NVIDIA's earnings call never says 'LPU'; TSMC's 244-page filing doesn't either. The analyst claim that Samsung scored 'first true inroads' making NVIDIA's LPU traces to a March 2026 conference: Groq's third-gen inference chip, Samsung-fabbed, absorbed into NVIDIA's stack as a low-latency accelerator. If that link holds — still a strong link, not confirmed — the win is a single co-opted accelerator, not a GPU or leading-edge logic slot. TSMC's CEO noted they're working with customers on next-gen LPUs anyway. Watch for broadening beyond one chip.