Edge AI Is Gaining on the Cloud

Chipmakers and device manufacturers used CES to argue that running AI locally cuts costs, reduces latency and addresses privacy constraints at a time when cloud-based inference is becoming expensive and operationally complex. Inference drives day-to-day AI usage, and where that computation happens shapes the economics of AI adoption across consumer and enterprise markets.

CES underscored that the era of “cloud-only” AI is giving way to a hybrid model where inference increasingly happens near the user or machine. Hyperscale training and model development will remain tethered to large data centers, but most interactive AI use cases, from personal assistants to robotics, will increasingly run locally.

Chipmakers Place Edge AI at the Forefront

Intel showcased how leading silicon companies are pivoting toward edge and on-device AI. The company used the show to launch its Intel Core Ultra Series 3 processors, the first AI PC platform built on its advanced Intel 18A process, designed to boost performance, graphics, efficiency and integrated AI compute across a spectrum of devices. The chips are already powering over 200 PC designs and are certified for industrial and embedded applications, including robotics, smart cities and automation, expanding local inference capabilities beyond traditional desktops and laptops.

Qualcomm also broadened the race for distributed intelligence with a suite of announcements highlighting its next-generation Snapdragon families and edge AI platforms. At CES, Qualcomm showcased the Snapdragon X2 Plus and Dragonwing IQ10 robotics stack, aiming to power next-generation wearables, personal AI devices, robots, vehicles and connected ecosystems that run AI locally without constant reliance on cloud back ends. The company’s emphasis on AI everywhere encompasses scaled intelligence from edge to cloud, signaling that on-device inference will be an integral part of its roadmap throughout 2026.

AMD used CES to expand its AI everywhere narrative, debuting the Ryzen AI 400 series and embedded processors that deliver up to 60 trillion AI operations per second of local neural performance and are targeted at laptops, desktops and edge applications, signaling that chip-level AI is becoming mainstream rather than experimental.

Arm’s trend overview for CES highlighted how physical and edge AI are extending into vehicles, robots, connected homes and industrial systems, suggesting a broader industry recognition that local compute is no longer niche but central to AI’s evolution.

Devices and PCs Drive Local Intelligence

Consumer device makers reinforced the point with announcements featuring on-device AI capabilities.

Samsung’s Galaxy devices and AI companion experiences emphasized localized personalization and convenience without constant cloud dependency, as seen in Samsung’s CES vision of “AI companions” across products.

Lenovo paired on-device AI with hybrid cloud services, unveiling Qira, a cross-device AI system designed to work seamlessly across PCs, tablets, wearables and phones, while also expanding its cloud infrastructure partnerships with Nvidia to accelerate enterprise AI deployments, as reported by Reuters.

Wearables and robotics also embraced local inference. XGIMI’s MemoMind AI glasses demonstrated multimodal inference for real-time translation and contextual assistance, and Boston Dynamics’ Atlas humanoid robot integrated local perception and decision logic that reduces reliance on remote compute.

The economics of inference are driving this evolution. Cloud-based inference introduces latency and recurring bandwidth costs that scale with usage, complicating pricing models for developers and users, CIO reported. Local inference reduces round-trip delays, lowers operating expenses and limits sensitive data from leaving the device, a combination that is compelling for enterprise and consumer segments.

This shift is already underway in financial infrastructure.

Hybrid AI architectures are emerging as the pragmatic default. Centralized cloud systems handle model training, coordination and heavyweight workloads, while on-device and edge compute handle real-time decisioning and interaction.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.