RISC-V is an open Instruction Set Architecture governed by a Swiss non-profit, RISC-V International. Royalty-free, no licensing fees, no single-vendor control. Anyone can build a RISC-V chip without permission. This is the foundational layer that makes "open silicon" possible at all; the other two major ISAs (x86 and ARM) are proprietary and tied to specific licensors. For the open-source AI conversation, RISC-V matters because the AI hardware stack is dominated by NVIDIA's proprietary GPU ISA on top of NVIDIA-controlled hardware. Even if the weights and runtime above are open, the silicon underneath remains a chokepoint. RISC-V is the only credible path to an open accelerator stack that does not depend on one company's goodwill. The lead bet at the AI-accelerator level is Tenstorrent, whose Wormhole and Blackhole accelerators combine RISC-V CPU cores with their Tensix tensor units. Production-ready as an ISA, yes; it ships in embedded and microcontroller chips at very high volume already. As a substrate for AI training or inference at frontier scale, not yet; the software ecosystem (compilers, optimized kernels, AI frameworks targeting it) lags NVIDIA's CUDA stack by years. SiFive is the leading commercial RISC-V CPU IP vendor; Tenstorrent is the leading RISC-V-anchored AI accelerator bet. Status: foundational, real, and slow to mature into the AI-frontier role its open-stack advocates want for it.
The Stack · Silicon · Open source
RISC-V
Open instruction set architecture; royalty-free; substrate for open silicon (CPUs and emerging AI accelerators).
Sources
- RISC-V International — About https://riscv.org/about/
- RISC-V Specifications https://riscv.org/specifications/
- SiFive — RISC-V CPU IP https://www.sifive.com/
- Tenstorrent — Vision (RISC-V positioning) https://tenstorrent.com/vision
Want a follow-up? Ask the chat about RISC-V in context. It will compare to siblings at the same layer and ground every claim in the wiki.
Other projects at the Silicon layer
6 siblings · ordered open first
- Tenstorrent (Wormhole, Blackhole) Open source
Open-trending AI accelerators on RISC-V; Jim Keller-led; tt-metal and tt-forge open.
- NVIDIA H100 / H200 Proprietary
Hyperscaler-class AI accelerator with CUDA software moat; default frontier-training and frontier-inference hardware.
- AMD MI300X / MI325X Proprietary
Highest-memory accelerator on the market (192 GB+ HBM); ROCm software stack open-source-adjacent.
- Cerebras CS-3 Proprietary
Wafer-scale accelerator; proprietary but disruptive on inference economics for specific model sizes.
- Groq LPU Proprietary
Language Processing Unit; proprietary; extraordinarily fast inference for small-to-medium models at low batch sizes.
- Apple Silicon (M-series) Proprietary
Unified memory architecture; closed silicon, but the strongest on-device inference platform via llama.cpp and MLX.