AI Picks & Shovels

This is the initial build-out of the 'sell the shovels' AI infrastructure portfolio, spreading capital across the four key layers of the AI buildout rather than betting on any single company. Memory (Micron/MU) and networking silicon (Marvell/MRVL) currently show by far the strongest price trends (huge 3- and 6-month gains driven by AI memory and custom-silicon demand), so they get the maximum allowed 25% each. Power/cooling infrastructure (Vertiv/VRT) is the third-strongest trend and a distinct, non-overlapping layer (data center power delivery), earning a 19% weight. Foundry manufacturing (TSMC/TSM) — the choke point that makes all AI chips possible — gets 14%. Networking/switching (Arista/ANET) adds another data-center layer at 9%. Broadcom (AVGO), which spans custom AI accelerators and networking, and Nvidia (NVDA), the GPU leader whose stock has cooled short-term despite its central role, round out the portfolio at smaller 4% and 3% weights respectively, since their recent momentum has been weaker even though they remain core to the theme. Together these seven names cover chips/GPUs, foundry equipment/manufacturing, memory, networking, and power/cooling — diversified across the supply chain instead of concentrated in one layer, with sizing tilted toward whichever layers currently show the strongest demand trends. About 1% is left in cash simply due to rounding.