{"slug": "five-developer-takeaways-from-arm-create-for-building-and-deploying-ai", "title": "Five developer takeaways from Arm Create for building and deploying AI", "summary": "Arm unveiled new hardware and software for AI development at its Arm Create 2026 events in Shanghai and Shenzhen, including the Arm AGI CPU and Neoverse CSS N4 for cloud infrastructure, Arm CSS for Mobile 2 for on-device AI, Arm Mali G2-Ultra NX for mobile graphics, and the Arm AI Portal for discovering models validated and optimized for Arm-based platforms. The company framed the developer challenge around four questions — which intelligence fits the task, where the work belongs, how the pieces work together as a system, and whether the application performs as intended on its real target — and distilled the sessions into five takeaways, including that agentic applications need a controlled execution environment around the model and that model choice should start with the workload rather than model size alone.", "body_md": "# Five developer takeaways from Arm Create for building and deploying AI\n\nA capable model is only the starting point. Building an AI application means deciding what sits around that model as latency, cost, privacy, power, and deployment targets begin shaping the design.\n\nAt [Arm Create](https://developer.arm.com/arm-create) 2026 in Shanghai and Shenzhen, developers explored those decisions across cloud, edge, mobile, graphics, and physical AI through keynotes, partner discussions, technical deep-dives, and hands-on labs.  \n\nArm’s latest hardware and software announcements were part of those discussions. [Arm AGI CPU and Neoverse CSS N4](https://newsroom.arm.com/news/arm-agi-cpu-neoverse-css-n4-agentic-ai) expand cloud infrastructure options, [Arm CSS for Mobile 2](https://newsroom.arm.com/blog/arm-css-for-mobile-2-and-c2-cpu-cluster) supports new mobile on-device AI experiences, and [Arm Mali G2-Ultra NX](https://newsroom.arm.com/blog/arm-mali-g2-ultra-nx-ai-native-mobile-graphics) brings neural capabilities to mobile. Meanwhile, the new [Arm AI Portal](https://newsroom.arm.com/news/arm-unveils-arm-ai-portal) helps developers discover models validated and optimized for Arm-based platforms, understand the target fit, and find code and deployment workflows. \n\nHowever, the useful question for developers is not simply what each platform or tool does, but what they help them make decisions, experiment and test, and improve their software.\n\nThe [Arm Create Shanghai keynote](https://youtu.be/6W_yIr8PrwM) framed the challenge into four questions:  \n\n- Which intelligence fits the task;\n- Where the work belongs;\n- How the pieces work together as a system; and\n- Whether the application performs as intended on its real target.\n\nThe sessions and labs turned those questions into five key developer takeaways.\n\n## 1. Decide what infrastructure an agentic application needs\n\nWhen an application becomes agentic, the model is only one part of the challenge. Agents can generate code, call tools, and take actions dynamically, placing new demands on the infrastructure underneath them.\n\nAt Arm Create Shanghai, Tencent Cloud’s Cube Sandbox provided a practical example of this challenge in action, with discussion around isolation, concurrency, security, state, resource management, cost, and which parts of the platform should handle each workload so developers can stay focused on the application.\n\nThe broader lesson for developers is that agentic applications need a controlled execution environment around the model, so actions, tools, state, and resources can be managed consistently as the application scales.\n\n## 2. Choose the model that fits the workload\n\nModel choice starts with the workload, not model size alone. A quick-reply feature in a mobile app, a cloud agent reasoning across large document sets, and an edge vision workload can push model choice in very different directions.\n\nThe Arm Create Shenzhen panel with Alibaba Qwen, ModelBest, Tencent Hunyuan, and Ultralytics brought that into focus. The discussion moved beyond model comparison toward application fit: what the workload needs, where it should run, and how it behaves once deployed. The principle is straightforward. Fit the model to the work, then put the work in the right place.\n\nAI Portal helps developers discover models validated and optimized for Arm-based platforms, understand the target fit, and find code and deployment information that can shorten the path from evaluation to implementation.\n\n## 3. Decide how to measure and optimize the complete application\n\nWhen an AI experience is slow or inefficient, the model is only one place to look. Time and resources can also be spent in the runtime, data movement, preprocessing, application logic, or underlying compute. A voice assistant can lose time in audio preprocessing, while a mobile AI feature can hit memory or thermal limits during sustained use.\n\nThat makes system-wide profiling and optimization part of the development loop. Arm tools and technologies support different stages of that process: [Arm Performix](https://developer.arm.com/servers-and-cloud-computing/arm-performix) and [Arm Performance Studio](https://developer.arm.com/tools-and-software/arm-performance-studio) can help developers understand where workloads spend time and resources, while [Arm Scalable Matrix Extension 2 (SME2)](https://www.arm.com/technologies/sme2) and [neural graphics tooling](https://developer.arm.com/mobile-graphics-and-gaming/neural-technology) provide different ways to improve performance on Arm-based platforms. The useful optimization starts with evidence from the workload and the constraint that matters, whether that is response time, power, cost, or sustained performance. \n\nFor cloud workloads, that may mean finding CPU or memory hotspots and then improving how the workload use the available compute. For mobile AI, it can mean balancing responsiveness and power and for graphics, it can mean preserving image quality, frame pacing, and thermal headroom.