MANAGEMENT ARTICLE

How to Assess AI Investment Payback Periods: 4 Steps for Evaluating CAPEX ROI

When Alibaba released its June quarter earnings on August 20, 2026—reporting a 75% drop in net profit to 10.444 billion yuan alongside a 75% surge in CAPEX to 67.678 billion yuan—the exact same data yielded completely opposite interpretations. Stage 1 breaks investments down into three layers: infrastructure, application, and capability, noting that internally, Alibaba's AI Cloud segment saw EBITA surge 133%, while its AI Labs segment saw losses quadruple. Stage 2 decomposes payback effects into cost, revenue, speed, and risk, with a PwC survey showing top 20% companies capturing 74% of AI's economic value, a gap driven by a focus on growth. Stage 3 requires finance and operations to first agree on the useful life that serves as the denominator for the payback period, which GPU depreciation debates prove is a management judgment rather than a hard fact. Stage 4 locks in stop criteria and re-evaluation points at the time of investment approval, noting that escalating monthly inference and token operating costs—rather than initial deployment expenses—most frequently break payback calculations.