ByteDance has approached several external partners to design custom AI chips, PCGamer reported on May 29, citing supply chain constraints as the driver. The TikTok parent company joins Alibaba, DeepSeek, and now Mistral on the list of large AI infrastructure operators moving to design their own silicon rather than wait on the open market.
The driver is operational. Current AI chip supply, particularly for Nvidia’s most advanced GPUs, has multi-month lead times that constrain how fast a buyer can scale capacity. For a company running AI inference at TikTok’s volume, that constraint is a binding limit on product velocity and a structural exposure on costs.
The strategic shape is becoming familiar. Hyperscalers and large model deployers are not waiting for Nvidia’s roadmap; they are building parallel silicon programs that target specific workloads. Amazon has Trainium, Google has TPU, Microsoft has Maia, Alibaba has T-Head, DeepSeek is reportedly working on something similar. ByteDance joining that list confirms the trajectory: AI compute is bifurcating between general-purpose Nvidia silicon and workload-specific custom silicon owned by the largest deployers.
Reported by PC Gamer on 2026-05-29.
