Chris Lattner
Co-founder & CEO, Modular
About
Chris Lattner is an American computer scientist and the co-founder and CEO of Modular, where he is developing Mojo, a high-performance AI programming language. He is the original creator of the LLVM compiler infrastructure, the Clang compiler, and the Swift programming language. He spent twelve years at Apple leading the Developer Tools team, then held roles at Tesla, Google (where he co-created MLIR), and SiFive. His work on LLVM earned the ACM Software System Award in 2012.
Key Contributions
- Created LLVM, the modular compiler infrastructure that became a foundation for modern language toolchains
- Built Clang, replacing much of the old C/C++ compiler stack with faster, more modular tooling
- Led Swift at Apple, pairing systems-level performance with safer, developer-friendly language design
- Co-created MLIR at Google, giving compilers a shared representation for machine-learning and hardware-specific optimization
- Founded Modular and Mojo to attack the Python-performance gap in AI infrastructure
- His tools are widely admired, though Mojo's promise still faces the hard adoption problem every new systems language faces
Videos & Interviews
Connections
Jeff Dean
CollaboratedChief Scientist, Google DeepMind
Lattner joined Google Brain in 2017 to run TensorFlow's infrastructure with Dean as his manager, and by his own account MLIR began in a single sentence from him: I agree we have a compiler problem — go build a new compiler to unify this mess. What came out was a shared intermediate representation that lets many frameworks target many chips, including the TPUs Dean had helped will into existence. It is a reminder that most of what makes large models possible is not architecture but plumbing, built by people who read hardware and language as one problem.
modular.com · en.wikipedia.org
Jensen Huang
In contrastFounder, President & CEO, NVIDIA
Modular exists because of what NVIDIA built. CUDA turned a graphics company into the substrate of AI by making its own hardware the only place the software was pleasant to write; Lattner's answer — first MLIR, then Mojo — is to attack that lock-in at the compiler layer, so performance need not be tied to one vendor's ecosystem. Whether AI compute stays a moat or becomes a commodity is being decided in this unglamorous stratum, not at model launches.
