New: Earlier today, Claude's Agents decided to take our $50K, sell out of Thermo Fisher $TMO, and buy Nvidia $NDVDA
Here's the full rationale on why:
"WHY CLAUDE SWAPPED TMO FOR NVDA
Thermo Fisher had been in the portfolio as a defensive life-sciences compounder: global instrument installed base, recurring reagent revenue, and a bioprocessing segment that was supposed to recover as pharma capex normalized. The thesis was intact on the surface.
But when Claude ran every researched name through the same expected-return framework this cycle, TMO came back at +5.3% over 12 months. That is the lowest number in the entire universe Claude evaluated. Holding the weakest-ER name in the book, ahead of an Apr 23 earnings binary with real tariff exposure on lab equipment imports and China revenue risk, for a company with an incremental proteomics partnership in Singapore as its biggest news of the day, is not a defensible position.
The entry point into NVDA has been building for months. The stock has been compressed by three overlapping forces: the Hormuz geopolitical overhang, a rotation out of growth into energy and defensives, and Michael Burry's widely-covered short position adding a bearish narrative layer. Strip those out and the fundamental setup is unusual. NVDA is trading at 17x FY28 EPS, which sits 51% below its own five-year average multiple and 22% below comparable AI-infrastructure peers.
The market is pricing in persistent disruption risk at a level that does not match what the hyperscalers are saying with their capex budgets.
The catalyst stack is what made the timing specific rather than generic. TSMC reports April 15-16 with CoWoS-S and HBM3E yield data that will directly read through to NVDA's supply constraints. The hyperscaler cluster reports between April 22 and May 2, and every one of those companies has guided toward accelerating AI infrastructure spend. NVDA's own Q1 prints May 20. Three independent confirmation windows in five weeks, each capable of resolving the uncertainty the market has been pricing in.
The risk is real: a Rubin architecture yield miss, an HBM supply disruption, or a hyperscaler capex trim on a trade-war demand shock would all hurt. But the expected return at current prices is +8.3% over one month, +16.3% over three, and +37.7% over 12, and those estimates were built on the current depressed multiple, not on a recovery. The asymmetry is there. Single swap. 60.5% of the cycle turnover budget used."
