The people shaping Mojo as AI developers seek Python-compatible speed
Modular aims to solve AI infrastructure fragmentation with Mojo, a programming language that can run up to 68,000 times faster than Python. Co-founded by Chris Lattner and Tim Davis, the startup has raised $130 million to bridge the gap between AI research and production hardware.
Modular builds tools to solve AI fragmentation
Modular secured $100 million in a funding round led by General Catalyst with participation from GV, SV Angel, Greylock, and Factory. This investment, which brings the total raised by the startup to $130 million, supports the expansion of the Mojo programming language and hardware support. Chris Lattner and Tim Davis, both former Google employees, cofounded Palo Alto-based Modular in 2022. Lattner and Davis worked in the Google Brain research division and wanted to remove the complexity of building AI systems at scale. Modular employs roughly 70 people to build an engine that improves the performance of AI models on CPUs. This engine, currently in closed preview, allows developers to import trained models and run them up to 7.5 times faster than on native frameworks. The company will add GPU support later this year. Lattner argues that current AI infrastructure is too complicated and fragmented. Modular aims to solve this by addressing where AI software meets AI hardware. AI demand is approaching the limits of sustainability because generative AI models are 10 to 100 times bigger than older models. This massive scale leads to skyrocketing costs and insufficient compute capacity. Lattner believes Modular will allow for AI products to be powered in a way that is more affordable and sustainable for any enterprise. Other companies like Deci and OctoML also provide technology to make trained AI models more efficient and performant. The demand for AI hardware rises as generative models grow. Microsoft has warned of server hardware shortages that might lead to service disruptions. This demand has driven the market cap of GPU provider Nvidia to $1 trillion. More than half of AI decision makers in top companies report facing barriers to deploying the latest AI tools.
Mojo aims for Python-level usability and C-level speed
Mojo aims to combine the usability of Python with capabilities like caching and metaprogramming. It functions as a superset of Python. Modular claims Mojo applications can run up to 68,000 times faster than Python applications. The language works with existing machine learning frameworks such as TensorFlow and PyTorch. Because it is built on Multi-Level Intermediate Representation, or MLIR, it targets heterogeneous systems like CPUs, GPUs, and AI ASICs. This foundation allows the language to target many different hardware types. This is the tech Lattner spearheaded at Google.
| Capability | Technical Detail |
|---|---|
| Performance Target | Up to 68,000x faster than Python |
| Framework Support | TensorFlow, PyTorch |
| Deployment Model | Open-source transition at ModCon ’26 |
| System Support | Linux (Mac and Windows coming soon) |
The Mojo SDK is currently available for download on Linux systems. Modular intends to release support for Mac and Windows in the future. The Mojo driver enables programming with REPL and VS Code. This accessibility aims to reduce the learning curve for developers moving from Python. Lattner and Davis designed Mojo to bridge the gap between research and production. The language allows for incremental typing-for-performance by adding type annotations and explicit variable declarations. Developers can manage dynamism through decorators, such as using @strict to opt-in to constrained dynamism using vtables. This approach uses class hierarchy analysis to devitalize and inline method calls. One user found that Mojo 0.26.1 failed to compile basic Python-like code, which indicates a current implementation gap. You already know that Python’s dominance in machine learning makes it difficult for any new language to gain traction. The company intends to address the fragmentation of the AI stack by providing a single language that manages diverse processing tasks and different hardware through the use of powerful compile-time metaprogramming and adaptive compilation techniques.
Community growth and the open-source pivot
The Modular community grew to over 120,000 developers in the four months following the product keynote in May. Other reports indicate that Mojo has captured the interest of 175,000 developers and 50,000 organizations. Since the launch in May, the language has been downloaded by more than 12,000 developers and 19,000 members in a Discord community. The company plans to release Mojo in general availability early next month. Modular will move Mojo to an open-source model at ModCon ’26 to leverage community contributions for faster development and bug resolution. This transition requires preparing the codebase and documentation. The language adopts the Apache 2 license, which is compatible with GPL2. This allows Mojo to integrate with open-source projects like the Linux kernel. Mojo can compile standalone executables without dependencies. This capability helps in production environments. The transition also aims to prevent the kind of migration difficulties seen when users moved from Python 2 to 3. The company currently has 30,000 people on the waitlist.
Will the transition to an open-source model at ModCon ’26 attract enough contributors to sustain the momentum?