Global GPU Computing News【20260214】

Domestic GPUs Accelerate Integration into Medical AI: On February 7th, a solution for applying domestic full-stack high-performance GPUs in the medical AI field was launched in Shanghai. Incubated by a pilot base led by Zhongshan Hospital affiliated with Fudan University, the solution aims to lower the barriers to medical AI adoption. MXMACA Technology Co., Ltd. signed a cooperation agreement with the Shanghai Health Data Industry Association at the event, offering diversified support such as computing vouchers.

Domestic Computing Power Sees New Products and Expansion: Domestic chips continue to make breakthroughs. Ascend plans to release the Ascend 950PR in the first quarter of 2026, incorporating self-developed HBM to respond to export restrictions. Concurrently, the core node of the National Supercomputing Internet, based on Sugon's scaleX 640, has launched a 30,000-card AI computing cluster, demonstrating the rapid expansion of domestic computing power capacity.

Tech Giants Increase Investment in Self-Development and Capex: Domestic cloud vendors are actively preparing for the AI market. Alibaba Cloud has set a target to capture 80% of the incremental growth in China's AI cloud market by 2026, with its semiconductor arm T-Head also planning an independent IPO. ByteDance's self-developed chip project aims to mass-produce 100,000 chips by 2026. Baidu's Kunlun Core has also submitted its listing application.

AI Demand Drives Up Inference Computing Prices: As AI application deployment exceeds expectations, the surge in inference demand has led to tight computing resources. Industry reports indicate that domestic model providers and cloud service providers have begun raising prices. This trend benefits industry chain companies and also signals that domestic large models are accelerating their substitution for overseas models in 2026.

Memory Shortage Impacts Nvidia's Gaming GPUs: Constrained by demand from AI infrastructure and Micron's exit from certain consumer market segments, the supply of memory chips remains tight. Nvidia has postponed the launch of its next-generation gaming graphics cards and plans to cut production of its GeForce RTX 50 series, prioritizing limited memory inventory for AI data center hardware.

Samsung First to Mass-Produce HBM4 for Nvidia: Samsung Electronics will begin the world's first mass production of 6th generation High Bandwidth Memory (HBM4) in the third week of February, delivering it to Nvidia for its next-generation AI accelerator platform, "Vera Rubin". This marks the market's transition from 5th generation HBM3E to HBM4.

HBM4 Supply Landscape Shifts: In Nvidia's next-generation "Vera Rubin" AI system, SK hynix will be the primary supplier of HBM4, holding an approximately 70% share. Samsung will receive the remaining 30%, while Micron Technology will not be part of the HBM4 supply chain for this system, instead supplying LPDDR5X memory.

Tech Giants Pledge $600 Billion for AI Infrastructure: Google, Amazon, Meta, and Microsoft plan to collectively invest approximately $610 billion in 2026 for data centers and AI infrastructure construction, a scale representing about 70% growth compared to 2025. Despite this, Google's CEO indicated that computing power supply remains constrained, and the market is adopting a wait-and-see approach regarding whether these massive investments will yield quick returns.

Nvidia Internally Boosts Developer Efficiency with AI: Nvidia has deployed generative AI tools for use by approximately 30,000 of its internal engineers. By partnering with Anysphere to customize the Cursor integrated development environment, engineers' code output is reportedly three times higher than previous development processes.

Film and Rendering Gets Dedicated GPU Cluster: QumulusAI is launching clusters based on NVIDIA RTX Pro 6000 Blackwell server editions. The goal is to provide film and creative studios with studio-grade AI rendering, advanced simulation, and generative production workflow support, addressing the growing demand for computing power in the creative industry.