Technology

Mobile computing and on-device AI

Our chosen course technology is mobile technology and computing. The trend redefining it right now is on-device AI — and it changes what a budget phone has to be.

The technology

Mobile technology and computing is the hardware, software, and networking stack that makes a modern smartphone possible. For Talky, this isn’t just a technology we use — it is the product. Our entire business sits on the trajectory of mobile computing: the chips we can buy, the software those chips can run, and the networks our customers live on.

The trend: AI moves onto the phone

The single most consequential recent development in mobile computing is the migration of artificial intelligence workloads from the cloud onto the phone itself. A new generation of Neural Processing Units (NPUs) is now embedded in mobile system-on-chip packages. Qualcomm’s Hexagon NPU in the Snapdragon 8 Gen 5 delivers up to 46% faster on-device AI performance than its predecessor. MediaTek’s ninth-generation NPU in the Dimensity 9500 doubles compute for generative workloads and cuts peak power draw by more than half.

Crucially, these capabilities are no longer confined to flagships. Mid-tier silicon from MediaTek and Qualcomm is pushing the same NPU architecture into the $200–$400 band. Gartner forecasts that generative AI smartphone spending will reach roughly $393 billion in 2026, with more than 500 million AI-capable phones shipping worldwide.

Two recent articles frame the debate. Deloitte argues that on-device AI could make smartphones meaningfully more exciting because it removes the latency, data cost, and privacy concerns of cloud inference (Lee et al., 2025). Light Reading argues the industry is over-marketing AI to people who just want a phone that works (Morris, 2025). Both are right, and both matter for Talky.

How on-device AI helps Talky

  • Camera differentiation. AI-driven computational photography can make a $249 phone with modest optics produce images competitive with a $499 phone’s hardware — a direct hit on the biggest weakness of the budget segment.
  • Offline translation and captioning. Running locally, these features work without cellular data. Enormously valuable in emerging markets where data is expensive.
  • Smarter warranty support. An on-device diagnostic assistant can run standardized hardware self-tests the moment a customer opens a claim, summarize the likely fault, and send structured results to a support agent — shortening our remote-troubleshooting step and reducing the share of cases that need a costly physical return.
  • Lower operating cost. Cloud AI is metered per query. On-device inference is paid for once, at manufacture. For a small brand, that difference scales powerfully as our install base grows.

Where the trend pushes back

  • BOM pressure. NPUs and the high-bandwidth memory they need are exactly the components whose prices are rising fastest right now. Adding meaningful AI pushes our bill of materials in the wrong direction.
  • Buyer indifference. Morris (2025) notes that many consumers ignore or disable the AI features that already ship on their phones. If we subsidize silicon our customers don’t care about, we’ve paid twice: once to Qualcomm, once in margin we didn’t have to give up.
  • Expectations set by flagships. Reviewers will compare anything we ship to the AI experience on a $1,200 Apple or Samsung. We have to pick narrow, concrete use cases where our silicon budget can produce a genuinely good experience, not an imitation.

How Talky will actually use it

Our approach is selective adoption. We ship on-device AI where it reinforces our core promises — reliability, long support, repairability, and offline usefulness — and we skip it where it would only add cost and marketing noise. Concretely, the first Talky AI features will be the diagnostic assistant in the warranty app, offline translation for international shoppers, and an AI-assisted battery-health prediction model that tells owners when a cell is actually degrading rather than just guessing from raw percentages. Everything else waits until our component budget and our customers ask for it.