Ex-Apple Star Aims to Build the Perfect GPU Software
▼ Summary
– Chris Lattner founded Modular in 2022 to create a unifying software layer that helps AI applications run efficiently across different types of GPUs and CPUs.
– The startup’s core product is a software platform and a new Python-based coding language designed to let developers build AI apps for one chip and run them on another without major modifications.
– Modular’s long-term ambition is to become the dominant AI software platform and loosen the control that companies like Nvidia and AMD have over the industry.
– The company has gained early traction, forming partnerships with major chipmakers and raising $250 million in funding, bringing its total valuation to $1.6 billion.
– A major challenge for Modular is competing with Nvidia’s established, proprietary CUDA software platform, which locks in developers, unlike AMD’s open-source alternative.
The soaring demand for artificial intelligence has created an unprecedented need for powerful computing hardware, yet the software required to fully utilize these advanced chips often remains fragmented and restrictive. Chris Lattner, a veteran of Apple, Google, and Tesla, identified this critical bottleneck and co-founded Modular in 2022 to build a universal software layer that unlocks the full potential of diverse processors. The company’s mission is to create a seamless development experience, allowing AI applications to run efficiently across different GPUs and CPUs without requiring extensive rewrites for each unique hardware platform.
Modular’s strategy involves two core components. First, it offers a unifying software stack designed to help cloud providers maximize performance from their computing resources. Second, the startup has developed a new programming language, built upon the familiar foundation of Python, that enables developers to write code once and deploy it across various chip architectures. The fundamental idea is to eliminate the cumbersome process of adapting software for different vendors, thereby accelerating the pace of AI innovation.
Looking beyond immediate technical solutions, the company harbors a much broader ambition. It aims to challenge the dominant software ecosystems controlled by industry giants like Nvidia and AMD. Lattner envisions Modular becoming the default software platform for AI chips, a necessary unifying layer in a future where specialized processors proliferate. He argues that as sovereign AI initiatives and massive computing projects like “Stargate” emerge, the industry will require a common software foundation to manage the resulting complexity and diversity of hardware.
Early industry reception suggests this vision is gaining traction. Modular has secured partnerships with major players including Nvidia, AMD, and Amazon, who are exploring its technology. In a notable collaboration, the startup worked with SF Compute to develop what is claimed to be the world’s most cost-effective API for running large AI models. The platform’s compatibility has also recently expanded to include Apple Silicon GPUs, adding to its existing support for Nvidia and AMD hardware.
This growing momentum is backed by significant financial support. The company recently closed a $250 million funding round, its third in three years, which values the business at $1.6 billion. The investment was led by the US Innovative Technology Fund, with participation from DFJ Growth and returning backers such as General Catalyst, Greylock, and GV. According to Dave Munichiello of GV, the investment thesis is clear: while many companies focus on building chips, the real long-term value lies in the software that makes them accessible and powerful for developers.
However, the path to achieving this goal is fraught with challenges, largely due to the entrenched position of Nvidia. The company’s proprietary CUDA software platform, developed over two decades, creates a powerful lock-in effect for the vast majority of the GPU market. In contrast, AMD’s ROCm platform is open source, offering developers more flexibility to port their code. Modular’s success hinges on its ability to provide a compelling alternative that offers both high performance and unparalleled portability, breaking down the software barriers that currently segment the AI hardware landscape.
(Source: Wired)
