Chips in a Nutshell is produced by an automated research pipeline, not a human analyst sitting at a Bloomberg terminal. Here is exactly what it does and how.
Semiconductor supply chain research is hard to do well. Identifying a bottleneck requires reading earnings transcripts for capacity guidance, cross-referencing SEC filings for CapEx commitments, mapping the supply chain to find who actually owns the constraint, and then stress-testing your own conclusions against counter-evidence. A thorough analyst might spend 30-40 hours on a single theme. We run this process monthly, automatically, on the themes with the highest institutional signal.
Each monthly edition goes through eight sequential stages. The first five build the research. The last three exist to challenge it.
Scans SEC 13F and 13D filings from tracked institutional funds, curated industry sources, and runs seven targeted web queries on semiconductor bottleneck indicators. Produces three candidate themes ranked by signal strength.
Five iterative research rounds per theme. After the initial sweep, the system identifies its own knowledge gaps and generates targeted follow-up queries to fill them. Each round builds on the last.
Google Gemini Deep Research autonomously plans, searches, reads full articles, and iterates for up to ten minutes — independent of the prior research. Serves as a second opinion on theme ranking and investability.
A dedicated deep research pass focused on finding the best publicly-traded vehicles for each theme. Evaluates revenue exposure, competitive moat, market share, financial trajectory, and valuation. Returns one to four candidates ranked by fit.
An autonomous agent with eight specialized research tools verifies each company pick against primary sources: earnings transcripts, SEC filings, financial data, supply chain maps, and peer comparisons. Budget-controlled at 20 tool calls maximum. Every claim requires a citation.
A separate agent is given one instruction: disprove the thesis. It searches for capacity expansions that would ease the bottleneck, signs of demand destruction, technology substitutes, and competitive threats. Findings are fed directly into the final conviction score.
The system imagines it is twelve months in the future and the thesis has failed. It writes a specific post-mortem: what assumption broke, what warning signs were missed, and what the mechanism of loss was. This forces honest engagement with failure scenarios rather than generic risk disclosures.
Final conviction is scored against six explicit inputs: primary source ratio, multi-source confirmation count, counter-evidence severity, pre-mortem plausibility, smart-money SEC filing alignment, and evidence quality. Conviction can only go down from stress-testing. Themes scoring below 7 out of 10 are rejected and do not appear in the report.
SEC 13F filings require institutional fund managers to disclose equity holdings quarterly. 13D and 13G filings disclose acquisitions of stakes above 5%. We track a curated set of funds with demonstrated expertise in technology and semiconductor infrastructure.
Their positioning is used as a calibration input, not as the primary signal. Strong alignment from multiple funds raises conviction. Absence of alignment does not disqualify a theme. The research has to stand on its own merits first.
Stocks are ranked by a composite score that blends multiple inputs:
Concentration limits apply: maximum three stocks per theme. Position sizing uses inverse-volatility weighting. Stocks the research identifies as secondary plays are capped at a maximum conviction of 6/10.
Get the research delivered free, monthly.