Chinese tech powerhouse ByteDance is in the process of developing its own central processing units (CPUs) to meet the increasing demands of its AI infrastructure, according to three insiders, as escalating chip prices and ongoing supply shortages hinder its growth ambitions.
This initiative highlights the industry’s swift transition towards “inference,” where AI models are used for complex tasks that require enhanced CPU capabilities, operating alongside the graphics chips produced by Nvidia that have been pivotal to the AI surge.
Recently, this shift has resulted in a CPU shortage, prompting global hyperscalers such as Alphabet’s Google, Amazon, and Microsoft to create custom CPUs aimed at cutting costs and optimizing performance for their unique workloads. This trend has positioned major CPU manufacturers like Intel and AMD as key competitors to Nvidia’s AI supremacy.
ByteDance, the parent company of TikTok, aims to implement its proprietary CPU within its own servers and data centers to bolster internal operations, as it gears up for an expansive launch of agent-driven products, including its Coze platform, as indicated by the first source.
The Beijing-based firm has reached out to various external partners for support in this venture, with those partners expected to assist in both the chip design and securing manufacturing capabilities at foundries, the sources mentioned. The project is still in its early stages, the first source added.
They requested anonymity, as the plan has not been publicly disclosed.
ByteDance did not provide a comment in response to Reuters’ inquiry.
CPU SHORTAGE
ByteDance’s efforts align it with a growing number of tech companies that believe the benefits of custom chips outweigh the complexities involved in their design.
The company is exploring two architectural paths for its CPU development — one leveraging SoftBank-owned Arm and another utilizing the open-source RISC-V instruction set architecture, as it evaluates which design will best serve its future data center needs, the sources stated.
Simultaneously pursuing two designs is a typical strategy for major tech companies, enabling them to trial options before committing to an expensive, large-scale production run.
Arm has not yet responded to a request for comment.
The initiative to create proprietary silicon comes as Intel has cautioned Chinese clients about potential server CPU delivery delays of up to six months, as reported by Reuters in February.
Intel noted last month that demand for its CPUs from AI companies was so robust in the first quarter that it sold chips it had initially deemed unfit.
AMD CEO Lisa Su expressed concern last week that the overall CPU market is “tight,” with demand surpassing projections and supply constraints likely to continue.
Currently, ByteDance procures its CPUs from Intel and AMD, whose prices have significantly increased, with quarter-over-quarter hikes ranging from 10% to as much as 35% in recent months, according to two of the sources, which has prompted ByteDance to expedite its development of in-house alternatives.
