
Tesla, Google TPUs, SiFive journey

From Swift to Mojo and high-performance AI Engineering with Chris Lattner
Topic 6 of 14
Published Nov 5, 2025 · Clip 47:19–52:22 (5:03)
Chris frames his career as 1.0 (Apple, CPUs, developer tools) and 2.0 (AI): falling in love with AI in 2016 via the Photos app cat/dog model, joining Tesla to learn Caffe and TensorFlow, then Google where he owned the bottom of TensorFlow and scaled the TPU software platform. He learned it's very hard to build AI software without CUDA, built MLIR as a 'LLVM 2.0' for domain-specific chips, then joined SiFive to learn chip design on the hardware side of the boundary. Across every leg he kept asking 'where's the software?', which led to founding Modular with his Google co-friend to build LLVM-for-AI.