Comparative Analysis of RISC-V and ARM in AI Applications

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The exponential growth of artificial intelligence (AI) applications in edge devices, embedded systems, and data centers has intensified the demand for efficient and scalable processor architectures. This paper provides a comprehensive comparative analysis of RISC-V and ARM architectures in AI applications, examining their architectural features, performance characteristics, energy efficiency, and ecosystem maturity. We analyze both architectures across multiple AI domains, including edge AI, datacenter inference, computer vision, natural language processing, and embedded systems. Our investigation encompasses recent developments from 2020 to 2025, including RISC-V Vector Extensions (RVV 1.0), ARMScalable Vector Extensions (SVE/SVE2), and commercial implementations from leading vendors. Performance analysis reveals that ARM maintains a 15× advantage in raw inference speed for large models, while consuming higher power. In contrast, RISC-V demonstrates 3-5× superior performance-per-watt for inference workloads. Both architectures offer distinct advantages: ARM provides a mature ecosystem and proven scalability from mobile to data center, while RISC-V offers customization flexibility and eliminates licensing costs. Through systematic evaluation of 40+ research papers and real-world implementations, we identify optimal deployment scenarios for each architecture and discuss future directions for AI-optimized processor designs.