Global GPU Computing News【20251012】

Microsoft Azure Deploys the Industry's First GB300 Super Cluster
Microsoft Azure announced the launch of the world's first production-grade NVIDIA GB300 NVL72 supercomputing cluster. This cluster comprises 4,608 GPUs and will be dedicated to running inference for OpenAI's most complex AI models.

OpenAI and AMD Form Strategic Partnership
OpenAI plans to procure tens of billions of dollars worth of GPU computing power from AMD and may acquire up to 10% of AMD's shares at a symbolic price. This deep collaboration aims to break the reliance on a single vendor (NVIDIA) and reshape the AI computing landscape.

Intel Discloses AI Chip Development Roadmap
Intel announced that it will update its AI GPU product portfolio annually and plans to launch new products focused on AI inference tasks and enterprise-grade workloads, enhancing its competitiveness in the data center AI chip market.

NVIDIA Blackwell Platform Sets New AI Inference Performance Records
In the latest Inference MAX v1 benchmark tests, NVIDIA's Blackwell platform demonstrated comprehensive leadership in both performance and energy efficiency. Its AI inference cost can be as low as $0.02 per million tokens.

US Approves Large-Scale NVIDIA AI GPU Exports to UAE
The US government has approved NVIDIA's export of tens of billions of dollars worth of AI GPUs to the United Arab Emirates, with an annual export quota increased to 500,000 accelerator cards. This move is seen as a significant part of US "AI diplomacy."

First Domestic Computing Power Adaptation Center in Greater Bay Area Begins Operation
The Domestic Computing Power Adaptation Center located in Zhuhai, part of the Greater Bay Area, was officially inaugurated. This center integrates domestic computing power supply, application validation, and ecosystem cultivation, with companies like Huawei and MetaX acting as the first batch of partners.

Domestic Computing Power Servers Account for Over 90% in Recent Bids
According to a CITIC Securities report, domestic computing power suppliers won over 90% of the shares in recent server tenders totaling around ten billion yuan combined from Industrial and Commercial Bank of China (ICBC) and China Unicom, indicating an acceleration in the autonomous and controllable procurement process.

Domestic Large Models Achieve Seamless Adaptation with Domestic Computing Power
Alibaba's newly released large model achieved "Day-0 support" on Huawei's Ascend chips, while Tencent's visual model ranked third globally on a relevant leaderboard. This marks the continued maturation of the domestic AI computing ecosystem.

Hon Hai Becomes Microsoft's GB300 Server Supplier
According to industry sources, Hon Hai (Foxconn) is the main supplier for the NVIDIA GB300 server racks, commencing volume production from September, which is expected to gradually contribute to this quarter's performance.

Domestic Computing Power Chips Realize Ecosystem Closure
CITIC Securities pointed out that domestic large models are accelerating their iteration, and domestic computing power chips can already achieve seamless adaptation, forming an ecosystem closed-loop from hardware to software. They express optimism about this driving the continuous development of domestic AI.