On-device AI on smartphones: How Arm brings generative AI to billions of mobile users
At Arm, we believe AI – and the benefits it provides – should be accessible to as many people as possible. This belief is driving the next chapter of AI, one which is personal, immediately available for users, and increasingly on device.
What is on-device AI on smartphones?
On-device AI allows smartphones to run AI workloads locally on the device instead of relying entirely on cloud processing. This can improve responsiveness, privacy, offline availability, and power efficiency for experiences such as real-time camera effects, speech recognition, translation, smart summaries, image search, multimodal assistants, and generative AI. Arm enables these capabilities through Armv9 CPUs, optimized AI software such as KleidiAI, and ecosystem partnerships across the mobile industry.
With Arm technology featuring in 99 percent of all smartphones worldwide, our role in this ongoing “democratization of AI” is critical. Through balancing performance, power and area, our technology powers the mobile platforms that enable advanced intelligence to run efficiently everywhere – not just today’s powerful flagship smartphones but also more affordable mid-range and premium devices.
Delivering on-device AI at this scale means billions of people worldwide experience tangible improvements across the mobile applications they rely on every day, like real-time image and video enhancement, speech recognition, multimodal assistants, translation, smart summaries and image search.
How Arm and TECNO enable real-time generative AI on smartphones
A clear example of AI democratization in action is Arm’s recent collaboration with TECNO, an OEM focused on delivering advanced mobile technology to users in emerging markets. Arm and TECNO are working together to bring on-device generative AI that powers advanced creativity experiences, like creative style transfer preview and immersive stylization, to its premium and mid-range smartphones.
Generative AI has traditionally faced barriers on mobile due to its high compute intensity, long latency, and reliance on cloud connectivity. However, the Arm TECNO partnership helps to solve these challenges by running generative AI on Armv9 CPUs and then optimizing through Arm KleidiAI to significantly enhance AI capabilities for TECNO’s Edge-Side AI Generated Content (AIGC) Preview Concept Technology, which eliminates flickering and jittering during split-screen viewfinder camera experiences. This integration delivers real-time AI artistic effects in the style transfer preview at 30 frames per second directly in the viewfinder.
At the core of these improvements is Integer 8-bit Matrix Multiply (i8MM), a hardware instruction extension embedded in Arm CPUs that accelerates General Matrix Multiply (GEMM), the core compute behind most AI workloads. This is especially beneficial for mid-tier and entry-level mobile devices where the CPU is often the primary AI engine. By leveraging i8MM instructions through KleidiAI, the Arm TECNO collaboration accelerates core machine learning (ML) workloads across common frameworks and applications, leading to 30% faster and more efficient AI compute for more responsive creative experiences, such as real-time photo effects, object detection, and augmented reality (AR) features.
The ongoing Arm TECNO collaboration has also led to several key engineering innovations including:
- Intelligent heterogeneous scheduling, which allocates workloads across CPUs to maximize performance and efficiency.
- Model compression and restoration techniques, which allows generative models to run efficiently on Arm CPUs, while preserving artistic quality.
- Real-time fluidity via CPU pipelines, with AI inference migrating to the Arm CPU to achieve video-level preview for creative style transfer, minimizing latency and enabling a responsive viewfinder experience.
As a result of this work, users can preview ideas instantly and experience immersive stylization as they compose.
Why Armv9 CPUs matter for on-device AI performance
A key enabler of powerful AI capabilities extending to mid-tier and premium smartphones is the growing availability of Armv9 CPUs. This is delivering the AI performance, efficiency, and ecosystem readiness required for democratizing AI capabilities across every tier of the smartphone market, from flagship all the way to entry-level.
On Armv9 CPU-based devices, these benefits show up in the following ways:
- Faster, more efficient AI performance that developers can seamlessly access through KleidiAI;
- Better image processing in camera and computer vision (CV) post-processing pipelines; and
- Faster application launches and improved responsiveness.
Armv9 CPUs are integrated across the widely adopted Arm-powered MediaTek Dimensity platforms, which power millions of mobile devices worldwide. The most recent Dimensity 8500 series push advanced AI even deeper across the mid-range and premium series of smartphones, supporting on-device generative AI workloads like style transfer effects, local language models and creative AI features that were previously reserved for flagship hardware.
How Arm KleidiAI helps developers optimize mobile AI apps
However, hardware alone isn’t enough to democratize AI – it also depends on how easily developers can access and use that performance. This is where KleidiAI plays a critical role.
