Video summary

Everyone Hates AI Right Now. Four Stocks That Are Bulletproof

Main summary

Key takeaways

Finance

Finance-Focused Summary (Markets / Investing / Portfolio Logic)

Macro / Market Thesis: “AI Slowdown Debate Won’t Stop Capex”

The speaker argues that even if AI regulation or “slow down” pledges occur, AI infrastructure spending cannot stop due to physical bottlenecks:

  • <4% of US data centers can accept a full rack of Nvidia’s newest chips; older chips keep running.
  • Grid/power connection lead times for major data center markets are said to be ~4 years.
  • New data center builds and equipment availability also take years.

The speaker frames AI as a multi-year buildout similar to prior eras (e.g., railroads, housing), measured by capex as a share of GDP:

  • Housing boom (2005): 6.6% of GDP
  • Railroads (1870s): ~5%
  • AI: ~1–2% of GDP in 2026, potentially reaching ~3% by 2028 (citing Goldman Sachs)

Portfolio Construction Framework: “Own the Bottleneck Parts”

Rather than trying to pick the single winning AI company, the video claims a more durable approach is to own data-center “parts” that multiple AI builders can’t replace quickly.

Four “data-center buckets”:

  • Processors (compute): “does all the thinking”
  • Memory (working/storage): holds conversation context/files
  • Optics (data transport): converts electrical to light; fiber/optics interconnect
  • Inference / finished machines: rent compute by the hour/day

Investing Methodology: Signals From a Nightly System

The presenter describes an internal system that runs nightly and:

  • Executes ~20 million calculations across a watchlist
  • Outputs a signal: buy a lot / buy / buy a little / hold / sell

Core logic:

  • Fair value vs. current price
    • If price < fair value → “buy”
    • If price > fair value → “sell” or “buy less”
  • Trend adjustment
    • Buying is preferred when price is turning up; less preferred when sliding
  • Sell-timing emphasis
    • Claims strong sell accuracy: 7 of the last 8 years
  • Lower conviction when problems are found
    • Buy size cutbacks occur on “three of seven names” when issues appear

Tickers / Assets Mentioned

Equities / Companies

  • Nvidia (NVDA)
  • Micron (MU)
  • SanDisk (described in context of SanDisk / Western Digital; ticker not provided in subtitles)
  • Lumenum (optics/lasers; ticker not provided in subtitles)
  • Marvell (ticker not provided in subtitles)
  • Cerebras (spelled inconsistently in subtitles; ticker not provided)
  • CoreWeave (private; no ticker)

Sponsor / Mining Equities

  • Copper One Resources Corp (sponsor; ticker not provided)

Commodities

  • Copper (referenced in $/pound)

Other References (No explicit tickers given)

  • “401k” referenced generally
  • Amazon / Google / Microsoft / Anthropic / Broadcom / AWS mentioned for demand/capex context
  • Only NVDA and the other named chip/storage/infrastructure equities have explicit ticker-style mentions

Key Numbers, Timelines, and Explicit Calls

Copper (Sponsor Segment + Commodity Context)

  • Copper price: all-time high > $6.70/lb (August); up >40% in 12 months
  • Copper One
    • Down ~75% in 2026
    • Cash / working capital: market cap ~CA$14.8M, with ~CA$10.4M as cash
    • Valuation framing after backing out cash:
      • “Pricing three copper projects ~CA$4M” total
      • Projects mentioned: Majuba Hill (Nevada) (past producer), Red Roanda and Red Hill (British Columbia); Roanda drilling “right now”
  • Standard promo tone: “do your own due diligence”

AI Infrastructure Spend Outlook (Demand Durability)

Cloud commitments

  • ~1.7 trillion of signed work across Google Cloud, Microsoft, and Amazon
  • Framed as ~3 years of AI buildout at the cited 2026 pace

Compute intensity

  • One AI agent job uses 15–100x the computing of a person doing the job

Chip utilization / constraints

  • Nvidia chips allegedly stay utilized even as older models remain deployed (discussion references “warm shells” / utilization constraints)

Investment Calls (7-Name Sequence With Explicit Labels)

1) Micron (MU) — Memory (fast/near-processor)

Operating profit margin impact

  • Two years ago: 11 cents operating profit per $1 sold
  • Latest quarter: 80 cents operating profit per $1 sold

Timing

  • Earnings “again on September 30th”

Trend / technical logic

  • Above 200-day moving average
  • Slipped under 50-day at one point

Fair value vs price conflict

  • System: trend says buy a lot
  • Historical valuation check: Micron has been this expensive only ~2% of the time in 10 years

Recommendation

  • “buy a little”
  • Wait for the September report to decide whether to buy more

2) SanDisk — Storage-side memory

Profit

  • Latest quarter: 78 cents operating profit per $1 sold

Valuation

  • Still under 8x expected earnings next year

Growth cheapness metric

  • Uses PEG = price / growth
  • PEG cited: ~0.3

Recommendation

  • Buy “a lot” (by both price and trend)

Caveat

  • Public history only ~19 months, so the presenter can’t run the 10-year expensive/cheap frequency check; conviction reduced
  • System says buy, but the presenter says “I don’t lean into it.”

