Qualcomm courts ByteDance for custom VPU chips amid data center pivot

Qualcomm's push to design custom VPU chips for ByteDance signals a concrete step into the custom ASIC services market currently dominated by Broadcom and Marvell. A signed deal would give Qualcomm its first flagship data center customer for this division, validating the AlphaWave Semi acquisition as more than a portfolio add-on. For streaming infrastructure, custom VPUs could reduce power consumption and improve encoding efficiency for ByteDance's massive video workloads at TikTok scale. Watch whether ByteDance commits to Qualcomm or opts for its parallel domestic suppliers — Iluvatar CoreX is expected to ship at least 50,000 inference chips to ByteDance this year, per Reuters.

Qualcomm's negotiations with ByteDance represent its most visible attempt to enter the custom ASIC design-services market, a segment currently dominated by Broadcom and Marvell. On its Q2 FY2026 earnings call, CEO Cristiano Amon stated that a leading hyperscaler custom silicon engagement is on track for initial shipments later this calendar year, confirming that the ByteDance talks are part of a broader data center strategy rather than a one-off deal. Qualcomm's handset revenue fell 13% year over year to $6.02 billion in that same quarter, while automotive revenue climbed 38% to a record $1.33 billion, underscoring the urgency of diversifying away from smartphone dependency.

ByteDance's willingness to engage Qualcomm sits within a larger capital expenditure program. ByteDance is in talks with several banks for a loan of approximately $20 billion, its largest offshore borrowing on record, with an initial three-year term and an option to extend to five years. The company's capex plan for 2026 is set at $70 billion for data center and AI infrastructure expansion, with potential growth to $100 billion in 2027 if conditions permit. That spending envelope gives ByteDance leverage to run parallel chip-design engagements with multiple partners simultaneously.

The competitive landscape for custom silicon design services remains concentrated. Broadcom and Marvell currently dominate the market of turning hyperscaler chip concepts into working silicon, and Qualcomm's entry adds a third merchant option for companies seeking alternatives. ByteDance itself is developing two CPU architecture lines concurrently, one on Arm and one on RISC-V, and is working to move custom processors into its servers for internal workloads including its AI-agent platform Coze, according to AI Weekly's analysis of the negotiations. The regulatory dimension adds complexity: design services occupy different legal territory than shipping finished advanced hardware under US export controls, but any deal must still stay within American legal ceilings on AI chip computing power for Chinese customers.