Front-running Washington's next stock pick (Using AI)
•By Dave Wang
A few days ago I saw the reports that the Trump administration was in talks to take an equity stake in more AI companies.
What caught my attention was the pattern sitting behind it....
The government already owns part of Intel. It became the largest shareholder in MP Materials. It holds stock and warrants in USA Rare Earth.
None of those were open-market buys. Each one was a surprise, and each one moved the stock hard.
So I asked a simple question. If there is a logic to who gets picked, can I reverse-engineer it and get ahead of the next one?
The hard part is that the criteria look political and ad hoc from the outside.
Who actually qualifies? What triggers a deal? Which AI names are in the blast radius?
This is where AI comes in.
I used Claude Code to do two things in one workflow. First, read the primary sources and pull out the real decision rule. Then score every US-listed AI company against it.
Here is the plan:
Pull the executive orders, deal terms, and officials’ own words
Extract the rule the government keeps following
Score the AI universe against that rule
Rank the names most likely to get a stake next
The Prompt
I ran this on Claude Code. You can run this on Codex as well - but the key is it should be running on one of the agentic platforms as you will require subagents to handle this amount of context.
In terms of the prompt, I like to frame the logic in a tops down format in order to force AI to search through logic I may not be considering. That's how you can get first principles style thinking rather than having overly prescriptive prompts (esp using tools like Claude Code).
Prompt:
I am a sophisticated public-equities investor focused on deploying into the AI theme. I keep seeing reports that the Trump administration is taking direct equity stakes in strategically important companies, starting with names like Intel and MP Materials, and now floating AI companies. I want to reverse-engineer the logic behind who the government picks. And come away with an actionable, rank-ordered list of US-listed AI names most likely to get a stake next so I can front-run the announcements. I only care about public companies, not privates.
Let's use the /plan mode and the ask user interview tool to make sure we are fully aligned on what I need from you and that there are no gaps of understanding between us and no gaps of detail that I am not thinking of for this analysis.
I want this to be incredibly top-down and first-principles. Start by getting inside the mind of the administration from primary sources (executive orders, official interviews, and the terms of the deals already done): what they care about, the criteria they look for, and any relationship or political-alignment signals that make a company a beneficiary. Then turn that into a screening framework, apply it to every relevant public company, and score each by likelihood, grouped by subsector within AI. Pattern-match from the historicals and pressure-test the thesis. I want a clean markdown report and an Excel with the rank-ordered list, plus deep dives on the top names, and potentially more if I am forgetting anything here. Don't worry about how many tokens you burn or how long the report gets, as long as the methodology is strong and it gets me to the result.
The most useful research the model gave me was the rule....
Here is that rule in one line:
The government doesn’t buy these companies on the open market. It converts a lever it already holds.
Intel started as an undisbursed CHIPS grant. The government turned it into a 9.9% stake.
With MP Materials, a defense relationship became preferred stock, a warrant, and a guaranteed price floor.
Lithium Americas was simpler. An existing Department of Energy loan turned into warrants.
The setup is the same every time. Take a strategically critical company that needs the government more than the government needs it, then convert existing support into ownership.
Once you see it, the prediction problem flips. You stop asking which companies are strategic, and start asking which ones already have a federal lever attached that nobody has converted yet.
And those levers are public. They sit in CHIPS award lists, Department of Energy and EXIM loan books, and Defense Production Act awards.
Interestingly, the market reads those as boring debt headlines, not as equity in waiting.
A few names that score near the top of the framework:
Wolfspeed ($WOLF). A distressed silicon carbide maker sitting on an undisbursed CHIPS award. The cleanest rhyme with the Intel setup.
Perpetua Resources ($PPTA). It just landed a $2.9B EXIM loan for the only US antimony mine. The lever is already attached.
GlobalFoundries ($GFS). A CHIPS recipient and the trusted foundry for US defense chips.
Rigetti ($RGTI) and D-Wave ($QBTS). Both already sit inside a $2B Commerce quantum program that takes minority equity.
One note: This is a probability map, not a buy list. A few honest caveats:
A lever on the books is not a signed deal. The timing is political, and some of these could sit for a year or never happen.
A government stake isn’t automatically good news. Warrants and golden shares can hang over a stock for years.
The framework is only as strong as the source behind each score, and some are cleaner than others.
So treat this as exposure to a pattern, not a recommendation.
Let’s check back in a few months and see which of these the government actually moves on.
Personal
I'm co-hosting a live panel with the AlphaSense team and Andrew Walker from Yet Another Value Podcast this month!
This will be a free event where I'll be sharing my thoughts about what a 'modern institutional AI stack' looks like + where I think certain platforms shine in different investing use cases.
I know many readers here from fund backgrounds (esp. equities) have mentioned to me that they're trialing AlphaSense - so I decided to do this event together with the AlphaSense team to hopefully provide some clarity on how they fit in my stack vs the other foundation model providers.
Check out the link below to listen in - hope to see you there!