ByteDance Building Its Own AI Chip With Groq's Playbook
TikTok's parent is teaming up with chipmaker InnoStar to build a low-cost AI inference chip inspired by Groq's architecture.
ByteDance isn't just running AI models — it wants to build the silicon that powers them. The TikTok parent company has partnered with chipmaker InnoStar to develop a custom AI inference chip, according to The Information.
The chip's design takes direct inspiration from Groq's Language Processing Units (LPUs), purpose-built hardware designed to run AI models cheaply and efficiently. It's a deliberate move away from the GPU-dominated status quo.
The project is part of ByteDance's broader push to build out its own AI infrastructure from the ground up. Designing custom silicon gives the company more control over costs and performance as it scales AI across its massive product ecosystem.
For ByteDance, the calculus is simple: if you're running inference at TikTok scale, owning your own chip architecture could save a fortune.