Taiwan’s MediaTek reportedly in final talks to design AI chips for ByteDance - Taiwan News

AI & Compute

MediaTek Reportedly in Final Talks to Design AI Chips for ByteDance

MediaTok is reportedly in final talks to design AI chips for TikTok parent ByteDance, a deal that would hedge Beijing's compute crunch and dent Nvidia's China demand.

By
Rebecca Stone
Filed
Channel
AI & Compute
Read
3 min read

Taiwan's MediaTek is in final-stage talks to design AI chips for ByteDance, the Beijing-based parent of TikTok, Taiwan News reports. If the deal closes, it would pair the world's largest smartphone SoC supplier with one of China's most aggressive AI buyers.

The report is thin on detail. What it establishes, though, is enough to redraw a few assumptions about who designs AI silicon for Chinese hyperscalers — and who doesn't.

What do we actually know?

The sourcing is explicit: MediaTek, the Hsinchu-based fabless designer whose Dimensity smartphone chips ship in hundreds of millions of handsets annually, is reportedly near an agreement to design custom AI chips for ByteDance. The talks are described as being in their final stages.

That's the confirmed core. No deal value, node target, or shipment volume has surfaced. Everything beyond the headline rests on inference, and this piece treats it that way.

Why would ByteDance want MediaTek silicon?

ByteDance runs one of the largest recommendation and video workloads on the planet. TikTok, Douyin and its advertising engine burn through compute. Training and inference demand keeps climbing, and U.S. export controls have made top-end Nvidia accelerators harder to obtain in China.

Custom silicon is the classic hedge. Google has its TPU line. Amazon builds Graviton and Trainium. Meta and Microsoft have both explored in-house designs. For ByteDance, a domestic-route accelerator designed with a proven partner like MediaTek would reduce exposure to Washington's licensing regime.

MediaTek brings something Chinese buyers can't easily get elsewhere: world-class design capability at a Taiwanese fab-facing entity, with deep experience shipping high-volume SoCs on TSMC's leading-edge nodes.

What's in it for MediaTek?

MediaTok has spent years trying to diversify beyond smartphones. Its Dimensity 9000-series flagship SoCs compete against Qualcomm at the top of the Android market, but handset chips carry thin margins and cyclical demand.

AI accelerators and custom design work offer a different profile. Design services revenue is sticky. It cushions the company through phone-market downturns. A ByteDance win would also burnish MediaTek's credentials for other hyperscaler contracts — a market where it has publicly signaled ambition.

The timing matters. Global AI chip demand has outrun supply since early 2024, and every major cloud operator is hunting for alternatives to a single-vendor dependency.

How solid is this?

Not very, by deal standards. The word "reportedly" carries weight. Final talks can collapse over pricing, export-control risk, or U.S. political pressure — Washington has already shown willingness to scrutinize any arrangement that funnels advanced AI capability toward Chinese firms.

There's also a design-scope question the report doesn't answer. "Design AI chips" could mean a full custom accelerator, an ASIC design-services engagement, or a co-developed part. Each implies a different revenue scale for MediaTek and a different capability transfer for ByteDance.

Who else is affected?

Nvidia, most obviously. ByteDance has been one of the largest buyers of Nvidia's China-market AI chips, and any shift toward custom silicon chips at that demand. Qualcomm loses too — it has courted Chinese cloud customers for custom silicon work of its own, and MediaTek would be taking a seat at that table first.

For TSMC, a MediaTek-designed AI chip would almost certainly land in its fabs, adding to an AI-related order book that already spans Nvidia, AMD and the major cloud in-house designs.

Nothing is signed. But if MediaTek and ByteDance close this deal, expect the custom-AI-silicon race to shift from U.S. hyperscalers to a genuinely global contest — with export-control politics deciding how far each side can push.

Source: Google News: AI chips

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering technology and product news in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

564 articles

Related articles

« Previous articleNext article »