In this week’s video, I break down one of the worst week for stocks since Liberation Day, the S&P down 2.6%, the Nasdaq down 4.5%, and make the case that this is a necessary rotation, not the start of a bear market. After nine straight up weeks, none of this should be a surprise. The 20-day finally broke, weekly RSI hit multi-year highs with no divergence yet, and Friday’s beta unwind was the sharpest since early 2000s, a genuine cleanse.
The bigger message: the fireworks show is over. The buy-anything phase of the agentic AI trade where memory, semis, and infrastructure gapped 10–30% on the way up on good news has moved from discovery to digestion. The agentic buildout itself has just begun, but the next three to six months will be two-sided and choppier; you have to do the homework now and manage through a much more crowded positioning running into bottlenecks. And the macro backdrop is the photographic negative of a sustained bear market: credit spreads near all-time tights, no jobless-claims spike, rising PMIs, record profit margins, and earnings revised up, not down.
That’s why I’m watching the rotation, not running from it. On Friday the S&P fell nearly 3% while five sectors rose, capital left the chip complex and moved into healthcare and financials. Structurally, we’re shifting from labor-vs-capital to compute-vs-energy: the input is electrons, the output is tokens. I think memory has likely topped for now and recursive self-improvement, DeepSeek, and government-stake risk all argue for increased caution moving forward so I prefer Marvell and optical over Micron. I’m adding energy (Exxon, Chevron), batteries (Fluence), Bitcoin near its 200-week, and leaning into the application layer, where Eli Lilly may be the most important AI company in the world.
Timestamps
- (00:00–02:20) Markets: rotation or bear market? The S&P fell 2.6% and the Nasdaq 4.5%, the worst week since near Liberation Day but after nine up weeks in a row, this looks like a necessary rotation, not a top.
- (02:21–06:09) Technicals: the Q’s and S&P finally closed below the 20-day; weekly RSI hit multi-year highs with no divergence yet; Friday’s Morgan Stanley beta factor fell ~10.5%, the sharpest move since 2000 a real cleanse.
- (06:10–07:45) “The fireworks show is over”: agentic AI moves from discovery to digestion. The easy gains are done; the next 3–6 months are two-sided, but the agentic buildout has only just begun.
- (07:46–12:46) How we got here: CES (Vera Rubin, “tokens per watt,” the five-layer cake), the March TMT sentiment shift, and Computex. Google’s $85B raise and Meta’s equity raise are to fund capex and that looks more like a possible short term capex top than a bottom.
- (12:47–18:17) The macro reality: earnings revised up, margins at record highs with no mean reversion, credit spreads at all-time tights, no jobless-claims spike, rising PMIs, “the photographic negative of a sustained bear market.”
- (18:18–19:59) The rotation in action: Friday the S&P fell ~3% but five sectors rose; capital left chips for healthcare and financials (deGraaf, Parker, Newton).
- (20:00–23:30) From labor-vs-capital to compute-vs-energy: the input is electrons, the output is tokens. Digital workers don’t buy homes or cars, so the old credit-and-labor cycle no longer applies.
- (23:31–27:17) Crypto rails & Bitcoin: agents need layer-1 rails to transact and settle. Buying Bitcoin in pieces near the 200-week MA, “the only high-quality stock that will exist for certain in a decade.”
- (27:18–34:24) The memory bear case: don’t extrapolate token demand into DRAM. Recursive self-improvement could solve the memory bottleneck; DeepSeek/edge substitution and government-stake risk pile on. Marvell and optical/CPO over Micron.
- (34:25–55:15) Capex vs. adoption and the application layer: $60–100B of semis per gigawatt while adoption lags; the market shifts from a capacity land-grab to ROI discrimination. Energy (Exxon, Chevron), batteries (Fluence), and Eli Lilly as the most important AI company in the world.
