{"slug": "qualcomm-open-sourced-the-mojo-compiler-three-weeks-after-paying-3-9-billion-for", "title": "Qualcomm Open-Sourced the Mojo Compiler Three Weeks After Paying $3.9 Billion for It", "summary": "Qualcomm open-sourced the Mojo compiler on GitHub under Apache 2.0 with LLVM exceptions on Tuesday, three weeks after closing its $3.9 billion acquisition of Modular on 29 July. The move, which makes the full toolchain buildable from source, is aimed at positioning Mojo as an alternative to NVIDIA's CUDA for heterogeneous computing. Modular is not yet accepting compiler pull requests but plans to open the compiler to outside patches by the end of the year.", "body_md": "*Apache 2.0 with LLVM exceptions, the full toolchain on GitHub, and a chip vendor now steering the language pitched as the way off CUDA.*\n\nModular pushed the Mojo compiler to GitHub on Tuesday under [Apache 2.0 with LLVM exceptions](https://www.modular.com/blog/mojo-open-source), ending four years in which the language was designed in public but built by a compiler nobody outside the company could read. The standard library went open in 2024, and hundreds of thousands of lines of kernel code followed. The part that actually turns Mojo source into machine code stayed shut until now.\n\nThe timing is what makes it interesting. Qualcomm [closed its acquisition of Modular](https://www.modular.com/blog/qualcomm-completes-acquisition-of-modular) on 29 July, a deal announced in June at roughly $3.9 billion in stock. Three weeks later, the new owner’s first visible move is to give the compiler away.\n\n## A chip company open-sourcing its own toolchain\n\nThat reads oddly until you look at what Mojo is for. It is Chris Lattner’s fourth act — he built LLVM, then Clang, then MLIR, then Swift — and Mojo leans on all of it. Python-shaped on the surface, a systems language underneath, and its whole reason to exist is compiling one codebase down to GPUs, NPUs and whatever accelerator ships next year. Write the kernel once instead of writing it in CUDA and rewriting it for everything else.\n\nQualcomm does not sell datacentre GPUs. It sells Snapdragon NPUs, Oryon cores and inference silicon that has to justify itself against racks of NVIDIA hardware. Every hour a developer spends locked into CUDA is an hour Qualcomm’s parts are not a candidate. Opening the toolchain costs it nothing it was ever going to earn in compiler licences, and buys the one thing it cannot buy outright: a chance that Mojo turns up in somebody else’s build system.\n\nThere is a detail in the org chart worth noting too. Modular’s engineering leadership includes Mostafa Hagog, who per the company’s own bio was an engineering director at NVIDIA running the teams behind cuDNN and CUTLASS — libraries that are a decent chunk of why CUDA is hard to leave.\n\n## What you actually get\n\nEverything sits in the [modular/modular](https://github.com/modular/modular) repository. Clone it and one Bazel command builds the compiler and standard library from scratch:\n\n`./bazelw run --config=build-mojo KGEN:mojo -- run hello.mojo`\n\nIf you would rather not wait on a full build, `--config=prebuilt-mojo`\n\npulls the latest nightly binary instead.\n\nThe licence choice matters more than it sounds. Apache 2.0 with LLVM exceptions is what LLVM itself uses: patent grant included, and the exception means binaries you compile do not drag attribution obligations along with them. Modular calls it “the gold standard for programming languages and compilers” — self-serving, and also true. It is the licence that lets Mojo end up inside a commercial product without a legal review first.\n\n## Readable, buildable, not yet collaborative\n\nOne caveat for anyone already reaching for a fork: Modular is not taking compiler pull requests. The standard library has accepted contributions since 2024, but the company says it wants to be deliberate about this “particularly in today’s era of AI coding”, and is aiming to open the compiler and tooling to outside patches by the end of the year.\n\nMojo hit [1.0 with source stability](https://www.modular.com/blog/modular-26-5-mojo-1-0-is-here) on 12 August, so the language is at least promising not to shift under anyone’s feet. Whether it dents CUDA is a harder question — NVIDIA’s moat was never really the compiler, it was fifteen years of libraries and everybody’s accumulated muscle memory. But the code is public now, and it was not a week ago.", "url": "https://wpnews.pro/news/qualcomm-open-sourced-the-mojo-compiler-three-weeks-after-paying-3-9-billion-for", "canonical_source": "https://hwbusters.com/news/qualcomm-open-sourced-the-mojo-compiler-three-weeks-after-paying-3-9-billion-for-it/", "published_at": "2026-08-19 06:12:30+00:00", "updated_at": "2026-08-19 07:41:12.316134+00:00", "lang": "en", "topics": ["developer-tools", "ai-infrastructure", "artificial-intelligence"], "entities": ["Qualcomm", "Modular", "Mojo", "GitHub", "LLVM", "NVIDIA", "CUDA", "Chris Lattner"], "alternates": {"html": "https://wpnews.pro/news/qualcomm-open-sourced-the-mojo-compiler-three-weeks-after-paying-3-9-billion-for", "markdown": "https://wpnews.pro/news/qualcomm-open-sourced-the-mojo-compiler-three-weeks-after-paying-3-9-billion-for.md", "text": "https://wpnews.pro/news/qualcomm-open-sourced-the-mojo-compiler-three-weeks-after-paying-3-9-billion-for.txt", "jsonld": "https://wpnews.pro/news/qualcomm-open-sourced-the-mojo-compiler-three-weeks-after-paying-3-9-billion-for.jsonld"}}