RustConf 2025 Interview: Nathaniel Simard, Co-founder & CEO, Tracel AI

At RustConf 2025 in Seattle, the Rust Project Content Team talks with Nathaniel Simard, creator of the Burn deep learning framework. Nathaniel shares how he went from trying Rust around 1.0 in Quebec City, to web consulting, to a master's in machine learning in Montreal, to a startup writing recommendation engines in Rust and ML in Python, and finally to building Burn on nights and weekends and founding a company around it. Topics include: • The state of AI in Rust: Burn for training and inference across NVIDIA, AMD, and Mac GPUs, Android, and more, plus Hugging Face's Candle for LLMs • Training on big hardware and deploying the same codebase to small devices with a different compile option • CubeCL: a just-in-time GPU compiler where you write Rust and target CUDA, Vulkan, AMD, and Mac GPUs, including Tensor Cores, for any numerical computing • On-device AI and keeping your data local, and where federated learning fits • Prototyping fearlessly in Rust: shipping an imperfect first version, then refactoring with confidence because the type system marks exactly where the shortcuts live • Breaking a type on purpose and following the compiler errors until it works • Building an open source business: a cloud platform complementary to the open source project, not a replacement for it Links mentioned: Burn: https://burn.dev/ Burn on GitHub: https://github.com/tracel-ai/burn CubeCL, GPU compute in Rust: https://github.com/tracel-ai/cubecl Candle, Hugging Face's deep learning framework: https://github.com/huggingface/candle wgpu: https://github.com/gfx-rs/wgpu Bevy game engine: https://bevyengine.org/ Interested in AI and Rust? Burn is open source, accepts contributions, and has a Discord for questions. Jump in and get involved. Interview conducted by Pete LeVasseur on behalf of the Rust Project Content Team.