1. Amazon is making a significant move in AI hardware with Trainium2, its second-generation AI accelerator; 2. Trainium2 is designed to compete with Nvidia's H100 and Google's TPUv6e; 3. Amazon's Project Rainier is a 400,000-chip Trainium2 cluster for Anthropic, a leading AI research company.

1. AMD shares fell to a 1-year low due to DeepSeek's AI chat assistant news; 2. I doubled my AMD position due to strong demand for AMD's AI products in 2025 and AMD's gains in the AI accelerator market; 3. AMD's aggressive AI investments are paying off with significant Data Center revenue growth and a near-50% discount to its historical P/E ratio.

➀ AMD showcases its new Instinct MI325X accelerator at CES with 256 GB of HBM3E memory; ➁ The new accelerator features 19,456 stream processors and 6 TB/s of bandwidth; ➂ AMD's move to 256 GB of memory from the previously announced 288 GB is unexplained.

➀ Intel CEO Pat Gelsinger visits Elon Musk’s Memphis data center and praises the deployment of Xeon processors; ➁ xAI team is commended for building the AI head node in a short amount of time; ➂ Intel’s financial struggles and its attempt to regain market share with Xeon chips are discussed.

➀ Raspberry Pi's AI HAT+ integrates Hailo AI acceleration technology, offering 13 TOPS and 26 TOPS models; ➁ The HAT+ is compatible with PCIe Gen 3.0 for enhanced compute power; ➂ The 26 TOPS model supports complex inferencing tasks like object detection and pose estimation.

➀ Google has unveiled its AlphaChip AI-assisted chip design technology, which promises to speed up chip layout design and make it more optimal. ➁ The technology has been used to design Google's TPUs and has been adopted by MediaTek. ➂ AlphaChip uses reinforcement learning to optimize chip layouts, potentially revolutionizing the chip design process.

➀ Banana Pi BPI-CM5 Pro is a Rockchip RK3576 system-on-module compatible with Raspberry Pi CM4, offering 16GB LPDDR5 RAM and 128GB eMMC flash; ➁ It features a 6 TOPS AI accelerator, WiFi 6, Bluetooth 5.3, and two 100-pin connectors; ➂ ArmSoM-CM5 module is in final stages of production with expected availability by mid-October; ➃ CM5-IO carrier board supports additional interfaces like USB 3.0 and PCIe.

➀ Microsoft unveiled its first custom AI accelerator, Maia 100, designed for large-scale AI workloads on Azure, featuring a 5nm process and COWOS-S technology. ➁ The chip integrates large on-chip SRAM and four HBM2E chips, offering 1.8TB/s bandwidth and 64GB memory. ➂ Maia 100 supports up to 700W TDP, optimized for high performance and efficient power management. ➃ The accelerator includes advanced tensor units and vector processors, enhancing AI computation efficiency. ➄ A comprehensive SDK is provided for seamless integration with Azure OpenAI services, supporting PyTorch and other development tools.