Topic
Covers chip design, semiconductor manufacturing, fab economics, supply chain dynamics, and the companies building the physical infrastructure of the AI era.
Nvidia is paying $3.5 billion to make sure the next wave of custom AI chips still plugs into its racks. MediaTek is the wedge.
Samsung's first LPDDR-based processing-in-memory chip hits 614 GB/s and triples Llama 3.1 8B token throughput, in a drop-in 561-ball package.
Google's warrant for up to 58.97 million Marvell shares turns a supply contract into equity, and the AMD-OpenAI deal already proved the model works.