KleidiAI is a suite of optimized AI libraries and tools designed to help developers automatically unlock the performance of Arm CPUs, without requiring any code changes or deep, hardware-specific expertise. With KleidiAI integrated across major AI frameworks, OEMs and a broad developer community have access to a mature, proven ecosystem that bridges the gap between cutting-edge CPU features, like Arm Scalable Matrix Extension 2 (SME2), i8MM and Scalable Vector Extension 2 (SVE2), and real-world applications, with performance benefits that they can access immediately. As a result, many leading global tier-1 applications across social media, shopping, payments, camera, AR, and multimodal, are already benefiting from Armv9 CPU’s AI capabilities.
How Arm enables on-device AI on smartphones
Arm enables on-device AI through four layers:
- Armv9 CPUs provide the compute foundation for AI workloads across mobile device tiers.
- i8MM, SVE2, and SME2 accelerate key AI operations such as matrix multiplication and vector processing.
- KleidiAI exposes these CPU capabilities through optimized AI libraries and frameworks.
- Ecosystem partnerships with OEMs, silicon vendors, and developers help bring these optimizations into real applications and devices.
How Arm scales on-device AI from flagship to mid-range smartphones
Democratizing AI isn’t about a single feature or device. It’s about building a compute foundation that allows intelligence to scale responsibly, efficiently, and universally. That means the AI experiences users care about most – camera pipelines, voice and speech features, and even emerging AIGC workflows – can scale efficiently from premium devices to the broad mid-range and beyond.
By enabling advanced AI capabilities on billions of smartphones through advanced architecture features, software optimizations, deep ecosystem collaboration with partners like MediaTek and TECNO, and decades of engineering innovation, Arm is helping ensure the next era of AI is not exclusive, but shared.
Intelligence everywhere, powered by Arm.
FAQs
What is on-device AI on smartphones?
On-device AI runs AI workloads locally on the smartphone, helping improve latency, responsiveness, privacy, and offline availability for features such as camera enhancement, speech recognition, translation, summaries, image search, and generative AI.
How does Arm enable on-device AI?
Arm enables on-device AI through Armv9 CPUs, AI-focused instruction extensions such as i8MM, SVE2, and SME2, optimized software such as KleidiAI, and ecosystem partnerships with OEMs, silicon providers, and app developers.
What is Arm KleidiAI?
Arm KleidiAI is a suite of optimized AI libraries and tools that helps developers access Arm CPU performance through common AI frameworks without requiring deep hardware-specific optimization.
Why is on-device AI important for mid-range smartphones?
On-device AI on mid-range smartphones helps bring advanced features such as real-time camera effects, translation, summaries, object detection, and creative AI tools to more users without limiting these experiences to flagship devices.
How can generative AI run on smartphones?
Generative AI can run on smartphones through efficient model design, compression, optimized CPU instructions, heterogeneous scheduling, and software libraries that accelerate AI workloads on local device hardware.
What mobile AI features can run on-device?
Examples include real-time image and video enhancement, speech recognition, translation, smart summaries, image search, multimodal assistants, object detection, AR effects, and creative style transfer.
99% of all smartphones powered by Arm
Arm powers the world’s smartphones, from entry-level to flagship, that are used by billions of people everyday.
Any re-use permitted for informational and non-commercial or personal use only.
Editorial Contact
Latest on X
Something new is coming soon.⌛
Every journey starts with a question.
This is just the beginning.
Stay tuned as we reveal what we've been building. 👀
What a week in Taipei.
From keynote moments and packed demos, to conversations with partners, developers and customers, #COMPUTEX2026 reminded us what makes this industry so exciting: the people building it together.
Thank you to everyone who watched our keynote, enjoyed the
Reachy isn't just being friendly at #COMPUTEX2026. 🤖
Powered by NVIDIA DGX Spark and Arm-based technology, it can see, listen and respond in real time using AI running locally at the edge.
This is AI everywhere—in action.
AI infrastructure today is all about the next generation of AI-ready data centers, calling for:
1️⃣ Efficiency
2️⃣ Performance
3️⃣ Scale
At #Computex2026, @Lenovo showcased the Arm AGI CPU-based Lenovo HR650a V3 Server, demonstrating how Arm-based platforms are expanding
What does it take to bring Physical AI from the lab into the real world?
At @GlobalSemi Tech Summit, we explored the convergence of AI, simulation and edge compute with leaders from @GlobalFoundries and @awscloud, and how it's enabling the next generation of robotics and
📣 82% of internet traffic is video.
That means the challenge isn’t just delivering every stream, it’s doing it without
❌Buffering
❌Dropped frames
❌Ballooning power demands
The Arm AGI CPU helps deliver more streams, more AI, and smoother video experiences at scale.
Not every day you get a robot explaining Arm AGI CPU. 🤖
At #COMPUTEX2026, the Aeolus aeo robot—a multi-talented robot built to help boost productivity and everyday convenience—showed how AI can understand, reason and interact in the real world.
Exciting to see how our
