The TikTok parent will buy millions of Qualcomm application-specific chips for AI data centres and use Qualcomm to take its own design from blueprint to production.
Qualcomm has struck a deal to supply ByteDance with millions of application-specific integrated circuits, the chip designer’s most prominent commitment yet to compete in the AI data-centre market it has spent the past two years trying to enter, according to a Bloomberg report on Tuesday.
ByteDance, the Chinese parent of TikTok, will use the Qualcomm ASICs to underpin its AI-agent software, the people familiar with the agreement said. Qualcomm’s shares closed up around 5% on the news, with intraday gains touching above 8%.
The agreement has two distinct components. The first is a straight ASIC-supply arrangement, with ByteDance committing to volumes large enough to make it Qualcomm’s earliest publicly named major buyer of AI-focused chips.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
The second, equally important, is a chip-manufacturing-services layer: Qualcomm will help ByteDance bring an internally designed semiconductor that the Chinese firm has already completed to volume production. ByteDance has, in effect, hired Qualcomm as both vendor and manufacturing partner.
The structure matters because it works around US export controls. Qualcomm’s ASICs, on current public reporting, sit within the legal computing-performance thresholds that the US Commerce Department has set for chip exports to Chinese firms. ByteDance’s own in-house design is being engineered to stay inside the same thresholds.
The deal is therefore commercially significant but politically defensible in a way that direct sales of frontier Nvidia GPUs into China are not.
The strategic context for Qualcomm is the harder one. The company has historically derived most of its revenue from smartphone modems and Snapdragon application processors. The AI-data-centre market is currently structured around Nvidia’s GPUs and a handful of custom ASIC programmes run by hyperscalers like Google (TPU), Amazon (Trainium), Meta (MTIA) and Microsoft (Maia).
Breaking into that market as an outside supplier is harder than it sounds because the workloads, software stacks and packaging requirements are all idiosyncratic. ByteDance is among the largest non-hyperscaler customers Qualcomm could realistically have landed, and its agent-software workload is exactly the kind of inference-heavy use case Qualcomm’s architecture is positioned for.
For ByteDance, the deal lands inside a tightening Beijing-side environment. Chinese authorities have spent recent weeks restricting top AI talent’s ability to travel and have instructed major firms including ByteDance to reject US capital in funding rounds without prior clearance.
Buying inference silicon from a US designer at the legal threshold is the kind of move ByteDance can defend domestically as a workaround for what Beijing is willing to tolerate, while still depending on Western design IP for the actual compute layer.
The deal also fits into the broader pattern of Chinese AI firms looking for non-Nvidia compute sources. Jensen Huang himself has argued that Chinese AI labs running on alternative silicon is the structurally important shift to watch. Huawei has been the headline alternative; Qualcomm now becomes a parallel one.
Whether that further fragments the Chinese AI hardware stack or accelerates indigenous design and production is the question that will define how the rest of the Chinese AI industry buys hardware over the back half of 2026.
Qualcomm declined to comment on the specifics of the ByteDance arrangement. The deal is expected to ramp through 2026 and 2027.
