Exclusive-ByteDance, Alibaba, Tencent place over $35 billion worth of Nvidia chip orders - The Lufkin Daily News

A $35 billion GPU procurement by China's top three internet firms is not merely an AI capex cycle—it is a strategic window-closing maneuver ahead of tightening US export controls. Technical ripple: The order almost certainly targets H100/H200-class silicon, directly stressing TSMC's (Taiwan, China) CoWoS packaging capacity and SK Hynix's HBM3E output. More critically, it deepens CUDA ecosystem lock-in. When all three anchor training clusters on Nvidia architecture, the PyTorch-CUDA developer flywheel accelerates, compressing the catch-up window for AMD MI300X and Huawei Ascend 910C. Compliance risk: The central tension is BIS export enforcement. If A100/H100 units are involved, delivery likely routes through Singapore or Malaysia, adding 15-30% premium and extended lead times that erode marginal ROI. Single-point supply chain dependency—US design, Taiwan, China fabrication, Japanese photoresist, Korean memory—creates systemic fragility. Market dynamics: AMD's China-revenue thesis weakens as customers vote with capital for CUDA. Huawei Ascend gains a paradoxical tailwind: the hoarding behavior validates the urgency of domestic substitution, likely accelerating 910C enterprise deployments in 2025. 12-24 month outlook: This is likely the last major tranche. Controls will escalate from banning sales to banning sales plus support. Nvidia's China revenue share may compress from roughly 20% to single digits. The true long-tail effect is not silicon but electrons—power infrastructure for this GPU fleet will force accelerated green-energy buildout in western China. The competition shifts from can you buy it to can you power it.