MediaTek's Rick Tsai announces 2nm chip production from September at Computex 2025

At Computex 2025, which will run through May 23, vice chairman Rick Tsai announced that MediaTek will debut its first 2-nanometre chip this September, a high-performance, low-power milestone developed by TSMC.

"We will be taking out our first 2 nanometers device in September this year. Of course, this is a high-volume chip, I'm not allowed to announce...but we are down this path for sure," said Tsai during his keynote address on Tuesday.

The upcoming 2nm chip is expected to deliver a 15% performance boost and 25% better power efficiency compared to its predecessor. The initial rollout will be a custom ASIC designed for high-volume applications, signalling MediaTek’s ambition beyond smartphones.

As AI-powered connectivity becomes central to next-generation networks and devices, MediaTek says it is prepared for the demands of an AI-native world. The new chip unlocks significant advances in machine learning and real-time AI inference, enabling use cases like autonomous driving, voice interaction, and instant language translation, all with longer battery life and reduced energy draw.

AI now sits at the heart of MediaTek’s roadmap. The company reports a 3.1x jump in CPU performance, a 7.4x improvement in GPU processing, and a staggering 29x leap in MPU capabilities—enhancements key to supporting on-device intelligence, immersive gaming, and efficient edge inference. During gameplay, MediaTek says its architecture can reduce system power consumption by as much as 65%.

Marking its most aggressive expansion yet, MediaTek is moving from mobile to massive-scale AI infrastructure.

A standout collaboration with NVIDIA further cements that ambition. Together, the companies co-developed a 20-core ARM CPU (10 Cortex-X925 + 10 A725) for NVIDIA’s next-gen DGX Spark AI supercomputer. With 1 petaflop of AI performance in a compact desktop form factor, the DGX Spark is expected to begin shipping later this year. It supports AI models up to 200 billion parameters and is powered by NVLink-C2C, a next-gen interconnect offering 5x the bandwidth of PCIe Gen5.

The partnership also includes the launch of NVLink Fusion, NVIDIA’s new high-speed, scalable computing fabric. Designed to support distributed AI workloads, NVLink Fusion connects CPUs, GPUs, and accelerators from a broad range of vendors, including Fujitsu, Qualcomm, Alchip, Astera Labs, Marvell, and MediaTek. The goal is to build semi-custom AI infrastructure systems that are modular, interoperable, and cloud-ready.

MediaTek’s chips now support over 240 AI models, including Gemini, LLaMA, and DeepSeek, running efficiently on edge devices like smartphones, tablets, and in-car systems.

Once known primarily for mobile chipsets, MediaTek is positioning itself as a full-spectrum technology player—from handsets and Chromebooks to cars, IoT, and global AI infrastructure.