Gpu

• Nvidia launched Blackwell Ultra B300 GPUs on April 25, 2026, at GTC US, delivering 20 petaflops FP4 performance for training models up to 10 trillion parameters. • New chips feature 4.5TB HBM4 memory and NVLink 6 interconnect, reducing training costs by 40% for hyperscalers like Microsoft Azure. • Announcement bolsters Nvidia's 92% AI accelerator market dominance amid US export controls on China sales.

• Nvidia launched the Blackwell Ultra B300 GPU series on April 19, 2026, claiming 50% faster AI training speeds compared to the prior B200 model. • The new chips feature 208 billion transistors and support FP4 precision, targeting data centers for large language model workloads. • This positions Nvidia to maintain dominance amid intensifying competition from AMD and custom silicon by hyperscalers like Amazon and Google.

• Nava, an AI infrastructure startup, raised $22 million in funding led by Greenoaks Capital to expand its GPU cloud and data center capabilities. • The investment targets growing US demand for high-performance computing to train and deploy advanced AI models. • Nava's platform supports AI developers with scalable resources amid chip shortages and cloud competition.