On August 18, Modular pushed the source code for Mojo's compiler and toolchain to GitHub, merging it into the modular/modular repository under an Apache 2.0 license with the LLVM exception. The same release included several hundred thousand lines of GPU kernel code written in Mojo. Until now only the standard library had been open — accepting outside contributions since 2024 — while the compiler stayed locked inside the company.
The timing is pointed. Mojo 1.0 shipped just a week earlier, on August 11, with source-level stability as its headline pitch: the syntax and type system are now considered settled, and invalidated references get proper diagnostics. Open-sourcing the compiler a week later completes the pitch that this is a language worth building long-term investment on.
A Promise Kept, by Someone Else
When Mojo debuted in May 2023, Chris Lattner's roadmap already listed open source as a goal, just without a deadline. The direction shifted along the way: Mojo was originally pitched as a superset of Python, aiming to win users by staying compatible with existing code. In August 2025, the official line changed — Mojo might never become a full Python superset, and that would be fine. The positioning moved from "a faster Python" to "a systems language built for GPUs," with migration handled by AI-assisted rewriting rather than syntax compatibility.
It was the acquirer that finally followed through on open-sourcing. Qualcomm signed a definitive agreement with Modular on June 21 and closed the deal on July 29, paying up to 19.2 million shares of Qualcomm common stock; financial terms beyond that weren't disclosed. Chris Lattner became Qualcomm Technologies' senior vice president of AI software and platforms, and the Mojo, MAX, and Modular Cloud brands were kept intact. The blog post announcing open-sourcing ran on modular.com, but the decision came out of San Diego.
Qualcomm's open-source track record has long been mixed — its Snapdragon driver support in the Linux kernel has drawn community complaints for years. Opening the compiler less than a month after closing the acquisition moves faster than most expected.
Code Is Open, the Development Process Isn't — Yet
One detail is easy to miss: the code is readable, modifiable, and redistributable, but outside patches can't get in yet. Modular said in its announcement that it isn't ready to accept contributions to the compiler and tools, with a target of opening that up by the end of this year. In other words, this step is a license conversion from source-available to open source; community collaboration is still the second half of the story.
The license choice itself sends a signal. Apache 2.0 with the LLVM exception is the same combination the LLVM project uses on itself, and it means binaries compiled with it aren't bound by the runtime library's license — commercial closed-source products can ship it directly. For a language aimed at AI infrastructure, that detail matters more than the open-source label itself: whether chipmakers and cloud providers are willing to embed it in their own toolchains hinges on exactly this.
Lattner's earlier project offers a useful comparison. Swift launched in 2014 and went open source in December 2015, a gap of about a year and a half; Apple, too, released the code before opening up governance. Rust took the opposite path, growing inside an open repository from day one. Mojo now sits between the two: three years of staying closed bought freedom to lock down the syntax, at the cost of a community that never got the chance to scale.
Where It Runs Today
Mojo supports macOS and Linux, with Windows support still in progress. Target hardware spans CPUs, GPUs, and various AI accelerators. Its language design draws visibly on Rust's memory model while keeping two-way interoperability with Python — Mojo code can call Python, and Python can call back into Mojo.
That positioning fits neatly into Qualcomm's roadmap. Compute across data centers and edge devices keeps getting more heterogeneous, and every accelerator vendor maintains its own software stack, with a striking amount of duplicated work. When the acquisition closed, Lattner said publicly that the merger would accelerate Modular's goal of delivering unified AI software across every vendor's hardware. As Qualcomm pushes products like Dragonfly into the data-center market, what it's been missing is exactly this kind of middle layer — one that lets developers avoid rewriting code every time they switch chips.
The harder test comes after contributions to the compiler open up. A language's fate isn't decided by GitHub stars but by how many people are willing to maintain third-party libraries for it. Mojo's standard library has been open for two years without pulling in a particularly large volume of community contributions; the compiler sits at a higher bar. How many outside maintainers show up by that year-end deadline will say more than this license announcement did.
Sources: Modular official blog, Qualcomm investor announcement, CocoLoop, Phoronix; verified the license combination, the 19.2 million share consideration figure, and the acquisition closing timeline.