IRCNF

RISC-V Crossed 25% Market Share. It's Not an IoT Story Anymore.

Share:
RISC-V Crossed 25% Market Share. It's Not an IoT Story Anymore.

RISC-V crossed 25% global market penetration by late 2025 — a milestone that would have seemed implausible five years ago when the architecture was still considered a niche option for microcontrollers and embedded systems. The number alone understates the shift. What changed isn't just the adoption curve; it's the segment mix. RISC-V is now showing up in data centers, AI inference accelerators, and automotive SoCs. The IoT story is no longer the whole story.

Why RISC-V Stayed in IoT for So Long

The open ISA (Instruction Set Architecture) that Berkeley researchers published in 2010 had obvious appeal for embedded and IoT applications: royalty-free silicon, a minimal base instruction set that reduced chip area and power consumption, and a modular extension system that let designers add exactly the instructions their workload needed and nothing else. For companies building billions of microcontrollers — Espressif, SiFive, Microchip — RISC-V made sense economically.

The barrier to data-center and high-performance workloads was more fundamental: the software ecosystem. ARM had decades of compiler optimization, OS tuning, and performance characterization across workloads that RISC-V lacked. A RISC-V server might have hardware competitive with ARM on paper, but if the compiler generates 15% worse code because the optimization passes haven't been tuned for RISC-V's microarchitecture, the hardware advantage disappears.

That gap is closing. GCC and LLVM support for RISC-V has matured significantly. The RISE (RISC-V Software Ecosystem) initiative under the Linux Foundation has coordinated optimization work across compilers, runtime libraries, and OS kernels in a way that produced measurable improvements in real-world benchmark performance between 2023 and 2025.

The Acquisition Signal

Corporate acquisitions are a reliable signal of strategic value. Qualcomm acquired Ventana Micro in December 2025. Ventana had been building high-performance RISC-V cores targeting server workloads — its V2 chip, a 32-core design on TSMC 3nm, was anticipated for 2026. Qualcomm didn't buy Ventana for its current products; it bought the engineering team and IP to bring RISC-V capability into its data-center and AI accelerator roadmap.

Meta's acquisition of Rivos in October 2025 tells a similar story. Rivos was building RISC-V chips specifically for hyperscaler workloads — memory-intensive inference, model serving, and large batch processing. Meta's internal compute needs are substantial enough that reducing per-unit silicon costs through royalty-free RISC-V designs has material economics at scale. Rivos gave them an accelerated path to custom silicon without ARM licensing.

Neither acquisition is about replacing ARM overnight. Both are about building optionality: when RISC-V performance parity arrives — projected near the end of 2026 for high-performance cores — these companies want the design capability in-house rather than paying ARM for the same compute they could own.

The Geopolitical Accelerant

US export controls on advanced semiconductors, combined with ARM's licensing model (headquartered in the UK, with licensing terms potentially subject to Western government pressure), have made RISC-V strategically attractive for countries pursuing semiconductor sovereignty. China, India, the EU, and Saudi Arabia have all funded domestic RISC-V design initiatives.

China's situation is the most acute: ARM designs are embedded throughout Chinese chip supply chains, and the risk of losing access to ARM IP has made RISC-V a national security priority. ByteDance's dual-track CPU development program — maintaining ARM designs while building RISC-V equivalents — is the hedge that any large Chinese tech company would want given the current geopolitical environment.

For the global RISC-V ecosystem, this concentration of sovereign investment has a positive externality: it funds toolchain development, compiler optimization, and silicon design work that benefits all RISC-V users.

AI Workloads as the Inflection Point

The AI inference workload is where RISC-V's modular ISA extension model has its clearest advantage. A custom AI accelerator built on RISC-V can add vector extensions precisely sized for its tensor operations — no wasted silicon area on instructions it will never execute. ARM's SVE/SVE2 extensions offer similar capabilities, but the licensing model means paying ARM for every unit of hardware even when the key innovation is in the custom extensions your team wrote.

SpacemiT's K3 (60 TOPS) and upcoming K5 (120 TOPS, TSMC 3nm) are examples of what RISC-V-based AI inference looks like at the commercial product level. Tenstorrent's Atlantis SoC, combining its own neural processing units with Andes' RISC-V Ascalon X CPU cores, is targeting the same space from a different angle. These are not experimental designs — they're commercial silicon with customer commitments.

What 25% Actually Means

Market penetration of 25% doesn't mean RISC-V is displacing ARM in high-performance workloads yet. The segment breakdown matters: IoT still holds 34.5% of revenue share, and automotive and data center segments are growing fastest from a smaller base.

But the trajectory is clear. The data center and AI segment is growing at 42.6% CAGR through 2031. Automotive is growing at 42.9%. The total RISC-V market is projected to reach $1.89 billion in 2026 on its way to much larger figures by the end of the decade. At some point — probably mid-decade — the conversation shifts from "RISC-V is promising" to "RISC-V is infrastructure."

The 25% milestone matters because it crossed the threshold where ecosystem fragility stops being a realistic concern. The toolchains, OS support, and developer expertise now exist in sufficient depth that building on RISC-V is a reasonable production choice, not an experimental one. That changes the calculus for anyone designing silicon over the next five years.

Share:
RISC-V Crossed 25% Market Share — Moving Beyond IoT | IRCNF | IRCNF - Intelligent Reliable Custom Next-gen Frameworks