On the shores of Shenzhen Bay, an industrial chain reaction triggered by open source is taking place.
At the Guangdong Provincial Artificial Intelligence Application Docking Conference, Chen Mai, founder and CEO of Shenzhen-based Shifang Ronghai, shared the topic of “Xiaozhi AI Open Source Ecosystem Empowering Intelligent Terminals”, which attracted the attention of the audience.
At the booth of Shenzhen Futian Convention and Exhibition Center, a seemingly ordinary plush toy was connected to an open source framework. In less than half an hour, the toy began to naturally talk to people – it could listen, speak, recognize people, and remember where you talked last time.
This ability to allow traditional toys to talk to each other brought Shifang Ronghai a cooperation contract worth over 40 million yuan in the next two months.
This points to an underlying pain point that has long plagued the hardware industry: the general large model on the software side is getting stronger and stronger, but the access cost on the hardware side is extremely high and the development cycle is long. There is a huge “fracture zone” between the two.
Shifang Ronghai has chosen a pragmatic path with more “Guangdong characteristics” – relying on Guangdong’s complete hardware supply chain advantages to allow Xiaozhi AI to “build roads and bridges” in the silicon-based world.
The value of this “road construction” has also been verified at the data level.
Shifang Ronghai founder and CEO Chen Mai gave a set of data: Xiaozhi AI has received more than 26,000 stars on GitHub, more than 100,000 developers have participated in the co-construction, more than 1.3 million access devices have been connected, the number of daily conversations is as high as 9 million, and the average daily Tokens consumption exceeds 120 billion.
Through open source protocols and standardized chip-level reference designs, Xiaozhi AI has significantly compressed the product verification cycle from concept to prototype from the 3 to 6 months generally required in the industry to 1 to 2 months; the average computing power cost of hardware products has been reduced by approximately 80%, and R&D efficiency has been increased by more than 70%.
This extreme cost reduction and efficiency increase directly activates the manufacturing vitality of the Greater Bay Area.
This efficient closed loop of “open source technology – hardware adaptation – mass production” allows traditional factories that originally lacked AI R&D capabilities to launch intelligent products in a short period of time.
At Shenzhen Mingyuan Experimental School, an AI learning terminal called “Moben Knowledge Card” is being piloted; in the field of elderly care, Shifang Ronghai’s cultural literacy smart service platform has launched the T10 PRO acoustic learning machine.
When the AI open source technology ecosystem meets the manufacturing heritage of Guangdong’s “half-hour supporting circle”, an explosion of terminal intelligence has just begun.
The outbreak of the Ten Directions Fusion Sea was no accident.
Looking at the history of this company, from winning the national gold medal in China’s “Internet +” College Student Innovation and Entrepreneurship Competition in 2017 to becoming the world’s number one open source project on GitHub in 2026, behind this is Guangdong’s strong support for continuing to optimize the business environment.
Guangdong Province requires that the scale of the province’s core artificial intelligence industry exceed 300 billion yuan in 2025, accounting for about a quarter of the country.
Combined with Guangdong’s 10 trillion-level industrial clusters and tens of thousands of industrial enterprises above designated size, these industrial foundations will eventually be transformed into a natural testing ground for the implementation of AI.
As AI is further implemented in smart homes, robots and other scenarios, this “open source technology + Made in Guangdong” model will promote the intelligent upgrading of the manufacturing industry in Guangdong and contribute more replicable enterprise samples.
