Articles from semianalysis.com

Articles from semianalysis.com

A look at Microsoft freezing ~1.5GW of data centers set for 2025 and 2026; Microsoft has walked away from significantly more than 2GW of non-binding contracts (semianalysis.com)
5 days ago Microsoft

A deep dive on AMD 2.0: a new sense of urgency, rapid AI software stack progress, a critical talent retention challenge, ROCm lags Nvidia's CUDA, and more (semianalysis.com)
9 days ago AI Nvidia AMD

A look at Huawei's AI CloudMatrix 384 rack-scale system, which is less power-efficient than Nvidia's GB200 NVL72, though this is not a limiting factor in China (semianalysis.com)
17 days ago China AI Nvidia

A deep dive on what Trump's Liberation Day tariffs mean for AI infrastructure: how the tariffs work, the USMCA's GPU loophole, GPU/XPU global trade, and more (semianalysis.com)
23 days ago Trump AI

A close look at Nvidia's GTC 2025 announcements, which focused on addressing pre- and post- training scaling, inference-time scaling working in tandem, and more (semianalysis.com)
2025-3-19 Nvidia

As automation and robotics undergo a revolution, a call to action for the US and the West, which are missing the new labor economy that China is set to capture (semianalysis.com)
2025-3-11 China Robot

A close look at DeepSeek, which is estimated to have access to ~50K Hopper GPUs, a total server capex of ~$1.3B, and a GPU spend of $500M+ over its history (semianalysis.com)
2025-1-31

Benchmarking AMD's MI300X and Nvidia's H100 and H200; in theory, AMD's GPU has advantages in specs and Total Cost of Ownership, but software bugs hold it back (semianalysis.com)
2024-12-25 Nvidia AMD

A deep dive into AI scaling laws, including an outline of the old pre-training trends as well as the new scaling trends for post-training and inference time (semianalysis.com)
2024-12-12 AI

A look at Intel's problems, starting with 2016's 10nm node: 7/11 board members have no chip experience, most CEO options are worse than Pat Gelsinger, and more (semianalysis.com)
2024-12-9 Intel

A deep dive into Amazon's Trainium2, which shows the company is on a path to eventually provide competitive custom AI chips, especially for LLM inference (semianalysis.com)
2024-12-5 Amazon AI

Previous Page Next Page