Reports Suggest ByteDance Developing Its Own CPU Chips to Accelerate AI Infrastructure

According to Reuters, ByteDance is developing custom CPU chips to meet the computational needs arising from the expansion of AI infrastructure. Three informed sources revealed that the rise in chip prices and ongoing supply chain tensions are key factors driving this decision.

This move reflects the profound changes underway in the AI industry. As AI models shift from the training phase to large-scale inference deployment, demand for CPUs has significantly increased. ByteDance's choice to develop CPUs in parallel with GPUs indicates that its AI strategy is extending to more fundamental computing autonomy.

ByteDance plans to deploy its self-developed CPUs in its own data centers and servers to support internal operations. Additionally, the company is advancing product iterations based on intelligent agents, covering multiple business lines including TikTok and cross-border e-commerce platforms.

Parallel Advancement of Dual Routes

Sources with knowledge of the matter have disclosed that ByteDance is currently advancing two chip architecture routes simultaneously: one based on the SoftBank-customized Arm architecture and the other based on the open-source RISC-V architecture. Each route has its focus, with the Arm solution being more mature and the RISC-V solution offering greater advantages in controlling long-term data center costs.

The simultaneous advancement of both architectures is a common hedging strategy among tech giants, facilitating a comprehensive assessment of their respective performance and cost performance before mass production.

ByteDance has approached several external partners seeking collaboration in chip design and production capacity. According to these sources, the partners are involved not only in the design phase but will also help address manufacturing capacity issues. The project is still in its early stages, and no specific plans have been disclosed yet.

Changes in the CPU Market's Supply and Demand Landscape

ByteDance is not alone. In recent months, there has been a noticeable supply crunch in the global server CPU market. Hyperscale cloud giants like Google, Amazon, and Microsoft have accelerated the development of their own chips, objectively contributing to the recovery of Intel and AMD's market shares. This also poses a certain dispersion effect on NVIDIA's absolute dominance in the data center field.

From a market price perspective, Intel has recently raised the prices of some of its products, stating that the delivery cycle for server CPUs to Chinese clients once reached as long as six months. AMD's CEO, Dr. Lisa Su, also mentioned last week that the global demand for CPUs is strong and expected the supply tightness to continue.

On NVIDIA's front, its business map is rapidly expanding into the CPU sector. CEO Jensen Huang previously revealed that NVIDIA is targeting a potential market size of $200 billion. Recently, NVIDIA also released its new generation of processors and AI systems based on the Arm architecture and has partnered with Taiwan's Foxconn Group in manufacturing to further solidify its layout across the AI chip industry chain.

ByteDance currently mainly purchases CPUs from Intel and AMD, both of which have shown a significant upward pricing trend in recent times, with quarter-on-quarter price increases ranging from 10% to 35%. The continuous increase in supply costs is considered a significant catalytic factor in accelerating ByteDance's plan to develop its own products.