Hyperscalers face a massive capital expenditure challenge for AI infrastructure, with current incremental AI revenue of $86-188B insufficient to justify the $13-19 trillion needed over the next decade. Memory chips, particularly DRAM from limited suppliers like Micron, represent 30-40% of hyperscaler capex and are a critical AI bottleneck, while off-balance-sheet financing structures like Meta's $30B Hyperion deal are increasingly moving AI infrastructure debt into retail investments with significant credit risk.