Microsoft’s second-generation custom AI accelerator focuses on low-cost, high-throughput inference for Azure, pairing native FP4/FP8 with massive HBM3e bandwidth. Early deployments start in the US with Azure service integration.

Read more

NVIDIA’s Earth-2 platform pairs GPUs with the CorrDiff generative downscaling model to turn coarse CMIP6 and ERA5 data into kilometer-scale fields suitable for local decisions. Here’s how it works and how teams can adopt it.

Read more

NVIDIA has released Earth-2, a fully open, GPU-accelerated family of AI models, libraries, and frameworks for high-resolution weather and climate forecasting. It targets kilometer‑scale accuracy, rapid runtime, and lower energy use—positioning AI as a complement or alternative to traditional numerical weather prediction.

Read more

NVIDIA is releasing Earth-2, a fully open, accelerated family of AI weather models, libraries, and tools that enterprises and agencies can deploy, fine-tune, and visualize on their own infrastructure. The platform spans medium-range forecasting, nowcasting, downscaling, and global-scale visualization.

Read more

NVIDIA will invest about $2 billion in CoreWeave to accelerate a network of AI-optimized data centers targeting 5–10GW by 2030, with a roadmap spanning Rubin platforms, Vera CPUs, BlueField systems, and Quantum-2 InfiniBand.

Read more

FlashAttention-4 retools attention for Blackwell GPUs by maximizing on‑chip reuse and alleviating memory bottlenecks, delivering petaflop‑class throughput and meaningful speedups for long‑context LLM inference.

Read more