The three-year compute payback estimate is the number I’d keep tracking. It converts “huge AI capex” into an ROIC question. If utilization rises and more workloads move onto Alibaba’s own silicon, the economics can improve quickly. But if inference pricing compresses faster than infrastructure costs, owning more of the stack concentrates the downside as well as the upside.
The three-year compute payback estimate is the number I’d keep tracking. It converts “huge AI capex” into an ROIC question. If utilization rises and more workloads move onto Alibaba’s own silicon, the economics can improve quickly. But if inference pricing compresses faster than infrastructure costs, owning more of the stack concentrates the downside as well as the upside.
Really enjoyed this read!!!