Global GPU Computing News【20251108】

Sugon Launches Super Node: Sugon unveiled the "scaleX640" super node, the world's first cabinet-level system integrating 640 accelerator cards. Designed for training trillion-parameter models, it delivers 630 PFlops of computing power and uses immersion phase-change liquid cooling to achieve a remarkably low PUE of 1.04.

Huawei & Alibaba Release High-Performance Solutions: Huawei launched its "Ascend 384" super node, integrating 384 Ascend NPUs. Alibaba Cloud introduced the "Panjiu AL128" super node, supporting 128 accelerator cards per cabinet with single-phase immersion liquid cooling, reducing data center PUE to 1.09.

"South Lake Computing Framework" Awarded: The "South Lake Computing Framework," developed by Zhejiang Lab, received an award at the World Internet Conference. It tackles the challenge of coordinating heterogeneous accelerator cards from different vendors, supporting mixed-precision training across various domestic accelerators and improving model training efficiency by over 30%.

Policy Supports Domestic Chips: New government procurement policies in China have clarified the definition of "domestic products" and offer a 20% price evaluation preference. This aims to promote the prioritized procurement of domestic hardware in critical infrastructure like data centers.

Jingye Intelligent & Moore Threads Collaborate: Jingye Intelligent and domestic GPU company Moore Threads announced a strategic partnership. They will jointly develop products in areas including embodied intelligence modules, AI servers, and AI computing centers.

Amazon & OpenAI Strategic Partnership: Amazon Web Services (AWS) and OpenAI entered a major strategic cooperation deal valued at $38 billion. OpenAI will utilize computing resources from hundreds of thousands of advanced NVIDIA GPUs provided by AWS, marking OpenAI's first such partnership with a cloud provider other than Microsoft.

Microsoft Signs Major Compute Orders: Microsoft signed multi-year compute capacity agreements this week with AI cloud providers IREN and Lambda. The combined value of these deals exceeds $10 billion, securing AI computing power based on NVIDIA's GB300 series GPUs.

Nebius AI Cloud Expands to UK: The Nebius Group launched its AI cloud services in the UK. The new facility utilizes NVIDIA's latest Blackwell Ultra GPUs and Quantum-X800 InfiniBand networking to support generative AI and large language model development.

OpenAI Infrastructure Investment Plan Revealed: Reports indicate OpenAI recently signed infrastructure contracts totaling approximately $1.4 trillion with several companies, aimed at jointly developing up to 30 gigawatts (GW) of computational resources.

NVIDIA Chips Approved for Export to UAE: Microsoft announced it has received U.S. government approval to export NVIDIA chips to the United Arab Emirates (UAE). As part of this plan, Microsoft's cumulative investment in the UAE is expected to reach $15.2 billion by 2029.