MediaTek

MediaTek is a fabless semiconductor designer that builds system-on-chip (SoC) products for smartphones, tablets, PCs, connected devices, automotive systems and data-centre infrastructure. It is the world's largest supplier of mobile chipsets by volume, shipping more than two billion connected devices' worth of chips annually, and it designs its own AI processors (NPUs) into most of them.^[1]

History and Background

MediaTek was founded in 1997 in Hsinchu, Taiwan, initially focused on optical-storage controller chips, and expanded into feature-phone chipsets and then smartphones, where its Dimensity line — introduced from 2019 onward — made it the default alternative to Qualcomm in the premium and mid-range Android market. The company established MediaTek Research, a dedicated AI research group, and has progressively moved from shipping connectivity silicon to shipping silicon whose headline features are AI: on-device generative models, agentic assistants and multimodal imaging.^[1]

Recent milestones mark the shift. The Dimensity 9500 arrived in September 2025 with a focus on on-device AI and power efficiency; the Dimensity 9600 Pro, announced in September 2026, became the first smartphone SoC announced on TSMC's 2nm process, paired with a "Native AI" architecture and an Agentic AI Engine designed to run assistant workloads on the handset rather than in the cloud.^[2] MediaTek pairs hardware with day-zero support for open models such as Alibaba's Qwen on mobile and automotive platforms, and in November 2026 the company hired a 37-year Intel veteran in advanced packaging to strengthen a data-centre strategy built around a dual CoWoS/EMIB approach — a clear signal that it intends to contest AI compute beyond the phone.^[5]

Key Concepts and Technology

A MediaTek NPU is built around a hardware multicore scheduler, a dedicated DMA engine that performs deep-layer fusion and data compression to reduce DRAM bandwidth demand, power-intelligence and shared-memory awareness, and a company-designed Network on Chip (NoC) for low-latency communication between cores.^[1] These design choices address the real constraint of edge AI: not peak throughput but energy per token — the model must run inside a phone's thermal and battery budget.

Three architectural ideas follow. On-device inference keeps prompts, photos and messages local, reducing latency and cloud cost; heterogeneous scheduling splits a workload between NPU, GPU and CPU depending on precision; and agentic workloads — multi-step tasks that call tools — are increasingly offloaded or split, with the device handling intent and the cloud handling long reasoning chains. Because MediaTek ships across phones, vehicles, smart home, retail and industrial IoT, its silicon is one of the cheapest routes to deploying the same model class at very different scales.

Applications and Impact

MediaTek chips sit in most non-Apple Android devices sold worldwide, which makes them the practical delivery mechanism for AI features to consumers in emerging markets: translation, call summaries, photo and video enhancement, and offline assistants. Its automotive arm (Dimensity Auto cockpits) and IoT platforms extend the same model to vehicles and industrial equipment, while its connectivity portfolio (5G, Wi-Fi 7/8) supplies the transport layer for cloud AI.

>See Also

• NVIDIA
• Edge AI
• Neural Processing Unit
• National Semiconductor Strategy
• Penang Semiconductor and AI Ecosystem