At the recently held 26th China International Industry Fair, D-Robotics made its debut exhibition. Unlike traditional chip vendors that showcase products independently, D-Robotics dedicated more than half of its booth space to ecosystem partners.
Source: D-Robotics
During the fair, Jiang Zhou, Vice President of Sales at D-Robotics, was interviewed by EEFOCUS reporters, elaborating on the strategic thinking and implementation paths of this AI chip company, which spun off from Horizon Robotics, across the dual battlefields of industrial automation and embodied intelligence.
From Consumer to Industrial: A "Proactive Expansion" Driven by Passive Momentum
"Over the past few years, without large-scale promotion in the industrial market, approximately 20% of D-Robotics' customers have passively come from the industrial sector," Jiang Zhou stated.
For a long time, traditional industrial solutions have relied heavily on general-purpose platforms such as Intel and NVIDIA, or generic main control solutions provided by competitors like Allwinner and Rockchip Electronics for edge computing and display control tasks. The former tend to have relatively high costs and room for improvement in localized services, while the latter face limitations when dealing with increasingly complex AI models.
Jiang Zhou believes that robotics is a composite industry encompassing various niche scenarios, where industrial applications and robotics are closely aligned. With the explosion of embodied intelligence, many traditional industrial clients have felt anxiety about transformation upon seeing new technologies, fearing their existing businesses could be disrupted by new paradigms. Consequently, there is an urgent need to integrate AI concepts into current production lines. This market demand has driven a rigid requirement for high-computing-power, cost-effective chip solutions. By bundling "chips, solution partners, and boards" to approach traditional industrial enterprises, D-Robotics aims to respond to this market shift.
Lowering the Barrier to Industrial AI with Tool Platforms
The process of industrial AI adoption has historically been slow, primarily due to two practical pain points: first, industrial scenarios are extremely fragmented, making it impossible for a single general-purpose large model to solve problems across all specific production lines; second, there is a shortage of AI talent, making it difficult for deep-vertical industrial enterprises to compete with internet giants or automakers for skilled personnel.
To address this, D-Robotics unveiled its RoboGo™ software platform for the first time at this year's Industry Fair. Covering data collection, model training, simulation verification, and quantized deployment, the platform serves the entire AI development lifecycle via the cloud.
Source: D-Robotics
Jiang Zhou pointed out that the core purpose of launching these software tools is to break the traditional inertia that equates digitalization with heavy investment. Industrial enterprises do not need to assemble specialized teams of dozens of high-end AI experts. Through new development methods such as AI Coding offered by RoboGo™, the labor cost of attempting AI integration can be significantly reduced, helping small and medium-sized traditional enterprises within ecosystems like KUKA and ABB achieve a cold start from zero to one with an asset-light approach.
Regarding external concerns about industrial data openness, Jiang Zhou noted that industrial assembly lines and instrumentation are relatively mature, with data accuracy and validity typically superior to early-stage embodied intelligence materials. Additionally, the RoboGo™ platform supports private deployment and has already been implemented for data collection companies such as Fanyu Zhiyuan, enabling data processing and model training within private enterprise environments to meet the security and compliance requirements of keeping industrial data "within the factory."
The Computing Power Sweet Spot
In terms of chip hardware, D-Robotics currently offers a product lineup covering the full range of computing power from 5 to 560 TOPS. Specifically, the Sunrise S100 and S600 chips, designed for industrial and embodied intelligence scenarios, have achieved mass delivery.
Addressing market concerns about computing power bottlenecks, Jiang Zhou stated frankly that computing power is currently insufficient across the entire industry, and the rapid iteration of embodied models poses a persistent challenge to edge-side computing power. However, beyond providing hard TOPS metrics, D-Robotics emphasizes the actual efficiency and accuracy of models running on edge chips, which relies on extensive adaptation engineering experience.
He noted that D-Robotics positions itself to "provide customers with more than twice the AI computing power on chips at the same price point." Taking the flagship high-computing-power platform, the Sunrise S600, as an example, it can stably run mainstream VLA (Vision-Language-Action) models in the industry at around 10 frames per second on the edge, ensuring smooth and stutter-free robot movements.
Compared to NVIDIA Thor, which has approximately twice the computing power of D-Robotics' flagship chip but commands a high unit price (over CNY 20,000 for the low configuration and nearly CNY 50,000 for the high configuration), most robot bodies delivered in bulk today sell for hundreds of thousands of RMB. Between electronic component Bill of Materials (BOM) costs and scenario constraints, the Sunrise S600 hits the "sweet spot" that encourages customers to actively choose it.
Data shows that cumulative shipments of the Sunrise series chips have exceeded 8 million units, and the mass production coverage rate among leading customers in the embodied intelligence field has surpassed 50%.
Adhering to the "Shovel Seller" Division of Labor
NVIDIA has built a deep moat in the AI chip domain relying on the CUDA ecosystem, but Jiang Zhou believes that as large models and Agent scheduling capabilities enhance, software-level barriers across the industry are rapidly lowering.
In the past, model optimization required engineers to deeply understand chip architecture and BPU operator structures. Now, through the democratization of intelligent toolchains, the threshold for model deployment has been drastically compressed. Details that once took months of one-on-one effort can now be resolved in one or two days, significantly reducing simple labor accumulation.
In its business model with ecosystem partners, D-Robotics has defined clear boundaries. Jiang Zhou positions D-Robotics as a "parent ecosystem company of computing infrastructure providers." D-Robotics is responsible for developing underlying technical capabilities, validating advanced algorithms, and creating "showroom" standards and "reinventing the wheel" efforts to bring algorithms to an 80-point level, without charging for these self-developed algorithms.
"D-Robotics does not intend to encroach on the ecological niches of our partners. We prefer to leave all profitable delivery businesses and last-mile scenario implementations to solution partners," Jiang Zhou emphasized. He added that the century-old niche know-how of traditional manufacturing resides with customers and solution providers, who deserve normal commercial returns for their close-proximity services.
Source: EEFOCUS
Overseas "Expedition": Exporting AI Capabilities with Chinese Solutions
As global attention to AI layout increases, overseas markets are becoming a significant source of incremental growth for D-Robotics. Jiang Zhou believes that in the edge AI segment, thanks to D-Robotics' accumulated background in large-scale mass production and engineering implementation in China, the company holds distinct technical and support advantages when facing overseas clients, demonstrating full technical confidence in international expansion.
Currently, D-Robotics' overseas strategy is divided into two tracks: developer communities and commercial implementation. In more than 20 countries worldwide, channels are available to purchase locally certified RDK robot developer kits. Offline Hackathons and joint activities with universities have also successively launched overseas.
In the overseas commercial process, Jiang Zhou observed an interesting phenomenon: many overseas clients initially recognize D-Robotics' complete AI capabilities and toolchain logic. However, due to China's stronger manufacturing foundation and comprehensive solution coordination capabilities, final contract signings often loop back to domestic partners with strong overseas expansion capabilities, such as Quectel, for execution.
In Conclusion
Looking ahead to industrial scenario planning for the next 3 to 5 years, D-Robotics has established a clear three-step strategy: First, persist in producing genuine "industrial-grade" chips that strictly meet national and industry standards, rather than screening consumer-grade chips as substitutes. Second, continuously expand the solution partner ecosystem, expecting over 100 industrial-grade solution partners by the end of this year and over 300 next year. Finally, leveraging the RoboGo™ platform and high-computing-power chips, select several specific large industrial segments to redo traditional automation with more cost-effective AI perception and computing solutions, helping industrial clients cross the hurdle of AI transformation with a lower barrier to entry.