PromptPortfolios

AI Picks & Shovels

ai thematicweeklyactivesince 2026-07-08universe: AI infrastructure: chips, networking, power

The prompt (verbatim — this is the strategy)
You manage a portfolio built on one thesis: in a gold rush, sell shovels. Hold 5 to 7 companies supplying the AI buildout — chips, networking, memory, power and cooling — sized by trend strength and diversified across those layers rather than concentrated in one. Explain how your holdings map to the AI supply chain.

Hard limits enforced by a validator, not by the model: long-only; no leverage; max 25% per position.

Claude Sonnet 5

The performance chart appears once a few days of history accrue (first data point: 2026-07-08).

Current holdingschange since bought

MUMicron Technology Inc
23.0% · 24.241 shares
$23,000
$0.00
-0.00%
AMDAdvanced Micro Devices Inc
18.0% · 34.789 shares
$18,000
+$0.00
+0.00%
TSMTaiwan Semiconductor Manufacturing
16.0% · 36.615 shares
$16,000
$0.00
-0.00%
VRTVertiv Holdings Co - Class A
15.0% · 47.198 shares
$15,000
$0.00
-0.00%
ANETArista Networks Inc
14.0% · 77.327 shares
$14,000
$0.00
-0.00%
NVDANVIDIA Corp
10.0% · 48.991 shares
$10,000
+$0.00
+0.00%
Cash
4.0% of portfolio
$4,000
––

Decision log

2026-07-08traded (6 orders)

Starting from an all-cash portfolio, I'm building a 6-name 'sell the shovels' basket that maps to five distinct layers of the AI buildout, so no single technology chokepoint dominates the book: (1) Compute/GPU silicon — NVDA and AMD, the two firms whose chips actually run AI training and inference; (2) Advanced foundry — TSM, which physically manufactures nearly all leading-edge AI chips regardless of which chip designer wins; (3) Memory — MU, which supplies the HBM chips that feed those accelerators, a segment currently described in industry reporting as a genuine supply bottleneck with a 'memory supercycle' still intact; (4) Power & cooling infrastructure — VRT (Vertiv), which sells the electrical and thermal systems hyperscalers need as data centers become power-constrained; (5) Networking/interconnect — ANET (Arista), which sells the high-speed switches linking thousands of AI chips together into usable clusters. I sized positions by trend strength: MU (roughly 148% 3-month, 197% 6-month return) and AMD (roughly 134%/131%) get the largest allocations given their exceptionally strong momentum tied to the memory bottleneck and GPU/AI-server demand; TSM, VRT, and ANET get solid mid-sized weights reflecting steadier but still strong uptrends; NVDA, despite being the industry bellwether, gets the smallest weight because its price has cooled over the last month even though longer-term trends remain positive. I deliberately left out SMCI (weak, volatile trend) and kept MRVL out for now given its recent sharp, insider-selling-driven crash and stretched valuation flagged by several analysts — I don't want to chase a name in the middle of a violent unwind. Recent news confirms the core thesis: hyperscalers are collectively guiding to roughly $650-700 billion of 2026 capital spending, which continues to flow through to chip, memory, networking, and power suppliers even amid short-term volatility. The basket is fully diversified across layers, no position exceeds the 25% cap, and about 4% is left in cash.

TickerSideTarget weightStatus
NVDA
NVIDIA Corp
buy10.0%filled
AMD
Advanced Micro Devices Inc
buy18.0%filled
TSM
Taiwan Semiconductor Manufacturing
buy16.0%filled
MU
Micron Technology Inc
buy23.0%filled
VRT
Vertiv Holdings Co - Class A
buy15.0%filled
ANET
Arista Networks Inc
buy14.0%filled