--- title: "Rockchip (瑞芯微) rides the edge-AI wave as its RK3588 SoC becomes the default brain for Chinese robots and cars" date: 2026-10-18 category: Chips & Compute site: NeuroAI canonical: https://neuroai.site/a/na-chips-rockchip-aiot-soc language: en --- # Rockchip (瑞芯微) rides the edge-AI wave as its RK3588 SoC becomes the default brain for Chinese robots and cars > Rockchip (瑞芯微) posted 2024 revenue of ≈ US$436M with profit up 341%, powered by its RK3588 AIoT SoC shipping into robots, vehicles, and vision gear. The robot on a factory floor does not think in a distant cloud. It has to see a part, decide in milliseconds, and move — all on a few watts of power. The chip making that possible is rarely the famous GPU in a data center. More often, it is a small, cheap, power-sipping system-on-chip sitting quietly on the robot's mainboard. In China, that chip is increasingly from **Rockchip (瑞芯微)**. ## A profit jump that signals a shift Rockchip (瑞芯微) closed 2024 with operating revenue of **RMB 3.136 billion (≈ US$436M / HK$3.4B)**, up **46.94%** year on year, and net profit of **RMB 595 million (≈ US$83M / HK$649M)**, up a striking **341.01%**. The company put the surge down to its AIoT SoC platform, led by the flagship **RK3588**, and to a "group-wide, multi-tier product portfolio" that lifted share across downstream lines — especially automotive electronics, machine vision, and industrial and embedded applications. The momentum carried into 2025. First-quarter revenue reached **RMB 885 million (≈ US$123M / HK$964M)**, up **62.95%**, with net profit of **RMB 209 million (≈ US$29M / HK$228M)**, up **209.65%**. ## The RK3588: Rockchip's technical calling card The RK3588 is the product that reframed how the market sees the company. Built on an **8nm** process, it pairs: - A CPU of **four Cortex-A76 and four Cortex-A55** cores. - A **Mali-G610** GPU. - A dedicated **6 TOPS NPU** for neural-network inference. - **8K** video encode/decode. Six trillion operations per second is not data-center scale, but for on-device vision, voice, and control it is plenty — and far cheaper and cooler than bolting a server GPU onto a robot or a dashboard. Rockchip positions a full ladder of parts around it: the RK3588 flagship, the mid-tier **RK3576** and **RK356X**, and the entry **RV11** series, covering a compute band from roughly **0.5 to 6 TOPS** to match wildly different price and power envelopes. ## Why edge AI is the real story The buzz about AI in China has been dominated by giant data centers and the large model (大模型) race. But a second, quieter revolution is happening at the edge — inside cars, cameras, factory arms, and home devices. This is where a general-purpose AI accelerator is overkill and a tailored SoC wins. Rockchip's own framing is that the drop in on-device AI development cost, plus the open-sourcing of domestic large models like DeepSeek, created a fresh opportunity for edge and endpoint AIoT. When a model can be shrunk and run locally, every camera and controller becomes a candidate for intelligence. According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, the edge-AI opportunity rewards chip vendors who can ship a *family* of parts — not one hero chip — because deployment spans everything from a sub-watt sensor node to a humanoid (人形机器人) torso computer. ## A customer list that proves breadth Rockchip is not a one-customer shop. Its disclosed customer base spans BYD, Baidu, Goertek, Inovance, Lenovo, Midea, Xiaomi, and Anker — names across autos, acoustics, industrial drives, PCs, appliances, and consumer tech. That "broad coverage plus deep binding" structure lets the company ride different downstream cycles without betting everything on a single market. For a China-AI reader, the takeaway is subtle but important: sovereignty in AI is not only about training clusters. It is also about the millions of inference chips embedded in products that actually ship to consumers and factories. Rockchip sits squarely in that layer. ## What comes next: coprocessors and RK3688 Two forward hints came out of the 2024 disclosures. First, Rockchip said it is developing **coprocessor chips** to address the gap between fast-rising AI compute demand and the slower iteration cycle of a full SoC — a coprocessor can upgrade a device's AI ability without redesigning the whole board. Second, it signaled a next-generation flagship, the **RK3688**, on an advanced process node, meant to open the company's next growth cycle. These are roadmap statements, not shipped products, so they should be read as direction, not spec. ## The quiet strategic meaning Step back from the quarterly numbers and the picture is about resilience as much as growth. China's AI story is often told as a contest of who has the biggest training cluster. But intelligence only delivers value when it reaches the physical world — a car that parks itself, a camera that flags a defect, a humanoid (人形机器人) that balances on uneven ground. Those moments happen at the edge, on chips measured in watts and dollars, not kilowatts and millions. A healthy domestic edge-AI SoC vendor means the country is not just building brains in the cloud; it is putting smaller, cheaper brains into everything it manufactures. ## What readers can do now - **Watch the RK3588 design-win count.** Each new robot, automotive, or vision product built on it is a data point on edge-AI adoption in China. - **Track the coprocessor and RK3688 roadmap.** If Rockchip ships a modular AI upgrade part, it changes how cheap devices get smarter. - **Separate edge from cloud in China-AI coverage.** Edge SoC vendors like Rockchip tell you about deployment; accelerator vendors tell you about training. Both matter. ## Honest limitations - **Viewpoint limits.** This article reads Rockchip as an edge-AI compute story, not as a stock recommendation or a benchmark shoot-out against Qualcomm, MediaTek, or Ambarella. - **Data-source limits.** Revenue, profit, and product specs come from Rockchip's 2024 annual report and exchange-connected Chinese financial media (STCN, Phoenix, Tencent finance). Customer names derive from brokerage summaries citing company disclosures; we have not independently verified each design win. - **Uncovered dimensions.** We do not quantify Rockchip's exact unit shipments, its overseas revenue share, or the node and performance of the unannounced RK3688. Supply-chain exposure to advanced packaging constraints is also outside this article's scope. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-chips-rockchip-aiot-soc Free to quote with attribution and a link to the original.