Huawei Releases Flex:AI Container Technology
Huawei has introduced an innovative software technology that can flexibly partition a single GPU or NPU computing card to form a shared computing power pool. It is reported that this can increase the utilization rate of AI computing resources from the industry average of 30-40% to 70%.
Moore Threads Initiates IPO Subscription on STAR Market
Domestic GPU company Moore Threads has officially launched its IPO subscription on the Sci-Tech Innovation Board (STAR Market). The company develops GPU products based on its self-developed MUSA architecture, aiming to provide computing acceleration for cutting-edge fields such as artificial intelligence and digital twins.
South Korea's Large-Scale Procurement of NVIDIA GPUs
The South Korean government, in collaboration with major corporations like Samsung and Hyundai, has established a task force to jointly plan and deploy 260,000 advanced GPUs procured from NVIDIA, aiming to enhance the country's overall artificial intelligence computing power.
First Batch of Hard Tech Theme Funds Approved
The first batch of Hard Tech and AI-themed ETFs, totaling 16 funds, were approved on the same day in China. This is expected to bring significant incremental funding into core sectors like domestic AI and GPU computing.
Hebei Plans AI Industry for the "15th Five-Year Plan"
Hebei Province has explicitly stated in its proposals for the "15th Five-Year Plan" that it will accelerate the cultivation and expansion of high-end and new-tech industries, including artificial intelligence, demonstrating local government's emphasis on AI industry development.
Google Officially Launches 7th Gen TPU Ironwood
Google has launched its seventh-generation TPU chip, Ironwood, claiming several-fold to ten-fold performance improvements compared to the previous generation. It also offers an integrated AI infrastructure solution combining chips, cloud services, and managed operations.
NVIDIA Responds to Google TPU Competition
In response to market attention on Google's TPU, NVIDIA issued a statement emphasizing that its GPUs still maintain a "one-generation lead" in absolute performance and versatility, and said it is the only platform that can run all AI models everywhere.
Meta Considers Procuring Google TPU Chips
Reports suggest that Meta is in discussions with Google to potentially deploy Google's TPU chips in its data centers starting in 2027. This potential deal, potentially worth billions of dollars, is seen as a signal to reduce reliance on a single supplier.
AMD Announces Collaboration Milestone
AMD announced that Zyphra's ZAYA1 model was successfully trained entirely on its MI300X GPUs and the ROCm software ecosystem. This is reported as the first large-scale Mixture of Experts (MoE) model trained solely on AMD hardware, marking progress in its high-performance computing area.
Beijing Plans Space-Based Computing Center
Relevant domestic institutions have unveiled plans to develop a large-scale space data center system, exploring the deployment of massive AI computing power in space environment. This aims to address resource constraints faced by ground-based data centers, such as energy consumption, land use, and cooling.