* With apologies to Benjamin Franklin, who, asked at the close of the Constitutional Convention in 1787 what sort of government the delegates had produced, is said to have replied: "A republic, if you can keep it."
On August 18, Qualcomm (NASDAQ: QCOM) open sourced the Mojo compiler under the Apache 2.0 license, three weeks after closing the acquisition that gave it the code.
Qualcomm announced its acquisition of Modular on June 24, 2026, and closed it on July 28 — an all-stock deal headlined at $3.9 billion, though all-stock deal math moves with the share price and the number actually booked was nearer $3.1 billion. Modular is the compiler company founded by Chris Lattner, who is now a Qualcomm executive vice president running advanced AI software and platforms.
Let’s spend some time reflecting on Qualcomm’s integration of Modular and how it intersects with their overall platform strategy. Spoiler alert: so far, considering Qualcomm is a hardware company, it has been uncharacteristically enlightened...
For those who do not know, Lattner is the legendary architect of LLVM, the open source compiler stack that underpins a multitude of translation technologies too numerous to name. At Modular, Lattner resolved to build Mojo, a DSL for machine learning that promised to reconcile the notoriously difficult tensions between expressiveness, performance, and hardware portability.
In the best of times, persuading developers to learn and adopt a new language is a formidable undertaking. As with all platform adoption, the defining figure of merit is return on investment, and learning a new language is arguably an even bigger investment than learning how to use a new API. Mojo did achieve some modest adoption, after building a cross-platform runtime that underpinned its cross-platform DSL — but modest is the operative word.
I’d been wondering whether Modular would yield a favorable exit, because we all have been conditioned — in no small part due to the success of Lattner’s own LLVM — to think of toolchains as a commons with indirect monetization. Purveyors of platforms often make the toolchains targeting those platforms free, because doing so decreases the “investment” in the aforementioned “return on investment.” A free compiler is not charity—it is a subsidy to drive adoption.
I shouldn’t really have been wondering, though.
The rest of this post is for members.
A CUDA Handbook membership is ad-free reading of the whole book plus full access to paid posts, from $10/month. Become a member. Already a member? Log in to read it here.