The AI bubble could burst when supply and capital spending outpace demand that can be monetised, rather than from technology failure, an analysis said. Persistent bottlenecks in memory, advanced packaging, networks, power and site readiness may delay a downturn. Strong semiconductor sales and AI-related revenue underscore demand, but rising memory prices may be lifting revenue faster than deployments. Oracle, OpenAI and the Stargate project were cited as showing a gap between commitments and realised revenue, with 2028-2029 flagged as a high-risk period.