Sovereign AI invests in Tendrils to develop general-purpose chips

The processors powering modern computers were designed around a very different era of computing. As AI and other demanding technologies advance, the limitations of existing architectures are becoming increasingly exposed.

Tendrils is rethinking how chips work, with the ambition of making parallel computing practical for a much wider range of applications. We're backing the team as they develop their approach to general-purpose chip design.

Rethinking how computers work

GPUs are exceptionally good at running thousands of calculations simultaneously, making them central to modern AI. But they are designed for particular types of workloads. CPUs can handle a much broader range of software, but become increasingly difficult to scale as more cores are added.

Tendrils is working to combine the flexibility of CPUs with the scalability of parallel computing, developing a processor architecture designed to scale to thousands of cores. The key challenge is how to coordinate all that computing power. Conventional processors spend considerable time managing how different cores communicate and access shared memory. As more cores are added, this complexity can limit performance.

Tendrils is taking a different approach, using a computational model called interaction nets. Rather than relying on traditional sequences of instructions, it represents programs as networks of connected operations, helping identify which tasks can run independently and simultaneously.

This could allow processors to handle much larger workloads in parallel, with less overhead.

Hardware and software designed together

Building a new processor means rethinking both the hardware and the software that runs on it.

Tendrils is developing its own open-source programming language, Vine, alongside its processor architecture. Both are designed from the ground up for parallel computing, allowing thousands of cores to work together more efficiently.

By developing the hardware and software in tandem, Tendrils aims to make its architecture easier to program, while giving developers new ways to take advantage of parallel computing.

Why we’re backing Tendrils

Tendrils is taking on a fundamental challenge in computing in a radically different way. We believe its approach has the potential to make parallel computing useful for a much wider range of applications, opening up new possibilities for how businesses use computing power.

Josephine Kant, Investment Partner at Sovereign AI:

“Nils is an exceptionally clear thinker. Between him, Hugo, and T6, they have an unusually strong founding team that can build a category-defining winner. They’re unapologetically ambitious, and we’re delighted to be partnering with them as they rethink general-purpose chips.”