In the new round of AI competition in the technology industry, ByteDance is consolidating its leading position in the AI field through a series of strategic layouts. The latest news shows that in order to reduce procurement costs and ensure stable supply of high-end AI chips, ByteDance is collaborating with the US semiconductor giant Broadcom to jointly develop an advanced 5nm ASIC (Application-Specific Integrated Circuit) AI processor.
According to insiders, this AI processor will comply with the latest US export control regulations, and the manufacturing work will be outsourced to the world's leading semiconductor manufacturer, TSMC. Currently, chip design work is progressing smoothly, but it has not entered the "tape-out" (trial production) stage, and TSMC is not expected to produce this chip this year.
ASIC, as a type of specialized chip for specific application scenarios, achieves maximum performance and energy efficiency through algorithm hardening, and has the characteristics of strong average performance, low power consumption, and high cost-effectiveness. However, ASIC requires large upfront investment and has a long development time, which reflects ByteDance's determination and long-term vision in the AI field.
As of the time of writing, ByteDance and Broadcom have not responded to this news, while TSMC has declined to comment. However, judging from ByteDance's layout in the AI field in recent years, this collaboration is not groundless.
Over the past year, with the rapid development of AI large model technology, ByteDance has caught up comprehensively and has made a comprehensive layout from the model layer to the application layer of AI large models. Among them, ByteDance has successfully launched multiple large language models and multimodal large models, which have been widely used on platforms such as Douyin. At the same time, ByteDance has also established a new AI department called Flow and launched multiple AI dialogue products such as Dou Bao, Kou Zi, and Cici, providing users with more intelligent and convenient services.
In 2024, ByteDance further accelerated its product development speed and launched multiple new products in the field of AI large models, including AI learning partner Hippo Love Learning, AI interactive drama product Cat Box, and AI image generation product PicPci. At the same time, ByteDance has actively laid out in the AI hardware field, launching new product lines such as wearable AI devices and handheld AI devices.
In terms of AI computing power at the underlying level, ByteDance has also demonstrated strong technical strength. Its subsidiary, the Volcano Engine, has released its self-developed video codec chip, providing powerful computing power support for AI large models. Although Yang Zhenyuan, Vice President of ByteDance, has stated that the company has no general-purpose chip commercial plan, the collaboration with Broadcom to develop AI processors undoubtedly further demonstrates ByteDance's determination and strength in the AI computing power field.
It is worth mentioning that the collaboration between ByteDance and Broadcom is not the first time. According to public information, ByteDance and Broadcom have been business partners since at least 2022. Broadcom stated in a public statement that ByteDance has purchased Broadcom's Tomahawk 5nm high-performance switch chips and Bailly switches for AI computer clusters. This collaboration to develop AI processors will further deepen the cooperation between the two parties and provide strong support for ByteDance's future development in the AI field.
