ByteDance is moving from “buying AI horsepower” to trying to own part of it. The TikTok parent is reportedly developing an in-house AI chip and is in talks with Samsung Electronics about manufacturing it — a notable shift as global AI demand tightens supply and geopolitics keeps advanced chips under pressure.
A chip aimed at inference, not training
The reported design focus is AI inference — the everyday work of running models (chatbots, recommendations, search, ad ranking) at scale. That’s where cost control and efficiency matter most, because inference is the “always on” bill that grows as users grow.
The timeline: samples first, then volume
The plan being discussed is to receive sample chips by late March, with production starting at at least 100,000 units this year and potentially scaling higher if testing and yields cooperate. It’s an ambitious ramp in a world where leading-edge capacity is scarce and manufacturing relationships take time to lock in.
Why Samsung is attractive: memory access is the prize
The talks reportedly include a key sweetener: access to memory chip supply, which has become a major constraint as AI infrastructure buildouts accelerate. For large AI deployments, compute isn’t the only bottleneck — memory bandwidth and availability can decide whether a rollout is smooth or stalled.
ByteDance is still buying Nvidia — a lot of it
Even with an in-house chip effort, ByteDance is expected to remain a massive buyer of Nvidia hardware. Reports suggest its 2026 AI-related procurement budget could exceed 160 billion yuan (about $22 billion), with more than half still going toward purchasing Nvidia chips while the internal chip program matures.
The bigger picture: China’s AI stack is going vertical
ByteDance wouldn’t be alone. Chinese tech rivals have been pushing hard into custom silicon as export controls and supply risks grow. The logic is simple: if you can’t guarantee access to the best external chips forever, you build what you can — then optimize your software around it.
What to watch next
If this effort is real, the next milestones are straightforward: confirmation of the chip’s specs, whether Samsung becomes the manufacturing partner, and whether ByteDance can turn a “chip project” into a reliable supply line. Because in AI, the companies that win aren’t just the ones with the best models — they’re the ones who can secure compute, memory, and cost-per-query at scale.
