ByteDance’s Bold Move: Self-Developed AI Chips

Recently, ByteDance's AI video generation model, Seedance 2.0, achieved sudden global viral success, with overseas users flocking to experience it. Simultaneously, reports from Reuters and other media outlets revealed that ByteDance is accelerating the development of its own AI chips, with an annual production target of at least 100,000 units. These two developments—one focused on software and the other on hardware—outline ByteDance’s strategic ambition to build a full-stack AI ecosystem.

Technological Breakthrough: Seedance 2.0 Redefines Video Creation

The viral success of Seedance 2.0 stems from its revolutionary user experience. It supports four types of input—text, image, video, and audio—and can generate coherent multi-shot, multi-scene videos with native audio in a single click. This has fundamentally shifted the video creation paradigm away from a reliance on complex traditional editing.

The core breakthrough lies in its powerful multimodal understanding and consistency retention. By providing only a small amount of reference material, the model can precisely replicate character traits, visual styles, and cinematic language. This solves a persistent "pain point" in previous AI videos: "morphing" characters and disjointed scenes.

Market reaction has validated its worth. Following the release, the media sector in the A-share market surged. This signifies that AI video is evolving from a mere "tech demo" into a stage of large-scale commercial application, bringing a genuine productivity revolution to the content industry.

Computational Autonomy: Self-Developed Chips to Combat Cost Pressures

Whether it is the stunning performance of Seedance 2.0 or the surge of the Doubao App’s monthly active users (MAU) to 227 million in Q4 last year, ByteDance is facing immense computational demands.

According to Reuters, sources familiar with the matter revealed that ByteDance is developing an AI chip and is in talks with Samsung Electronics for foundry services. Sources claim ByteDance aims to receive chip samples by late March. This chip is specifically designed for AI inference tasks, with a production plan of at least 100,000 units this year, and hopes to eventually scale production to 350,000 units.

One source noted that the negotiations with Samsung also include securing a steady supply of memory chips. Amid the global AI infrastructure boom, this "foundry + memory" integrated supply chain strategy underscores ByteDance’s determination to ensure computational autonomy.

Earlier, the Financial Times reported that ByteDance’s projected investment in AI-related fields for 2026 will reach approximately 160 billion RMB, with 85 billion RMB dedicated specifically to AI processors for developing models and applications. As the world’s largest short-video and AI application provider, ByteDance’s core businesses—including Douyin, TikTok, and Doubao—face exponential growth in demand for AI inference power.

Developing in-house chips has become an inevitable choice for ByteDance to reduce costs and guarantee computational security. Through customized design, these chips can be perfectly optimized for core scenarios such as video processing, recommendation algorithms, and large model inference, significantly improving the power-efficiency ratio.

As the competition in AI large models intensifies, computational autonomy has become a "must-answer" question for tech giants. With Google, Amazon, and Microsoft already deploying self-developed AI chips—and domestic peers like Alibaba and Baidu actively pushing similar projects—ByteDance’s entry is set to further intensify the rivalry in this arena.