\n\n## 4. Decide where each workload should run\n\nNot every part of an AI application belongs in the same place. Developers must decide which workload should stay on device, which benefits from nearby edge infrastructure, and which requires scale and shared services in the cloud.\n\nCompute capability shows where a workload can run; product requirements help determine where it should run. Latency, privacy, offline operation, local control, cost, and available compute all shape that choice, and a single application can split work across several locations. The goal is not to choose one location for the whole application, but to place each part where its requirements are best met.\n\nThe Arm compute platform spans those deployment choices across cloud, edge, mobile, and graphics. AI Portal can help connect model selection to specific Arm-based targets and deployment paths.\n\n## 5. Keep the workflow portable, then validate on the real target\n\nAI applications rarely stay in one environment. A team might prototype locally, test on cloud infrastructure, move part of a workload onto a device, and then optimize for a mobile, edge, graphics, or physical AI target. If the model, runtime, tooling, and performance workflow change completely each time, that movement creates more work.\n\nAt the Arm Create Shanghai fireside, Eric Ernst from Aston Martin Formula One™ Team and Irving Chen from Unitree used Formula One cars and humanoid robotics as two examples of the same engineering challenge. In both cases, software, hardware, data, sensing, and control must work together in physical environments where real world conditions cannot be fully predicted.\n\nThis makes portability and validation connected problems. Every movement between cloud, edge, mobile, graphics, and physical AI adds friction if the model, runtime, tooling, and performance workflow keeps changing. Developers benefit from tools, libraries, models, and workflows that can move with the application, but the final proof still has to come from the environment where it will run. The physical AI labs at the Arm Create events extended that idea through simulation-to-robot workflows, deployment, trust, telemetry, and device discovery.\n\nSME2, KleidiAI, Performix, [Arm MCP Server](https://developer.arm.com/servers-and-cloud-computing/arm-mcp-server), neural graphics tooling, [Arm Device Connect](https://learn.arm.com/learning-paths/embedded-and-microcontrollers/device-connect-d2d/overview/), and Arm AI Portal each support a different part of that workflow, from finding models and optimizing workloads to measuring performance and connecting intelligent software with devices. \n\nThe practical takeaway is to think about portability early, keep as much of the workflow consistent as deployment choices change**,** and validate the complete system on the target that matters. \n\n## Build for what’s next\n\nShowcasing practical approaches to AI development is the point of Arm Create. Across the Shanghai and Shenzhen events, developers moved from keynote ideas into technical deep-dives, live demos, hands-on labs, and direct conversations with Arm engineers and ecosystem partners. That hands-on experience also created a path forward for developers, with 97% of Arm Create Shanghai survey respondents saying the event improved their understanding of how to build, optimize and deploy AI applications on Arm.\n\nAs AI applications span more targets, the goal is not to avoid platform-specific optimization, but to preserve as much continuity as possible in the code, tools, skills, and performance workflow while deployment choices change. Developers need a workflow that can travel with the application rather than starting over each time the target changes. A common Arm architecture and software foundation help developers carry more of that work forward across different deployment environments, including cloud, edge, mobile, graphics, and physical AI.\n\nThe next Arm Create event moves to the US, where developers can continue working through these same practical questions with Arm experts and ecosystem partners, and get hands-on with the tools and workflows behind AI applications from cloud to edge.\n\n#### Join an upcoming Arm Create near you.\n\nArm Create gives developers access to hands-on projects, live events, and challenges.\n\nAny re-use permitted for informational and non-commercial or personal use only.", "url": "https://wpnews.pro/news/five-developer-takeaways-from-arm-create-for-building-and-deploying-ai", "canonical_source": "https://newsroom.arm.com/blog/takeaways-from-arm-create-china-2026", "published_at": "2026-09-11 10:00:03+00:00", "updated_at": "2026-09-11 10:04:28.565562+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-chips", "ai-products", "ai-agents", "ai-tools"], "entities": ["Arm", "Arm Create", "Arm AGI CPU", "Neoverse CSS N4", "Arm CSS for Mobile 2", "Arm Mali G2-Ultra NX", "Arm AI Portal", "Tencent Cloud"], "alternates": {"html": "https://wpnews.pro/news/five-developer-takeaways-from-arm-create-for-building-and-deploying-ai", "markdown": "https://wpnews.pro/news/five-developer-takeaways-from-arm-create-for-building-and-deploying-ai.md", "text": "https://wpnews.pro/news/five-developer-takeaways-from-arm-create-for-building-and-deploying-ai.txt", "jsonld": "https://wpnews.pro/news/five-developer-takeaways-from-arm-create-for-building-and-deploying-ai.jsonld"}}