ByteDance, the Chinese conglomerate behind TikTok, has partnered with Broadcom, a U.S. semiconductor firm, to create an innovative artificial intelligence (AI) chip.
This partnership ensures a consistent supply of high-performance chips for ByteDance’s AI endeavors amidst ongoing U.S.-China tensions.
Strategic development of AI hardware
The new collaboration focuses on developing a 5-nanometer application-specific integrated circuit (ASIC) that conforms to U.S. export regulations. This move is significant as it marks one of the first major initiatives to develop a cutting-edge 5nm chip since the U.S. imposed stringent export controls on advanced semiconductors to China in 2022.
Reuters cites insiders who reveal that Taiwanese giant TSMC is set to manufacture the chips, although production is expected to commence later this year. ByteDance, primarily recognized for its popular apps TikTok and Douyin, is venturing further into generative AI.
The Beijing-based company is reportedly shifting from reliance on Nvidia’s A100 graphics processing units, used for running large language models (LLMs), to developing its proprietary chip technology. This strategic pivot is seen as a way to mitigate the impact of U.S. sanctions, which have restricted the export of high-end chips to China, and to lessen its dependency on Nvidia’s GPUs, which are currently scarce due to high global demand and U.S. export limitations.
Expanding capabilities in AI technology
The partnership between ByteDance and Broadcom is not new; it dates back to 2022 when ByteDance acquired Broadcom’s Tomahawk 5nm high-performance switch chip and its Bailly switch for AI computer clusters. With the planned AI chip, ByteDance aims to enhance the power of its algorithms significantly, thereby advancing its AI product capabilities.
This development is particularly crucial as ByteDance seeks to secure more efficient and cost-effective solutions for its hardware needs. This will enable the firm to maintain a stable chip supply for its growing range of AI projects. Such advancements are vital in an era when AI applications are becoming increasingly prevalent and require substantial computational power.
Global demand for AI chips skyrockets
The demand for AI processors has surged as companies like OpenAI and Microsoft intensify their use of LLMs for applications such as ChatGPT, which require extensive computational resources to train and operate. This rising demand has led to a notable shortage of AI chips, pushing the prices upward, with Nvidia’s A100 GPU reportedly priced at around $15,000 each.
A 2020 TrendForce report highlighted the intense resource needs of OpenAI’s GPT models, which required approximately 20,000 of Nvidia’s GPUs for processing training data. As AI technologies like ChatGPT continue to evolve, the demand for such hardware is only expected to increase, making collaborations like that of ByteDance and Broadcom more significant in the tech industry.
This strategic move by ByteDance not only aims at self-reliance in terms of chip supply but also marks a significant step in the global AI technology race, highlighting the intersection of technology development and international trade policies.