The core thesis is to find asymmetric winners years early, before capex, product road maps, insider buying, and customer demand show up in the numbers. The playbook runs from secular-trend selection to S-curve timing, then to who captures hyperscaler capex, with AI, HBM, custom silicon, connectivity, and Tesla treated as the main examples. Financials, debt, capex, moats, ownership, analyst revisions, PEG, and technicals are all filters; the point is to buy before the crowd, size small enough to survive, and keep rotating when the thesis weakens.
The strategy is position trading in asymmetric bets over years, not day trading or scalping; weekly trading and aggressive short trades like oil and AVIS are allowed when the setup is there.
The first filter is secular trend: AI has been expected for four years, with next themes named as space-based data centers, humanoid robots, more tokens, and AI agents.
S-curves are the timing edge: TSLA after IPO was flat, then went vertical; AMD, MU, ALAB, and MAR were framed as early before AI, HBM, and data-center interconnectivity were widely recognized.
Hyperscaler capex through 2028 is the money map; an “i813” built 14 months ago and created in June 2025 was linked to that spend path.
The capex winners named are NVDA and AMD for GPUs/accelerators, MU and SK Hynix for HBM, and ALAB, MRVL, and AVGO for AI racks, retimers, optical connectivity, and custom chips.
A team of Tesla people in China is allegedly working on final approvals for the supply chain for the Optimus human robot.
The process says not to wait for revenue, profit, or P&L proof; by then it is too late, and winners are found instead by emotes, designs, switching costs, road maps, and investment intensity.
The top 0.3% of stocks over the last 100 years allegedly generated half the total return, so the target is to find the few names that matter rather than the broad market.
If NVDA is missed early, the playbook says the opportunity is still there in second-tier names like ALAB, MRVL, and custom AS6; missing one winner does not end the trade set.
Solana was found in March 2021 before “anybody else” was talking about it, and the same crypto tools are said to distinguish winners from losers better than stock tools can.
AMD’s data-center gains are tied to the MI300 roadmap and a software stack catching up to NVDA/CUDA, but custom chips are still described as the huge growth segment.
MU’s HBM setup is tied to signals from Elon Musk and Jensen Huang, hyperscaler capex, and the memory line items in NVDA and AMD BOMs; the trickle-down is described as enough to keep shortages going for the next three to five years.
TSLA is framed as a decade-plus investing story with a claimed $120 trillion TAM, roughly global GDP, and a moat from vertical integration plus exposure to all the top S-curves.
Free cash flow is the “golden indicator”; the financial screen also checks revenue growth, three-year kagger, gross margin trends, operating leverage, and the rule of 40.
Red flags are cash burn with no share gains, serial dilution, permanent one-time adjustments, revenue up with worsening unit economics, and especially debt; CoreWeave debt versus annual revenue or profit was called a reason to run.
Capex is usually scary to Wall Street because it hits free cash flow, but capex over sales rising into a demand wave is described as very bullish when it builds a moat; examples were MU fabs, AMD capacity, and TSLA gigafactories and lithium factories.
The moat test is whether the advantage can be copied in three years or less; named moats include vertical integration, scale, CUDA, process node, owning HBM supply, bottleneck supply, and custom relationships, while data was called fuel in the AGI age.
Ownership and sentiment matter: 13F tracking, Stanley Druckenmiller as a smart-money buyer, insider buying as extremely bullish, short interest as squeeze fuel, and insider selling as the first rats off the ship; CoreWeave insiders were said to have sold billions in 12 months while nobody was buying.
Analyst targets are treated as sentiment and information flow: half of Wall Street analysts were dismissed as idiots, revisions and spread matter, and a crowded target like SpaceX at $800 from Brian Gwal was said to lose edge as more people pile in.
SK Hynix at a 0.5 PEG was called stupid cheap versus a dream PEG of 1, and the technical-trading section said never buy a broken thesis just because it is cheap; oversold stocks are usually dead or dying.
The exit rule is to know the sell price before entry, size with 80% in safer names and 20% in faster movers, review quarterly, and rotate if the thesis changes — including a US AI ban scenario tied to slowing AI capex.
The workflow is built around scanning 200 names, picking the top five, layering in on dips, using AI to replace weeks of 10-Q/10-K reading, and staying patient enough to let one asymmetric winner change the outcome.
The close is simple: be the early bird, sit on cash, expect to be early and wrong-looking for a while, and accept that one good asymmetric bet can be life-changing.