Intense scrutiny of AI infrastructure economics is shaping trader views on when the bubble might burst. Hyperscalers face a widening gap between trillion-dollar data center capex and slower revenue growth, with inference prices halving since mid-2026 amid Chinese open-weight models like Qwen and GLM undercutting U.S. labs by up to 90%. Recent analyst notes from Panmure Liberum and Capital Economics flag a potential 2027 inflection as compute costs rise sharply while adoption metrics lag, echoed by Michael Burry's expanded shorts on Nvidia and peers. Ongoing frontier releases—Claude Opus 5.5, GPT-6 variants, and Gemini 4 Argon—demonstrate capability gains, yet safety warnings from Anthropic and OpenAI's IPO delay to 2027 add uncertainty. Key near-term catalysts include Q4 earnings, regulatory moves in China and the U.S., and any further price compression that could accelerate or defer a correction.