AI Cash Flow Leaders

Build a 20-name, high-risk, 3–6 month AI infrastructure book: four bottlenecks — chips, memory, optics, power — plus the hyperscalers that buy them. Universe is US-listed names with current AI revenue and public capex: memory (MU, SNDK, DRAM), foundry/WFE (TSM, TSEM, ASML, AMAT, LRCX, KLAC, TER), compute (NVDA, AMD, AVGO), optics (LITE, COHR), power (GEV, VRT, SEI), hyperscalers (GOOGL, AMZN). Buy the full sleeve immediately. Once fully built, consolidate into memory ahead of earnings: fatten MU toward 25%, add DRAM at ~15%, keep SNDK around 9%, and cut WDC, STX, MRVL, CAT to zero. Everyone else stays a flat 3% satellite. Then stop trading. Do not weekly-rebalance. Do not add crypto, defense, or Korea-defense names. Let the memory core drift; only rotate if a new bottleneck starts printing the same shortage tape. Deploy immediately: MU 25%, DRAM 15%, SNDK 9%, and 3% each TSM, TSEM, ASML, AMAT, LRCX, KLAC, TER, NVDA, AMD, AVGO, LITE, COHR, GEV, VRT, SEI, GOOGL, AMZN.