Qualcomm Designs Custom Chips for ByteDance as CPU Lead Times Hit Six Months

Published by James Harris on June 24, 2026June 24, 2026

Qualcomm is in talks to design custom video processing chips for ByteDance, with mass production targeted for late 2026, according to anonymous sources cited by Reuters. The design work would draw on IP cores from AlphaWave Semi, the high-speed connectivity firm Qualcomm acquired in 2023 for $2.4 billion.

The timing is not incidental. ByteDance is under simultaneous pressure from multiple directions: CPU lead times from Intel reportedly stretch to six months for Chinese customers, AMD has described global CPU availability as very “tight,” and ByteDance’s own chip sourcing costs have risen between 10% and 35% quarter-over-quarter according to Reuters sources. The company is reportedly negotiating a $20 billion term loan to fund continued hardware purchases, which would be the largest financing package in its history. That scale of infrastructure spending explains why ByteDance is pursuing both internal chip development using Arm and RISC-V architectures and external design partnerships simultaneously. It is also why Beijing has pushed firms including ByteDance toward tighter control over their technology supply chains, making a Qualcomm partnership politically complicated even as it becomes operationally necessary.

Qualcomm already signed a deal in May to supply ByteDance with millions of ASICs for AI data center applications, so this is an extension of an existing commercial relationship, not a new bet.

The broader dynamic here mirrors what is happening across every major hyperscaler. Google, Amazon, Meta, and Microsoft have all built proprietary silicon to reduce per-token inference costs and limit exposure to constrained GPU supply, particularly from Nvidia. South Korean infrastructure buildouts illustrate exactly the kind of Nvidia dependency that custom silicon is designed to reduce over time. Qualcomm entering custom ASIC design for Chinese clients puts it in direct competition with Broadcom and Marvell, both of which already generate billions annually from comparable hyperscaler relationships. The difference is that Qualcomm is doing this while navigating a softening smartphone cycle and the ever-present risk that U.S. export controls could invalidate the business case entirely. The 2023 Commerce Department rules use compute density thresholds, specifically 4,800 TOPS, to define what falls under restriction, and any chips designed for ByteDance would need to be structured carefully around those limits.

Neither Qualcomm nor ByteDance has confirmed the negotiations, and the Reuters sources explicitly cautioned that no outcome is guaranteed. If the deal does proceed to production by late 2026, it would represent Qualcomm’s clearest signal yet that it intends to compete seriously in the custom data center silicon market rather than treat it as a secondary line of business.