system layers that route work between device and cloud.

Adoption economics favour the chain broadly: higher silicon and memory content per device raises component revenue even if unit volumes only recover modestly, and premium mix shifts upward as AI features cluster in higher-priced models.

The beneficiaries include device makers, mobile and PC processor designers, memory manufacturers benefiting from higher content per unit, camera and sensor suppliers, and the contract manufacturers assembling it all.

The value chain runs from semiconductors and components through assembly to brands and retail. Component content per device is the metric that matters — it rises even when unit shipments are flat.

The overlooked layer includes specialty component suppliers, thermal-management and materials firms, testing and assembly equipment makers, and the repair and refurbishment businesses extending device lifecycles.

Competitive dynamics are dominated by a few platform owners, with the interesting competition happening in silicon — where AI capability per watt has become the specification that sells devices.

Risks: consumer demand is discretionary and price-sensitive; upgrade cycles have disappointed before when new features failed to justify replacement; tariffs and supply-chain disruption hit hardware hard; and cloud AI may prove good enough for most users, undercutting the on-device premise.

What to watch: memory and silicon content per device, premium mix shifts, replacement-cycle length, and whether AI-capable device shipments grow as a share of total. The research follows content per unit rather than headline unit shipments.