3) Lumenum — Optics / lasers

Profit trend

  • Two years ago: losing money per $1 sold
  • Now: 27 cents operating profit per $1 sold

Growth claim

  • Sales grew 83% in a year (with an “August” earnings headline miss mentioned)

Valuation

  • Pricing cited around ~195x last year’s profit
  • Speaker claims that normally it costs around ~70x
  • PEG mentioned as under 1 (growth “not expensive”)

Recommendation

  • Possibly sell some
  • Rationale: business looks fine; price ran ahead of fundamentals

4) Marvell — Custom AI chips (clouds building in-house vs Nvidia)

Demand visibility

  • In August 2026, Google gets a right to buy ~6.7% of Marvell
  • Google “earns the stake” by buying ~120B chips through 2033
  • Implied: next ~7 years of buying locked in

Valuation

  • Paying ~80x last year’s profit
  • “Normally” around ~33x

Recommendation

  • Strong sell
  • Rationale: price already reflects the next 7 years of growth

5) Cerebras — (spelled inconsistently in subtitles)

Backlog / order visibility

  • Backlog: $25B in signed orders
  • Under $1B worth of sales happening this year
  • Key risk: converting orders on schedule (or thesis pauses)

Financials

  • Latest quarter loss: -$450M
  • Presenter says much is non-operating/stock-related (employee stock)
  • Cash: $8.6B

Recommendation

  • Buy a lot by the price signal
  • But system is cautious due to limited trading history:
    • Company only public ~4 months
    • System relies on analyst-implied fair value → signal notched down one
  • Practical stance: buy depends on backlog conversion timing; if delayed, likely sidelines

6) CoreWeave — Private inference / renting compute

Demand vs capacity

  • Two years ago quarterly sales: $400M
  • Now quarterly sales: $2.6B
  • Customers signed for ~4.2 gigawatts of power
  • CoreWeave only has ~1.5 GW running
  • Implied oversubscription today: ~3x

Risk / cash burn (explicit)

  • Last 12 months cash burned: $13.7B
  • Sales: $7.66B (interpreted from subtitles showing “7.6 6”)

Recommendation status

  • Presented as a high-risk inference bet with material funding/capex exposure
  • “Touches on risk”; a clear buy/sell label was described as less explicit than for the other names

7) Nvidia (NVDA) — “Anchor” position

Scale and profitability

  • Two years ago: $30B quarterly revenue
  • Now: $96B quarterly revenue
  • Operating profit margin: keeps 66 cents per $1 (vs 62 cents two years ago)

Supply/demand constraint

  • Speaker claims Nvidia supplies about ~70% of what customers ask for

Lease backstop

  • Nvidia guaranteed up to $18B of customer data center leases (co-signer if a customer can’t pay)

Valuation

  • Paying ~28x last year’s profit (video wording: “as of today”)
  • Historically “normally cost about 52x”
  • “Pay ratio” during filming: 0.35 (interpreted as a low valuation/fair-value indicator in their system)

Recommendation

  • Buy a lot / “by a lot”
  • Technical timing: above 200-day moving average even after a “very soft week”

Key thesis risk

  • Could unravel if signed orders (earlier referenced ~$1.7T) begin to shrink

Disclosures / Disclaimers

  • Presenter states: “I am not a financial adviser and I do this for educational purposes.”
  • Additional promo: like the video; Patreon support encouraged
  • Copper sponsor disclosure:
    • “This segment is disseminated on behalf of Copper One Resources Corp”
    • “please do your own due diligence”

Key Presenter / Source List (As Stated in Subtitles)

  • Jensen Wang (Nvidia) — quoted
  • Anthropic CEO — referenced (name not given in subtitles)
  • Broadcom CEO — referenced (name not given in subtitles)
  • Satya Nadella (Microsoft) — quoted
  • Goldman Sachs — referenced (AI buildout projection)
  • The presenter of the video — not named in subtitles
  • Copper One Resources Corp — sponsor entity

Original video