China is leveraging its three major innovation centers to drive industrial upgrading and foster new quality productive forces by tapping their respective strengths, said the guests on the latest episode of China Economic Roundtable, an all-media talk show hosted by Xinhua News Agency.

MediaTek says the memory-saving features in its Dimensity 9600 Pro are designed to ease pressure from rising chip and memory costs. The company says these features let phones run larger AI models without additional RAM. According to TechNews, the chip supports mixture-of-experts models with up to 30 billion parameters, keeping roughly 3 billion in memory and loading others as needed.

MediaTek just launched two flagship phone chips at once. That is a first for it. The Dimensity 9600 Pro is built on TSMC’s 2nm process. The Dimensity 9600M uses TSMC 3nm. Phones with both are due soon. $TSM The Pro is the top chip. It uses an all-big-core design: two ultra cores at 4.55GHz and six more big cores beneath them. Single-core speed is up 17%. Multi-core is up 15%. At peak multi-core load it uses 61% less power than the last generation. On-device AI is the pitch. A second-gen low-power brain uses 40% less energy. A tenth-gen high-power brain makes 55% more tokens per watt. The chip is meant to notice and act, not just sit in an app. The 9600M is the volume flagship. Same idea, lower cost. Better camera processing and a lighter AI agent for more phones. MediaTek says its Dimensity line has been the world’s top phone chip for six years. Two chips at one event is how it tries to keep that. 2nm for the tip. 3nm for the crowd.

MEDIATEK UNVEILS ITS FIRST 2NM SMARTPHONE CHIP MediaTek launched the Dimensity 9600 Pro, its first mobile processor built on TSMC's 2-nanometer process, alongside the 3-nanometer Dimensity 9600M for the broader flagship tier, per Reuters. The first phones using both chips launch soon. The 9600 Pro's neural processing unit handles more generative-AI work on the device and processes a user's prompt 51% faster than the previous generation, the company says. The push is aimed at Qualcomm $QCOM in premium handsets, where MediaTek's market value already passed Qualcomm's earlier this year. MediaTek's first AI accelerator for a major US cloud provider enters mass production in the fourth quarter. Last month it raised $3.9B in a convertible sale in which $NVDA put in $3.5B and $GOOGL participated. August revenue rose 44% y/y.

MediaTek’s latest announcement on the deepening of its long-term partnership with Nvidia repeatedly uses the term “rack-scale.” This is the key to understanding the announcement, and it directly validates the central conclusion of my research from more than two months ago: MediaTek has upgraded the strategic positioning of its AI business from "IC / ASIC design" to "system-level design." The announcement link: The announcement also explains how MediaTek plans to put this new positioning into practice. By working with Nvidia, MediaTek can bring custom chips and XPUs into Nvidia’s mature rack-scale ecosystem, so that once chip development is complete, they can be integrated more quickly into rack-scale systems for deployment in data centers and AI factories. For MediaTek, Nvidia’s mature rack-scale systems ecosystem means it does not need to start from scratch, giving this new strategy a much clearer path to execution. For Nvidia, MediaTek’s custom silicon design capabilities and customer base can help extend Nvidia’s rack-scale ecosystem further into the custom silicon market. This also explains why “rack-scale” is the key term in the announcement: it is where the two companies need each other most. MediaTek can move from chips to systems, while Nvidia can extend its rack-scale ecosystem into the custom silicon market, creating a win-win that opens up new growth opportunities for both.

MediaTek's Rick Tsai on their biggest bottleneck: "Supply chain does present challenges. Not just HBM but go down to substrate, etc. Being in Taiwan, it's almost like the center of AI infrastructure ecosystem really helps." There is not enough substrate at all... Via @BloombergTV

MediaTek: “400G or 448G SerDes IP is really going well. We certainly expect that to be ready sometime in the next year, maybe second half of next year. Very interesting comment.