Qualcomm CEO says phones should run 100B-parameter AI models by 2028

By Splitfeed EditorialLast updated Oct 12, 2026, 06:10 UTC

Qualcomm CEO Cristiano Amon made the claim around October 10 to 11, 2026, stating that AI companies have expressed a desire for smartphones capable of running 100-billion-parameter models locally by 2028. Amon also predicted what he called a coming "battle of the agents" among AI firms competing at the device level. The remarks were widely circulated across financial and technology platforms in that window. Amon did not name the specific AI companies that had communicated the 100-billion-parameter target, and he did not detail the hardware roadmap Qualcomm would use to meet it.

The significance lies in the gap between where on-device AI currently stands and where Amon is pointing. Smartphones today can run models in the low-to-mid billions of parameters at most, and those already require dedicated neural processing hardware. A 100-billion-parameter model is roughly the scale of the largest publicly known frontier models, which currently require server-grade infrastructure and significant power budgets. If that workload genuinely moved to a handheld device by 2028, it would shift the competitive dynamics of AI deployment, reducing dependence on cloud inference and the data-center operators behind it. What is genuinely in dispute is whether the physics of battery life, heat dissipation and memory bandwidth can be solved at consumer price points within that timeline.

None of the three camps have published reactions to this story yet. The Pro-AI camp would typically treat a claim like this as evidence of rapid, democratizing progress, arguing that local inference protects privacy and reduces latency. The Anti-AI camp would be expected to question the energy and resource costs of building chips powerful enough to run such models in a pocket device, and might scrutinize whether the framing serves Qualcomm's commercial interests more than any broader public benefit. The Middle Ground camp would likely focus on the feasibility gap, welcoming the directional ambition while pressing for specifics on how quantization, hardware efficiency and model compression could realistically close the distance between current capability and the 2028 target.

The clearest near-term signal to watch would be Qualcomm's next Snapdragon platform announcement, where any disclosed on-device parameter benchmarks or memory architecture details would either lend credibility to Amon's timeline or expose it as aspirational marketing.