A 1.6 trillion parameter model called LongCat-2.0 was trained entirely on Chinese silicon. Meituan bypassed Nvidia hardware to prove domestic chip viability at scale. This development challenges the assumption that US export controls can halt massive model training. It provides a blueprint for other Chinese firms to scale without Western GPUs.