Corrected plan — $20k/month profit from physical goods in the AI trade's path

Prepared Oct 13, 2026. Supersedes the securities-based outline (Chad already has that fund — Leadsure). Model: buy and sell physical items the AI buildout cannot wait for. Scarcity + lead times + impatient buyers = the spread.

Why this works right now (checked Oct 13, 2026)

• Memory is the epicenter. DRAM spot prices up ~700% in a year per Bloomberg (July 2026); a 16Gb DDR5 chip went ~$6.84 → ~$27.20 in one quarter; a 32GB DDR5 kit went from under $90 to ~$529. Samsung/SK hynix inventories under 10 days of supply. Micron says 2027 supply is already 75% sold out and conditions stay tighter through 2027–2028 (TrendForce: meaningful new capacity only from H2 2027).
• Power gear is the hard ceiling. Large transformer lead times stretched from ~50 weeks (2021) to 140–160+ weeks (Wood Mackenzie, 2026); switchgear sold out through 2028. Anything available now commands a speed premium.
• Used GPUs are volatile, not scarce-priced. H100s fell from ~$40k (late 2023) to ~$12–22k used / $21–34k refurbished in mid-2026, with March 2026 auctions printing $8,200–$12,500. The spread lives in auction-vs-resale mispricing, not in holding — values are still sliding as Blackwell fleets rotate.

What to buy and sell — ranked by fit for ~$100k and a small operation

1. Memory & storage (best fit). Server DDR5 RDIMMs, enterprise NVMe SSDs, and legacy DDR4 (EOL scarcity pricing). Small, shippable, price-transparent, testable, deep buyer pool (servers, workstations, repair, homelab, smaller data centers). This is the Lonis "memory at 3.4x revenue" chokepoint in physical form.
2. Part-out arbitrage. Buy decommissioned enterprise/AI servers in liquidation lots; the DIMMs + SSDs + NICs + GPUs sold separately are routinely worth more than the whole unit. Chad's team knows servers (Cloudways/DigitalOcean/Forge operations) — testing and listing is in-house skill.
3. Networking. 400G/800G optics (QSFP-DD/OSFP transceivers), InfiniBand/ConnectX NICs, high-speed switches, DAC/AOC cables. High value density, chronic allocation shortages, buyers who cannot wait.
4. Power & cooling at tradable scale. Not utility transformers (institutional size) — but rack PDUs, UPS units, CDUs, liquid-cooling components (cold plates, manifolds, quick-disconnects), and used commercial generators. The trade is lead-time arbitrage: available today beats cheaper in 2028.
5. GPUs — small, rules-based allocation only. Buy only at auction/liquidation lows with verified serials and test capacity; sell into the reseller market fast. No inventory aging: prices are falling ~10–20% per generation rotation. Strictly domestic, fully export-control compliant — the offshore price gap exists because of legal restrictions and is not a trade to touch.

Operating models (can combine)

• Broker first, inventory second. Source deals and take margin without holding stock while learning which categories actually move — protects the $100k.
• Buy → test → refurb → resell through eBay plus an own storefront. Chad runs a Shopify agency (Ambaum) — building the storefront/catalog is free in-house expertise competitors pay for.
• Liquidation lots + part-out as the supply engine for models above.

The math to $20k/month profit

Profit = gross sales × net margin. With ~$100k of working capital: | Net margin | Monthly sales needed | Implied inventory turns/mo on $100k | |---|---|---| | 15% | ~$133k | ~1.3x | | 20% | $100k | 1.0x | | 25% | $80k | 0.8x | | 30% | ~$67k | ~0.7x | So the plan needs roughly one full inventory turn per month at ~20% net — aggressive but coherent for scarce components with fast resale, and it scales by adding turns (faster cycle) before adding capital.

First 30 days

1. Pick the beachhead: memory/storage + part-out (lowest ticket, fastest turns, easiest testing).
2. Set up sourcing: liquidation/auction platforms, ITAD and decommission brokers, reseller wholesale lists; register as a buyer; watch 2 weeks of sold prices before buying.
3. Stand up the sell side: eBay seller account + simple storefront, testing bench (memtest/NVMe health/GPU stress), serial-number verification and photo process.
4. First trades small: 3–5 lots, target proof of cycle (buy → test → list → sold → profit per day of capital) before scaling ticket size.
5. Add one category at a time (networking next, then power/cooling, GPUs last and smallest).

Risks / guardrails

• Prices can round-trip: memory will eventually normalize (2027–28 capacity) and GPUs are already deflating — inventory must turn in weeks, not quarters; no "investing" in stock.
• Counterfeit/re-marked chips and dead-on-arrival parts — testing and serial verification are the business, not overhead.
• Export controls on advanced GPUs/memory are real law (the offshore premium is a legal wall, not an arbitrage) — domestic, compliant sales only.
• Sales tax, business licensing, and warranty/returns policy need setup up front — Chad's CPA background covers the bookkeeping discipline; a licensed advisor should bless the tax setup.

What Muse can do next, on go-ahead

• Build the sold-price tracker for target SKUs (DDR5 RDIMM sizes, enterprise SSD models) from marketplace data.
• Draft the sourcing watchlist and first-lot criteria.
• Stand up the weekly profit/turn dashboard against the $20k target.