Someone wired up seven ESP32s to create a ~0.4B LLM, and so can you

Credit: Major-Nebula1743 Published Aug 23, 2026, 3:44 AM EDT Simon is a Computer Science BSc graduate who has been writing about technology since 2014, and using Windows machines since 3.1. After working for an indie game studio and acting as the family's go-to technician for all computer issues, he found his passion for writing and decided to use his skill set to write about all things tech. Since beginning his writing career, he has written for many different publications such as WorldStart, Listverse, and MakeTechEasier. However, after finding his home at MakeUseOf in February 2019, he would eventually move on to its sister site, XDA, to bring the latest and greatest in Windows, Linux, and DIY electronics. Sign in to your XDA account Summary This 7-node ESP32-S3 cluster runs ~0.4B-parameter LLM via SPI daisy-chain Master node runs BPE tokenizer and INT4 embeddings; 6 compute nodes handle transformer layers Slow but fun: about 9s per token—not for heavy use, but perfect as a DIY tinkering project We're getting to the point with LLMs where we can squeeze them onto an ESP32. Granted, it's not going to take down the bigger, stronger LLM models running on powerful hardware anytime soon, but as a fun thing you can make at home to handle simple queries, they can be a lot of fun. You can even hook up an external LLM to one, too; our own Adam Conway connected a local LLM to a $30 ESP32 display, and it designs a new screen for every question he asks. But what if you hooked up several ESP32s into a cluster and ran an LLM off it? Well, someone has created a ~0.4B parameter LLM that does just that. And while it's not winning any awards for speed, it's a cool DIY project you can do at home. This ESP32 cluster with seven nodes runs an LLM in tandem It's not very quick, but it is very cool In a post on the ESP32 subreddit, user Major-Nebula1743 shared details of their ESP32 AI cluster. It built off a previous project that got a 56M parameter model running off an ESP32 via ESP-NOW, and with the power of sticking more ESP32s together, they managed to make something bigger. Here's how it works: - 1 master node + 6 compute nodes (all ESP32-S3 N16R8) - I use SPI daisy-chain to link between ESP32s for high-speed communication. (no WiFi overhead) - The master node handles the BPE tokenizer and INT4 token embeddings. - Compute nodes each process, 4 layers of the Transformer blocks, and transfer the intermediate X to the next node to compute further layers. So, is this going to replace ChatGPT? Well, not quite. It's quite slow; it takes about 9 seconds to crunch through a single token on the 0.4B parameter model, so you won't be using this to vibe code anything anytime soon. However, as a cool little project, it certainly gets the job done. You can read more about this project, and perhaps even build your own, over on the project's GitHub repo.