Five developer takeaways from Arm Create for building and deploying AI
Arm’s 2026 Shanghai and Shenzhen events outlined five developer takeaways for building and deploying AI across cloud, edge, mobile, graphics, and physical platforms, emphasizing workload-specific decisions and system-wide optimization.
At Arm Create 2026 in Shanghai and Shenzhen, developers examined how latency, cost, privacy, power, and deployment targets shape AI application design across cloud, edge, mobile, graphics, and physical AI. Arm’s latest hardware and software updates—including Arm AGI CPU, Neoverse CSS N4, CSS for Mobile 2, Mali G2-Ultra NX, and the new AI Portal—were central to these discussions, providing expanded infrastructure options and tools for model discovery and deployment workflows.
Agentic applications require a controlled execution environment around the model to manage code generation, tool calls, actions, state, and resources efficiently. Tencent Cloud’s Cube Sandbox at Arm Create Shanghai illustrated these challenges, highlighting the need for isolation, concurrency, security, and resource management to scale applications while keeping development focused on the application layer.
Model selection should prioritize workload requirements over size alone, with deployment location and behavior shaping the choice. A panel featuring Alibaba Qwen, ModelBest, Tencent Hunyuan, and Ultralytics emphasized matching models to specific workloads and environments, while the AI Portal helps developers identify validated and optimized models for Arm-based platforms to streamline evaluation and implementation.
Optimizing AI performance extends beyond the model to runtime, data movement, preprocessing, application logic, and underlying compute. Arm’s Performix and Performance Studio tools, alongside SME2 and neural graphics tooling, support profiling and improvement across cloud, mobile, and graphics workloads, enabling developers to address constraints like response time, power, cost, and sustained performance.