Global GPU Computing News【20250321】

NVIDIA Launches Blackwell Ultra GPU
At GTC 2025, NVIDIA unveiled its flagship Blackwell Ultra GPU, featuring 12-layer stacked HBM3e memory with 288GB VRAM. It delivers 15 PetaFLOPS at FP4 precision, a 2.5x performance leap over the Hopper architecture.

GPU Architecture Roadmap Revealed
NVIDIA disclosed its 4-year plan: Rubin architecture (2026), Rubin Ultra (2027), and Feynman (2028), with iterative HBM advancements (e.g., HBM4e).

Breakthrough in AI Inference Efficiency
The Blackwell Ultra NVL72 rack, optimized for inference, outperforms the GB200 by 1.5x. Tasks taking 1.5 minutes on Hopper now finish in 15 seconds. NVIDIA’s Dynamo OS boosts inference performance to 40x Hopper levels via dynamic task allocation.

Expanded Autonomous Driving Partnership with GM
NVIDIA introduced the full-stack safety system NVIDIA Halos and deepened AI collaboration with General Motors in production design and simulation.

6G and AI-Native Wireless Network Development
Partnering with Cisco and T-Mobile, NVIDIA is advancing 6G wireless tech using its AI Aerial platform for high spectral efficiency.

Silicon-Photonics Switch Breakthrough
The Quantum-X silicon-photonic co-packaged switch achieves 3.5x higher performance and 1.3x deployment efficiency over traditional switches.

Embodied AI Model Upgrade
NVIDIA released the humanoid robot foundation model Isaac GR00T N1, trained with synthetic data for 40% better performance than real-data-only models.

**Global Data Center Spending to Exceed $\mathrm{<span\; style="color:\#FFFFFF;">1</span>}\mathrm{<span\; style="color:\#FFFFFF;">T</span>}\mathrm{<span\; style="color:\#FFFFFF;">r</span>}\mathrm{<span\; style="color:\#FFFFFF;">i</span>}\mathrm{<span\; style="color:\#FFFFFF;">l</span>}\mathrm{<span\; style="color:\#FFFFFF;">l</span>}\mathrm{<span\; style="color:\#FFFFFF;">i</span>}\mathrm{<span\; style="color:\#FFFFFF;">o</span>}\mathrm{<span\; style="color:\#FFFFFF;">n</span>}<span\; style="color:\#FFFFFF;">\ast </span><span\; style="color:\#FFFFFF;">\ast </span>\mathrm{<span\; style="color:\#FFFFFF;">J</span>}\mathrm{<span\; style="color:\#FFFFFF;">e</span>}\mathrm{<span\; style="color:\#FFFFFF;">n</span>}\mathrm{<span\; style="color:\#FFFFFF;">s</span>}\mathrm{<span\; style="color:\#FFFFFF;">e</span>}\mathrm{<span\; style="color:\#FFFFFF;">n</span>}\mathrm{<span\; style="color:\#FFFFFF;">H</span>}\mathrm{<span\; style="color:\#FFFFFF;">u</span>}\mathrm{<span\; style="color:\#FFFFFF;">a</span>}\mathrm{<span\; style="color:\#FFFFFF;">n</span>}\mathrm{<span\; style="color:\#FFFFFF;">g</span>}\mathrm{<span\; style="color:\#FFFFFF;">p</span>}\mathrm{<span\; style="color:\#FFFFFF;">r</span>}\mathrm{<span\; style="color:\#FFFFFF;">e</span>}\mathrm{<span\; style="color:\#FFFFFF;">d</span>}\mathrm{<span\; style="color:\#FFFFFF;">i</span>}\mathrm{<span\; style="color:\#FFFFFF;">c</span>}\mathrm{<span\; style="color:\#FFFFFF;">t</span>}\mathrm{<span\; style="color:\#FFFFFF;">s</span>}$1 trillion in data center infrastructure spending by 2028, driven by AI inference. U.S. cloud providers have ordered 3.6 million Blackwell GPUs for 2025.

China’s Domestic Computing Ecosystem Updates
NVIDIA’s breakthroughs spurred gains in Chinese GPU-related stocks (e.g., Meili Cloud hitting limit-up). CPO tech (Co-Packaged Optics) gained traction, with firms like Accelink and InnoLight viewed as key players.

Chinese HBM Technology Accelerates
Amid U.S.-China tech competition, domestic firms are advancing high-end HBM R&D under a state-backed collaborative model to strengthen supply chain autonomy.