Qualcomm CEO: Smartphones Will Run 100-Billion-Parameter AI Models by 2028
nashnova research
Qualcomm's CEO revealed that leading AI companies have demanded phones capable of running 100-billion-parameter models near-continuously by 2028; Elon Musk responded that the real bottleneck is not chip power but RAM supply.
Who is demanding 100-billion-parameter models on a phone?
Qualcomm's CEO said in an October 10 interview that several frontier AI companies have told Qualcomm they need phones running at least 100 billion parameters by 2028.
The requirement goes further: the model must stay active nearly around the clock. This means → on-device AI is shifting from "wake on demand" to "always on" — a step-change in hardware requirements.
He also confirmed that some of these AI companies are developing their own phones — making them both Qualcomm customers and potential competitors.
How hard is it to run 100 billion parameters on a phone?
Today's flagship phones handle on-device models of roughly a few billion parameters. A 100-billion-parameter target is more than a tenfold jump in model size.
In plain terms = today's phone AI is a pocket-sized assistant; the 2028 goal is to fit a full-scale brain in your pocket — and keep it running all day.
This reflects a sharp escalation in AI companies' bets on on-device inference — running AI locally on the phone rather than on cloud servers — driven by privacy, latency, and offline availability.
Why does Musk say the bottleneck is memory?
Elon Musk responded on X on October 11, pointing out that the key constraint is RAM supply, not raw chip compute.
This means → running a 100-billion-parameter model requires loading vast quantities of parameters into memory. Current phones typically carry 12–24 GB of RAM; even a compressed 100-billion-parameter model far exceeds that capacity.
In plain terms = a fast enough chip does not guarantee the model can run — like an engine with plenty of horsepower but a fuel tank too small to go the distance. RAM is that fuel tank.
What does this mean for the supply chain?
If the 2028 target holds, phone memory capacity must multiply several times over within three years — a direct tailwind for memory chipmakers.
Qualcomm, as the dominant mobile chip supplier, faces a dual squeeze: it must meet AI companies' hardware demands while some of those same clients are building their own phones, potentially cutting Qualcomm out.
This reflects a deeper signal: the AI race is extending from the cloud to the pocket, and whoever solves the on-device hardware bottleneck first controls the next entry point.
市场有风险，内容仅供研究参考，不构成投资建议。