Qualcomm CEO Cristiano Amon says artificial intelligence companies are already pressing for smartphones capable of running 100-billion-parameter models continuously by 2028 — a request that would more than triple the on-device capacity of the chips Qualcomm has just begun shipping.

Amon made the remarks in an interview on October 8, 2026, describing a shift toward always-on “agentic” AI: software that acts on a user's behalf rather than waiting for a tap on an app.

What AI Companies Are Asking For

According to Amon, the models AI firms want would operate with personal context. They would understand what a user intends, reason through the request, and plan actions on their own.

Parameters — the learned values inside a model that determine how it processes information — are the standard yardstick for model size, and larger counts generally demand more memory and compute from the hardware that runs them. That is why the number carries weight in a phone, where both resources are tightly constrained.

The scale of the request becomes clearer when measured against what Qualcomm just shipped. At its Snapdragon Summit on September 22, 6, the company unveiled the Snapdragon 8 Elite Gen 6 and Extreme Gen 6 platforms. Those chips, built on 2nm technology, are designed to run mixture-of-experts models of up to 30 billion parameters directly on the device — an architecture that activates only the portions of a model needed for each task, which is what allows large networks to operate within a phone's limits. Moving from 30 billion to 100 billion parameters represents more than a threefold jump — inside the same rough physical footprint of a phone.

How Qualcomm Plans to Fit Large Models Into Small Devices

Qualcomm is leaning on several techniques to make the math work. One involves loading model parameters from flash storage, so the full model does not need to sit in active memory all at once. The new platforms also expand the memory available to the Neural Processing Unit, the part of the chip dedicated to AI workloads. The company additionally pointed to intelligent flash management and a low-power memory architecture as ways to support larger models.

Complementing these efforts are always-on sensing hubs, which run far smaller models — approximately 200 million parameters — continuously in the background. Qualcomm says its new chips can process approximately 1 million tokens per day, tokens being the chunks of text a model reads and writes.

Why the Smartphone Still Matters

Amon has been consistent on one point: the smartphone is not going anywhere. He argued that phones will remain the central hub for personal AI, since they hold the context about a person's life that makes an assistant useful.

He also tied the AI push directly to hardware demand, linking anticipated growth in smartphone sales to these capabilities and noting that rising memory prices have held users back from upgrading.

Qualcomm is not alone in eyeing this timeline. OpenAI is reportedly also exploring hardware aimed at similar AI capabilities by 2028, though firm details have not been disclosed.

What It Means for the Industry

The case for on-device AI rests on three advantages over cloud-dependent systems: privacy, lower latency, and better power efficiency. Data that never leaves a phone cannot leak from someone else's server, and answers do not have to make a round trip across the internet.

The risks are real, however. Reaching 100-billion-parameter models on a phone is a multi-year engineering challenge, and the 2028 target reflects what AI companies are asking for — not a product anyone has promised to deliver.

Memory pricing is the most obvious pressure point. If costs stay elevated, the hardware required for these models could push phone prices higher at exactly the moment the industry hopes to lure buyers back.

Battery life is the other open question hanging over the word “continuously.” A model that runs all day must do so without draining the phone by lunchtime, which is why Qualcomm keeps emphasizing low-power architecture and small background models.

The markers to watch are concrete: how quickly on-device model sizes climb beyond the 30 billion parameters Qualcomm supports today, whether memory prices ease, and whether OpenAI or other AI firms put real hardware details on the table before 2028.