Under a scenario where, after the 2026 US midterm elections, the Democratic Party wins both chambers of Congress and also takes the governor seats in Texas and Ohio—states where data center construction is concentrated—the new governors introduce policies such as suspending approvals for new data centers and requiring projects to bear the full resource costs. Combined with existing local power grid interconnection bottlenecks, these policies cause about 5GW of US data center capacity originally planned to come online in 2028 to be delayed to 2029, accounting for 20% of the US newly added capacity that year and 10% of the global newly added capacity. Affected by this, US newly added data center capacity falls 4% year on year in 2028, while global newly added capacity grows 9% year on year; in 2029, the two rebound by 38% and 20% respectively, forming a gap period in the AI upstream supply chain in 2028, which may trigger problems such as weakened pricing power and inventory accumulation. Target prices or ratings of cloud semiconductor suppliers and related names in the legacy memory sector may be significantly cut, with the core of the downgrade lying in cycle inflection management rather than earnings collapse. The overall capital expenditure plans of leading cloud vendors remain unchanged, with only the construction pace misaligned; some projects may be shifted to other US states with more friendly policies, while another 20GW of potential spillover demand flows to Asia, Africa and Europe, and improvements in computing efficiency can also buffer capacity constraints. The market generally believes that AI demand has not disappeared, but the construction cycle has been lengthened; enterprises in different segments, locations and capital structures face differentiated risks, the scarcity of data center assets that have been completed and have power security rises, and the advantages of large cloud vendors will be further consolidated, with focus on large cloud vendors, selected data center supply chain names and opportunities related to AI software applications.