One number from @jpmorgan Asset Management's latest AI outlook caught my attention. Major hyperscalers are expected to spend around 9️⃣3️⃣% of operating cash flow on capex in 2026, up from roughly 33% in 2023. The implication is, investors will increasingly demand evidence that higher AI spending can translate into revenue, margins and free cash flow. But I think there is another layer to this: 👉 AI capex is starting to look like a prisoner’s dilemma. Each hyperscaler may recognize the risk of overbuilding. But no company wants to be the first to slow down, because falling behind in compute capacity, models or cloud infrastructure could carry an even greater long-term cost. So some of this spending may not be driven purely by expected returns. It may also be defensive capex, where companies keep investing because the strategic cost of not investing feels too high. That creates an unusual setup: AI demand can remain strong, while industry-wide returns still disappoint if everyone builds similar capacity at the same time. In that sense, AI capex is no longer just about technology. It is becoming a broader capital allocation story, affecting liquidity, credit markets, energy demand, commodities and risk assets including crypto. The next phase of the AI trade will be less about who spends the most, and more about who can turn that spending into sustainable revenue and free cash flow. here's the full report 👇