NVIDIA unveils next-generation Blackwell architecture GPUs: NVIDIA officially launched the B200 GPU based on the Blackwell architecture. It boasts a significant performance leap, claiming up to 25x improvements in performance and energy efficiency for training trillion-parameter large language models compared to the previous generation, solidifying its leadership in AI training and inference.
AMD expands MI300X series supply to compete in AI market: AMD announced that shipments of its Instinct MI300X accelerators are increasing, with more orders from cloud service providers and large tech companies, aiming to capture more market share in AI computing from NVIDIA.
Microsoft Azure announces deployment of new in-house AI accelerators: Microsoft announced the large-scale deployment of its second-generation custom AI chip, Maia, in its Azure data centers. This aims to provide more efficient and cost-optimized computing power for AI services like Copilot, reducing reliance on third-party GPUs.
China's Biren launches new BR-series general-purpose GPUs: Chinese GPU company Biren Technology released its new generation of BR104/102 general-purpose GPU products, highlighting breakthroughs in AI training and inference performance. Samples have been sent to specific clients for testing, aiming to advance domestic high-performance computing.
Google Cloud TPU v5p becomes generally available: Google announced that its latest AI accelerator, Cloud TPU v5p, is now generally available, providing customers with powerful computing options for training large-scale cutting-edge AI models, competing with NVIDIA GPUs in the cloud service market.
OpenAI seeks partnerships to address global GPU shortage: Reports indicate that OpenAI's CEO Sam Altman is in talks with global investors and chip manufacturers to raise vast sums of money for building a global network of AI chip fabrication facilities, addressing the fundamental computing bottleneck hindering AI development.
US government considers tightening AI chip export controls on China: Reports suggest the Biden administration is urging allied nations to strengthen controls on AI technology exports to China and might further tighten licensing rules for high-end GPU exports (like NVIDIA's specially designed chips for China), aiming to slow down China's AI computing development.
Amazon AWS launches instances powered by NVIDIA Grace Hopper: Amazon Web Services (AWS) launched new EC2 instances featuring the NVIDIA GH200 Grace Hopper Superchip. This chip combines CPU and GPU tightly and is designed to accelerate giant AI models and massive data processing.
Moore Threads releases new MTT S4000 data center GPU: Chinese GPU maker Moore Threads released a new MTT S4000 GPU designed for AI computing and data center scenarios, supporting mainstream AI frameworks and models, boosting the construction of domestic AI computing infrastructure.
Soaring AI computing demand keeps TSMC's CoWoS advanced packaging capacity tight: Due to overwhelming orders from AI chip giants like NVIDIA and AMD, TSMC's CoWoS advanced packaging capacity remains a critical bottleneck. The company is actively expanding production, but supply shortages are expected to persist into 2025, constraining global GPU supply.