This startup is creating a tool to break Nvidia's chip lock-in. Read the pitch deck Spectral Compute used to raise $6 million.

Spectral Compute, a London-based startup, has successfully raised $6 million in funding to develop a groundbreaking software framework aimed at liberating programmers from reliance on Nvidia chips. The company is working on technology that will enable applications originally designed for Nvidia's graphics processing units (GPUs) to run seamlessly on any compatible hardware. Nvidia's GPUs have become synonymous with the artificial intelligence revolution, largely due to their proprietary programming environment, CUDA. While this platform allows applications to leverage GPUs for a variety of computing tasks beyond graphics, it also creates a significant hurdle: programs developed using CUDA are not compatible with chips from competitors like AMD or Intel. This situation effectively establishes a protective barrier for Nvidia, complicating the process for companies wanting to transition their CUDA-based applications to alternative hardware. In response to this challenge, Spectral Compute aims to simplify the migration process for businesses looking to utilize non-Nvidia chips. CEO Michael Søndergaard explained, "We're building something that's compatible with CUDA, enabling software that operates within Nvidia's ecosystem to work effortlessly on competing hardware." Founded in 2018 by Søndergaard and his team, including Chris Kitching, Nicholas Tomlinson, and Francois Souchay, the startup's current technology is designed to work with specific AMD architectures, with plans to broaden its compatibility to include Intel chips and those from emerging semiconductor firms. The framework, named SCALE, is available for free use in non-commercial settings, and the company has formed partnerships with various research institutions for academic applications. For commercial use, Spectral Compute offers a licensing model tailored to the specific deployment type, collaborating with cloud service providers and aiming to create custom solutions for large-scale enterprise needs. Notably, there have been other initiatives, such as ZLUDA, an open-source project once backed by AMD, which faced challenges when AMD withdrew its support. In contrast, Søndergaard emphasizes that Spectral Compute's method is distinct. "We're taking a source-by-code approach, ensuring that we respect the original CUDA code authored by the developers, while adhering to all necessary licensing terms." The recent funding round was led by Costanoa and included contributions from Crucible and several notable angel investors. This financial boost will facilitate product development, expedite market entry, and support the expansion of Spectral Compute's growing team of 19 employees. An exclusive glimpse into the 13-page pitch deck that helped secure this funding reveals the startup's ambitious vision moving forward.