Mojo is Apache 2.0 now. The licence was never the hard part

Modular released the entire Mojo language, compiler and developer tooling under Apache 2.0 on August 18, three weeks after Qualcomm completed its acquisition of the company ^[1][2]. That settles the licence question and leaves the harder one open: RuntimeWire's account of the release argues Mojo's credibility now depends on compiler governance and on continued support for hardware that competes with its owner's ^[23].

The sequence is tight. Qualcomm closed the purchase on July 29 and made Chris Lattner executive vice president of advanced AI software and platforms, with Mojo, MAX and Modular Cloud keeping their brands ^[4]. Mojo 1.0 shipped on August 11, establishing a more stable language contract after years of frequent change ^[3], and the source release followed seven days later ^[20]. Simon Willison notes the open-source release had been promised since May 2023, which makes this roughly a 39-month wait ^[14][21].

What landed is substantial. The public repository now carries the compiler, standard library, examples and other platform components ^[5]; the standard library had already been open before this ^[6]. Phoronix, which had not expected much opening given what it called Qualcomm's mixed history, treats the full-language Apache 2.0 grant as the surprise ^[18].

The opening stops at a commercial line. MAX, Modular's inference framework and serving layer, is not covered by the Apache 2.0 release ^[7]. Modular said at ModCon that MAX's licence no longer contains device-use restrictions, that MAX will be source-available, and that an open alliance program is in development, with the final scope of that program still unclear ^[8]. Source-available plus an undefined alliance is a statement of intent, not a governance structure.

This is where the licence and the question diverge. Apache 2.0 gives anyone the right to fork a compiler; it says nothing about who reviews patches, who owns the roadmap, or whether a competitor's backend gets merged. RuntimeWire frames the test the same way: the treatment of outside compiler changes will determine whether Mojo develops shared governance or stays a Qualcomm-owned codebase that the public is free to fork ^[9]. Lattner's track record is the reason the question is worth asking rather than assuming. He created LLVM, helped create Clang, drove Swift and helped create MLIR, projects whose influence came from broad industrial participation ^[10]. None of those were owned by a chip vendor competing with their contributors.

The structural tension is plain. Qualcomm benefits when workloads run on Qualcomm silicon, while Modular's relevance depends on making the chip underneath interchangeable ^[13]. Modular addressed it at ModCon by saying the platform will keep supporting competing hardware, and announcing support for AWS silicon ^[12]. Watch which non-Qualcomm backends actually get maintained, not which get announced.

One thing to price in before treating Mojo as an escape hatch from CUDA and C++: it is no longer trying to be a Python superset. Willison records that the plan changed around August 2025, with Modular saying Mojo "may or may not evolve into a full superset of Python, and it's okay if it doesn't" ^[15]. Mojo is now its own language with Python-inspired syntax aimed at GPU programming, not full compatibility ^[16], though calls in both directions between Python and Mojo are supported ^[19]. Migration is work, and the work is yours.

Three markers over the next few releases: the scope of the alliance program ^[8], the first merged compiler change from someone outside Qualcomm ^[9], and whether the in-progress Windows port lands alongside existing macOS and Linux support ^[17].