Open cluster runs a 0.5B BitNet LLM across 7 ESP32-S3 boards
A 1.58-bit model split over daisy-chained microcontrollers puts offline LLM inference on cheap dev boards — as AMD buys Fei-Fei Li's World Labs for $8.2B.
Copy markdown
A transformer split across seven microcontrollers
The open-source ESP32s3-LLM-Cluster runs a 0.5B, 1.58-bit BitNet model across seven ESP32-S3 boards — one master handling tokenization and embeddings, six compute nodes daisy-chained over SPI for the attention and MLP layers. It ships as MIT-licensed ESP-IDF firmware plus Python quantization tools, so a stack of cheap dev boards becomes a fully offline inference cluster with no GPU and no cloud. A concrete weekend build if you want an LLM running at the edge.
AMD buys Fei-Fei Li's World Labs for $8.2B
AMD is acquiring spatial-intelligence startup World Labs in an ~$8.2B all-stock deal, with Fei-Fei Li joining as EVP and chief scientist under Lisa Su. World Labs' models generate and simulate interactive 3D worlds from text, image, or video, and AMD says it will fold the work into its CPU, GPU, and software roadmap for 'an open ecosystem.' It's the clearest sign yet that AMD wants to compete on the AI software stack, not just the silicon — worth tracking if you're weighing ROCm against CUDA.