ByteDance Eyes 100,000 Samsung-Made AI Chips, Secures Scarce Memory Supply
ByteDance is in talks with Samsung to manufacture 100,000 AI inference chips by year-end with plans to scale production to 350,000 units, and expects to receive samples by late March. The company is also negotiating memory chip supply deals to secure scarce HBM capacity as global AI infrastructure demand climbs, potentially reducing reliance on Nvidia GPUs.
1. Manufacturing Partnership and Timeline
ByteDance has opened negotiations with Samsung Electronics to produce an initial 100,000 AI inference chips, with sample deliveries expected by the end of March. The company plans to ramp output to 350,000 units over the course of the year, marking its first major step into in-house AI processor manufacturing.
2. Memory Chip Supply Negotiations
Alongside chip fabrication, ByteDance is securing commitments for high-bandwidth memory supplies to meet surging demand in AI data centers. Access to scarce HBM capacity will be critical for the performance of its custom AI processors and for reducing exposure to market shortages.
3. Implications for Nvidia GPU Sales
By developing its own AI inference chips and locking in memory sources, ByteDance aims to lessen dependence on Nvidia GPUs. This shift could dampen growth in Nvidia’s Chinese hardware revenue if more domestic firms replicate ByteDance’s strategy.