Qualcomm’s CEO says AI companies want phones running 100-billion-parameter models by 2028
THE SO WHAT
Vendors asking for 100B-parameter models on-device by 2028 means handset silicon, memory, and power budgets are being re-architected around local inference. If your product assumes cloud-only AI, start planning for a world where the premium UX is offline, private, and hardware-gated.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIDebt Worries Keep Driving Down CoreWeave Stock
When a core AI infra provider trades like a distressed credit, every long-dated GPU and capacity contract tied to it becomes counterparty risk, not just cost. If you’re relying on non-hyperscaler GPU lessors, treat their balance sheets and debt covenants as part of your infra due diligence this quarter.
Applied AIA week with RTX Spark: everything we’ve learned using Nvidia’s new laptops
If RTX Spark-class Arm laptops deliver desktop-grade AI locally, the default assumption of “AI = cloud call” starts to erode for client-heavy workflows. CIOs should start segmenting use cases by what can move on-device over the next 24–36 months to reduce latency, cost, and data exposure.
Applied AIMaybe Chatbots Shouldn’t Get to Say ‘I’ or ‘Me’
If mainstream voices like Sherry Turkle are arguing that first-person pronouns in chatbots are harmful, UX and compliance teams will eventually be pulled into that debate. Treat anthropomorphism—“I,” “me,” emotions—as a policy choice with regulatory and brand risk, not just a copywriting flourish.
Applied AISchwab and JPMorgan are blocking AI agents from moving customer money
Top brokers drawing a hard line—AI agents can see balances but not move funds—sets a de facto standard for autonomous finance. If you’re building agentic flows in regulated domains, assume read-only first and design explicit human-in-the-loop checkpoints for any transaction authority.