ByteDance shifts to in-house AI chips: rollout to begin next year

ByteDance, the parent company of TikTok, is preparing to transition to proprietary AI chips. According to industry sources, they will appear in the company's production computing infrastructure as early as next year.

Qualcomm Reveals Its Hand

At a recent quarterly event, Qualcomm for the first time openly confirmed key details about its customers. The company stated that it supplies finished chips to Meta — the name was obscured as M**a in official slides, but the context left no doubt. Additionally, Qualcomm announced that it is developing custom chips for two other major cloud companies, though it did not name them publicly. Industry sources have long discussed the possibility of a Qualcomm partnership with ByteDance — now this story has taken concrete shape.

Qualcomm positions itself as a manufacturer not only of consumer mobile processors, but also of custom AI accelerators for major platforms. This is a strategic pivot: previously, the company was primarily associated with Snapdragon for smartphones, but now it is actively entering the enterprise segment.

Qualcomm's involvement in the project, it appears, does not mean a full transfer of intellectual property. Rather, it likely involves joint design tailored to ByteDance's workloads and subsequent manufacturing at third-party fab facilities — presumably TSMC or Samsung.

Why ByteDance Needs Its Own Chip

For ByteDance, this transition is not a technological whim, but a forced strategic move. Several reasons drive it:

• Geopolitical protection. U.S. export restrictions have already cut Chinese companies off from the latest NVIDIA GPUs. Proprietary chips offer greater control over the supply chain.
• Workload optimization. TikTok's recommendation algorithm and language model inference have requirements that a universal GPU serves inefficiently.
• Economy of scale. At ByteDance's volumes — billions of requests daily — even small reductions in per-operation costs yield enormous annual savings.
• Control over the roadmap. A proprietary chip allows hardware to be synchronized with software priorities without waiting for NVIDIA or AMD release cycles.
• Reduced sanctions risk. In the context of a trade war, supply chain independence becomes critically important.

China Builds Chip Independence

ByteDance is not a pioneer in this direction. Baidu, Alibaba, and Huawei have long been investing in proprietary AI accelerators. Huawei's Ascend chips are already replacing NVIDIA products in some Chinese data centers, while Alibaba is developing its Yitian series for its cloud infrastructure. The Chinese government actively subsidizes chip design: Biren Technology, Cambricon, and Moore Threads receive state support.

Notably, even in collaboration with American Qualcomm, ByteDance operates in a regulatory gray zone. U.S. export regulations govern the transfer of certain technologies, and every such deal requires careful legal structuring.

For comparison: the largest Western platforms — Amazon with Trainium and Inferentia, Google with TPU, Microsoft with Maia — have long been pursuing vertical integration of their hardware stacks. ByteDance follows the same logic, only in a significantly more complex geopolitical context.

What This Means

ByteDance's move to proprietary AI chips signals that the era when companies could simply buy a rack of GPUs is coming to an end. For Qualcomm, the situation is mixed: large orders now strengthen its position in the corporate AI segment, but a client that has built a chip with its help, in the long term, gains incentive to bypass the intermediary.

*Meta is recognized as an extremist organization and is banned in Russia.