美国AI数据中心“叙事变化”：从“大交易”的兴奋感转向“不断延误”的“推锅大战”

Core Insights - The market sentiment in the U.S. AI data center sector is experiencing a significant reversal, with excitement from large-scale transactions and record contracts fading due to project delays and blame-shifting among stakeholders [1][2] - Supply chain tensions are beginning to materially impact corporate performance, as evidenced by CoreWeave's warning of a potential revenue hit of $100 million to $200 million due to delays from third-party developers [1][3] - The delays are not isolated to individual companies but reveal systemic bottlenecks within the industry, as GPU delivery speeds have outpaced facility construction, leading to idle hardware waiting for deployment [1][6] Project Delays and Accountability - CoreWeave has become a case study in the blame game, as CEO Mike Intrator indicated that revenue would be severely impacted due to delays attributed to third-party data center developers, speculated to be Core Scientific [3][4] - Microsoft previously reduced contracts with CoreWeave due to delays at a Texas data center, which is powered by Core Scientific, highlighting the interconnected nature of these delays [3][4] - Core Scientific's CEO Adam Sullivan criticized the unrealistic timelines for AI data centers and noted that public disclosures of delays can create market confusion and erode confidence [4] Financial Pressures and Contractual Obligations - The stakes in the AI compute race are high, with strict financial terms in contracts often including penalties for missed deadlines or operational failures, which can significantly impact the already thin margins of GPU cloud leasing businesses [5] - The urgency to assign blame for missed deadlines is heightened, as even minor delays can have substantial financial repercussions for companies committed to rapid delivery schedules [5] Hardware Backlog and Strategic Adjustments - The mismatch between hardware supply and facility readiness has led to a backlog of GPUs, with some companies forced to store idle units while awaiting deployment instructions [6] - Major tech companies are adjusting their strategies to build buffers, with Meta acknowledging the need for phased data center construction to quickly scale capacity when needed [6] - The increasing difficulty in securing power and other resources may lead more clients to diversify their partnerships with multiple data center providers to mitigate risks [6]