The market’s biggest headwind in the world of AI is no longer whether hyperscalers can fund their investments, but rather whether ever-rising capex will continue to justify premium valuations across AI infrastructure and semiconductor stocks. While some of these concerns have spilled into credit markets, JPM estimates that hyperscalers still have significant debt issuance capacity – potentially around $50bn. However, the bank argues that funding itself may not be the constraint.

Using JPM’s analysis, we can estimate the revenue required to justify AI model providers’ investments. Assuming $6tn of cumulative AI capex through 2030, AI model providers would need roughly $1.8tn of annual revenue by 2030, equivalent to around 8.5% of projected S&P 500 revenues. While this may seem ambitious, it’s not entirely implausible.

The debate is therefore shifting from whether AI can be financed to whether the returns on that investment will ultimately justify the spending. As JPM notes, “the market’s willingness to pay for those returns is the key issue.” In other words, the real challenge lies in demonstrating that the expected returns from AI investments are sufficient to justify their high valuations.