UK chip startup Tendrils raises $30m seed round led by Benchmark
Cambridge-based Tendrils has secured US-led funding to develop a general-purpose processor designed for thousands of parallel cores, alongside its own programming language.

Cambridge-based semiconductor startup Tendrils has raised $30 million (€26 million) in seed funding led by US venture firm Benchmark to develop a general-purpose processor architecture designed to support thousands of cores working in parallel, EU-Startups reported on November 11.
Jack Altman, a general partner at Benchmark and co-founder of HR technology company Lattice, has joined Tendrils’ board. Other investors in the round include Visionaries Club, the UK Sovereign AI Fund, FR8, Antimatter, Cloudberry VC, Conviction, Embassy Ventures and SV Angels, alongside angel investors.
The deal establishes a US investment and governance connection for the British company, rather than a disclosed operational expansion. The report does not identify a US office, local hiring programme or American customers. Its immediate significance is the backing of a US lead investor for an effort to redesign both processors and the software that runs on them.
A different route to parallel computing
Tendrils argues that modern CPUs remain constrained by design choices inherited from sequential, single-core computing. Maintaining compatibility with older software, in its view, adds complexity and limits how far processors can scale. Its proposed alternative is an architecture built around parallel operation rather than incremental improvements to those existing mechanisms.
Conventional processors use mechanisms such as cache coherency protocols to keep shared data consistent across cores, and reorder buffers to execute instructions more efficiently without changing a programme’s intended behaviour. Tendrils is instead moving away from traditional shared mutable memory and linear instruction sequences.
The foundation is a computational model called interaction nets, which represents programmes as networks of connected elements. Independent operations can run concurrently, potentially reducing the need for the coordination mechanisms used in conventional architectures. Thousands of parallel cores are the design ambition—not a demonstrated performance result in the supplied report.
Building the software alongside the chip
Founded in Cambridge in 2025, Tendrils is led by CEO and co-founder Nils Cremer, who previously researched interaction nets at ETH Zürich. His co-founders are T6, the developer of the open-source programming language Vine, and Hugo McNally, who previously worked at lowRISC on open-source semiconductor projects including OpenTitan.
Vine is central to the company’s approach. Its syntax resembles Rust, but it translates programmes into interaction nets. The aim is to let developers exploit the proposed processor’s parallel capabilities without manually allocating tasks across individual cores.
Developing the language and processor together addresses a basic adoption challenge for alternative chip architectures: software must be able to use the hardware effectively. For Tendrils, that makes the programming environment part of the product strategy, rather than a separate layer to be added after the processor design.
The seed round follows a pre-seed investment led by Visionaries Club in summer 2025, with participation from FR8 and angel investors. According to the EU-Startups account of Visionaries Club’s involvement, the firm first supported Cremer with a grant while he assembled the founding team.
The AI opportunity beyond GPUs
The potential market extends to AI workloads that require more varied processing than the large-scale mathematical calculations typically accelerated by GPUs. AI agents can execute code, search documents, run tests and perform interconnected operations—activities that can place substantial demands on CPUs even when a GPU runs the underlying model.
That gives Tendrils a potential role alongside AI accelerators, rather than simply as a replacement for them. The competitive proposition rests on handling irregular, parallel work more effectively than conventional processor designs. Benchmark’s investment finances further development of that proposition; it does not yet establish commercial adoption or a US market entry.


