The Mojo language (by Modular, now Qualcomm) is now open-source
Modular has made a significant move to unify the fragmented AI hardware landscape, open-sourcing its Mojo 1.0 language and making its Modular Cloud platform publicly available. This strategic push provides a single, open software stack for deploying AI workloads across diverse accelerators like TPUs, Trainium, and Qualcomm's chips. The developer-focused announcements, including native Windows support for Mojo, resonate with the HN crowd looking for robust solutions to AI's infrastructure challenges.
The Lowdown
Modular, recently acquired by Qualcomm, has revealed a series of major announcements at ModCon, furthering its mission to provide a unified software platform for the increasingly heterogeneous AI hardware ecosystem. Four and a half years after betting on the need for such a re-architecture, the company states its platform is now production-ready, serving billions of tokens per minute.
Key announcements include:
- Mojo 1.0 Open Source: The entire Mojo language, including its compiler and tooling, is now fully open source under an Apache 2.0 license, offering a stable foundation and allowing for broad community contributions and extensions.
- Modular Cloud Public Availability: The full Modular stack is now publicly accessible as a production service, offering industry-leading inference performance and abstracting infrastructure complexity. It has already demonstrated strong performance in benchmarks and serves flagship customers like MiniMax.
- Expanded Hardware Support: Beyond NVIDIA and AMD GPUs, the Modular Platform now supports AWS Trainium, Google TPUs, Qualcomm Cloud AI 100, and Qualcomm Dragonfly accelerators. This enables developers to write models once and deploy them across varied architectures with significantly reduced engineering effort.
- Native Windows Support for Mojo: Modular is collaborating with Microsoft's Windows team to bring native support for Mojo to Windows, addressing a long-standing developer request.
- MAX Licensing Changes: The MAX license no longer restricts device usage and will become source-available with an open alliance program, encouraging broader ecosystem participation in platform development.
Modular's vision is to foster an open, horizontal platform where hardware vendors, model providers, and cloud companies can collaborate, allowing competition to focus on silicon merits rather than proprietary software stacks. This move aims to simplify AI development and deployment in a rapidly evolving hardware landscape.