# MediaTek Launches 2nm Dimensity 9600 Pro with 61% Lower Power Consumption

Flagship chipset introduces native AI architecture and major efficiency gains for next-gen smartphones.

- **Source:** [Unite.AI](https://www.unite.ai)
- **Published:** 2026-09-15
- **Original article:** <https://www.unite.ai/mediatek-unveils-2nm-dimensity-9600-pro-with-61-lower-multi-core-power/>
- **On Brifing:** https://brifing.si/s/4723a9b7-7fc5-47b1-8f94-e0aec810ef73

## Breakthrough in chip miniaturization

MediaTek's Dimensity 9600 Pro marks the industry's first 2nm smartphone SoC, achieving a 61% reduction in multi-core power consumption. The chip uses TSMC's N2P process with nanosheet transistors, following their 2025 partnership announcement. Compared to N3E, the new process offers 18% better performance at same power or 36% lower power at same speed. This leap enables more efficient AI processing in mobile devices.

## Revolutionary CPU architecture

The 9600 Pro features a 2+3+3 All Big Core design with C2-Ultra cores clocked at 4.55GHz and C2-Pro cores at 4.35GHz/3.1GHz. Benchmarks show 17% better single-core and 15% improved multi-core performance versus previous generation. MediaTek's testing demonstrated these gains through lab evaluations on demo devices. The architecture specifically targets AI workloads while maintaining thermal efficiency for sustained performance.

## AI-optimized memory system

A 34.5MB cache supports the chip's AI Computing Fusion Architecture, complemented by industry-first LPDDR6 memory and UFS 5.0 storage. The Dimensity Scheduling Engine dynamically allocates compute, power and memory resources between hardware and software. This system allows smartphones to maintain multiple background applications while delivering responsive AI features without excessive battery drain.

## Dual-NPU for on-device AI

MediaTek's second-gen Super Efficient NPU cuts always-on AI power by 40%, while the NPU 1090 boosts LLM performance by 51% in prefill and 55% in token generation per watt. The Agentic AI Engine supports models up to 30B parameters, enabling digital twin capabilities. These advancements allow for real-time analysis of device data and personalized responses that learn user behavior patterns.

## Graphics and imaging leaps

The G2-Ultra NX GPU delivers 27% higher peak performance with 24% lower power consumption. New Dimensity Neural Graphics combines GPU and AI computing for enhanced gaming up to 185fps. For photography, the Imagiq 1290 ISP enables 4K240 slow-motion video and 17EV dynamic range. AI Imaging Fusion merges ISP and NPU capabilities for computational photography previously requiring desktop hardware.

## Accessible flagship performance

The companion Dimensity 9600M brings Pro-level AI and efficiency to more devices, retaining the Agentic AI Engine and scheduling system. Its NPU maintains always-on perception for real-time imaging enhancements while the ISP supports professional tracking features. HyperEngine ensures similar 185fps gaming performance as the Pro variant, democratizing high-end mobile experiences.

## Industry positioning

MediaTek emphasizes its position as the world's top mobile chipset provider, being first to announce 2nm technology. The company highlights how rapidly evolving user demands require simultaneous improvements in performance, efficiency and intelligence. Their Native AI architecture directly addresses these needs through hardware-software co-design rather than just transistor scaling.

## Market implications

Volume production begins late 2026, with first devices expected that quarter. The 9600 series could redefine premium smartphone capabilities, particularly for on-device AI applications that typically require cloud connectivity. MediaTek's advancements may pressure competitors to accelerate their 2nm roadmaps while forcing app developers to leverage new local AI features.

## Sources covering this story

- https://www.unite.ai/mediatek-unveils-2nm-dimensity-9600-pro-with-61-lower-multi-core-power/ — 

---

_Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._
