China Mobile's US$75 billion 2027 procurement plan signals a seismic shift towards AI computing

China Mobile's 100-billion-yuan AI compute allocation is not a procurement event—it is a demand-side forcing function. The two-year doubling target will trigger a production arms race across HBM, advanced packaging, and liquid cooling. Under sustained US export restrictions, I estimate 60%+ of this spend will flow to Ascend and Cambricon, compressing NVIDIA's enterprise share in China below 30% within 18 months. The gigawatt-scale data center network is the real strategic anchor: it elevates compute competition from the silicon layer to the energy layer. Over the next 24 months, power grid coordination and domestic HBM (CXMT) will become the binding constraints. China Telecom and Unicom will almost certainly mirror this scale, creating a combined 200-billion-yuan AI infrastructure demand pool—enough to keep SMIC's 7nm lines at full utilization and force accelerated EDA localization. For global semiconductors, this is not one customer's order book. It is the formation signal of a parallel ecosystem operating on its own supply chain logic, its own performance benchmarks, and its own capital allocation cycle. The decoupling is no longer theoretical; it is now priced into capex.