Tendrils raises €26M seed to rebuild the CPU in 2026

Lead

Tendrils, a semiconductor startup founded in 2025 in Cambridge, UK, announced a €26M seed round on November 11, 2026. Benchmark led the round, and its general partner Jack Altman, the Lattice co-founder, joins the board. The company is building a general-purpose processor in which thousands of cores run simultaneously, rather than a few fast cores working through instructions in sequence.

Who Backed the Round?

Benchmark led, with participation from Visionaries Club, the UK Sovereign AI Fund, FR8, Antimatter, Cloudberry VC, Conviction, Embassy Ventures and SV Angels, plus a group of angel investors. Tendrils has not disclosed the valuation.

Visionaries and FR8 backed the company earlier, in a pre-seed round in summer 2025. The seed therefore follows that first check by roughly 15 months and arrives before any silicon has been shown publicly. That is a large bet on architecture alone. The presence of the UK Sovereign AI Fund, a state-linked vehicle, adds a domestic-capability angle to what is otherwise a US-led deal.

What Is Tendrils Building?

Tendrils is building a CPU that runs everything it can in parallel, using a computation model called interaction nets. In that model, each step is a small, local rewrite of a graph that does not depend on the others. The design avoids shared mutable memory and linear instruction sequences.

Conventional processors carry heavy machinery to keep sequential software fast: cache coherency protocols, reorder buffers and out-of-order execution. Tendrils argues that this machinery, bound to decades of backwards compatibility, is what has capped general-purpose CPU gains since the mid-2000s. Removing it, in the company's framing, also removes race conditions as a class of bug.

The company is also developing Vine, an open-source, Rust-like programming language built for parallel-first hardware. Developers would write in Vine without managing threads or learning CUDA. The language gives the hardware a software on-ramp, which has been the usual point of failure for architectures that break with x86 and Arm.

Why Is a CPU Startup Raising Now?

Investors are funding CPU alternatives now because AI agent workloads have shifted attention from GPUs to the processors that surround them. Tendrils' backers point to tool execution, which runs on CPUs, accounting for up to 90% of latency in some agent workflows. If that figure holds broadly, the sequential CPU becomes the bottleneck in a stack where accelerators have already been optimized heavily.

That argument deserves some skepticism. The company's own market framing is a pitch, not a measurement, and agent latency varies widely by task. Still, it explains why a seed-stage hardware company with no disclosed product could attract a top-tier US venture firm.

What Are the Risks?

The central risk is software compatibility. A parallel-first processor cannot run existing x86 or Arm binaries efficiently by design, so every program must be rewritten or recompiled through a new toolchain. Past attempts at massively parallel general-purpose designs, from the Transputer era onward, have struggled against the inertia of existing code.

Capital is the second constraint. A €26M seed covers architecture work, compilers and early prototypes, but a fabricated chip on an advanced process node costs far more. Tendrils has not said how it will fund tape-out or on which node it plans to build. The founding team is also young: CEO Nils Cremer was an ETH Zurich student, co-founder T6 created Vine and contributed to Parity, and co-founder Hugo McNally previously worked at open-source chip group lowRISC.

What Comes Next for Tendrils?

Tendrils' next milestone is a working demonstration of the architecture, likely through simulation or FPGA prototypes, before any custom silicon. The company has not published a timeline. Vine, already open source, offers the nearest evidence: adoption by outside developers would indicate whether the programming model has appeal independent of the hardware.

A second seed-to-Series A test will be whether benchmarks show gains on real workloads rather than hand-picked parallel ones. Investors will look at performance per watt and per dollar against established server CPUs, not core counts alone.

Outlook

Tendrils has secured €26M and a Benchmark board seat to challenge the sequential CPU design that has defined computing for decades. The funding is sizeable for a pre-product hardware company, though valuation, use of funds and chip timeline remain undisclosed. Success depends on proving real-world speedups and winning developers to Vine, while capital for fabrication remains an open question.