Situational Awareness LP just filed its Q4 2025 13F. I've been waiting for this one.
Situational Awareness is widely regarded as the top AI thinker amongst public equity investors.
The fund is run by Leopold Aschenbrenner, former OpenAI researcher who wrote the 165-page "Situational Awareness" memo that shook the AI industry. He grew the fund from $255M to $5.5B in a single year.
I've talked to a lot of top hedge fund PMs about this space.
The consistent thing I hear is that semis and AI infrastructure investing is brutally technical. It's hard to understand "what's next" if you aren't close to the research. Aschenbrenner was inside the labs where frontier models were built. That's a different vantage point than most. So when I see his 13F drop, I pay attention.
Where is he rotating capital? What sub-sectors is he building conviction in? What's he exiting?
I personally am placing significant % of my capital towards AI demand themes based on my vantage point .... and I like to study Situational Awareness to see how they're expressing this theme.
I built a single AI prompt to pull his raw 13F data, crunch the numbers, and spit out an interactive dashboard that answers all of this.
Here's how I set it up:
- Pull raw Q4 and Q3 holdings data directly from SEC EDGAR XML filings
- Compute quarter-over-quarter position changes across every holding
- Classify each position as a new initiation, add, trim, or exit
- Build an interactive HTML dashboard with sub-sector rotation analysis
One more thing. I ran this entire workflow in a single prompt start to finish.
I wanted to stress-test this because Claude Code, Perplexity, and Manus all shipped significant improvements to context management this week. Previously, multi-step agentic prompts had quality degradation... steps 8 through 12 would get sloppy and the AI would lose the thread. That ceiling is moving fast and I wanted to see what's possible now on a real finance workflow.
The Prompt
We are using Claude Code to copy paste one shot this prompt below. Previously we had to split up multiple tasks due to agent skill limitations + context limits .... but parallel agent improvements now fix this issue.
Prompt 1: Situational_Awareness_Prompt.pdf
P.S. If you want to learn how to use Claude Code for Finance, I'm setting up a LIVE bootcamp ... sign up to get on the waitlist:
The Result
Full output here: Link
Three names entered core conviction status this quarter. Two exited.
- Bloom Energy (BE) entered core at 15.87% of the portfolio, up from 1.06% last quarter. He added +$831.6M. Now his single largest holding.
- Lumentum (LITE) entered core at 8.67%, up from 1.64%. Added +$410.6M. Lumentum makes the optical transceivers that connect AI clusters.
- Applied Digital (APLD) entered core at 5.04%, up from 3.36%. Nearly doubled the position at +$138.9M.
On the exit side, he closed his Nvidia put hedge (was 7.21% of Q3) and sold out of Vistra entirely. Every other put from Q3 is gone too... Broadcom, TSMC, Micron, and the VanEck Semiconductor ETF.
When I mapped the energy rotation I thought something was off at first. Energy went from 10.2% to 21.5% of the portfolio in a single quarter, a +11.3 percentage point swing. He stripped out all the defensive semiconductor hedges and redeployed into Bloom Energy fuel cells, Lumentum optical networking, and power infrastructure names.
The thesis I'm reading... he's moving to "go long the energy and connectivity layer feeding AI data centers."
Now zooming out on what happened on how we created this dashboard.
With just one prompt, what did we do? We instructed AI to pull raw SEC EDGAR XML filings, normalize 57 rows of holdings data across two quarters, compute every delta, classify every position change, run sub-sector analytics, and generate a full interactive dashboard with charts, filters, and CSV exports.
Why does that matter to me?
Because if you can one-shot it, you can schedule it. Set this to run automatically every quarter when 13F filings drop. No manual scrubbing. No re-prompting. The analysis just shows up... as if you had a research engineer on staff.
Personal
I just crossed 20,000 subscribers on our newsletter this week!
I appreciate each one of you .... I never would have thought to grow into one of the largest AI for investors newsletters in just a single year.
To welcome the new faces, I have a question for you!
- Where are you on your AI skill level? What are you looking to learn?
- What area of finance or investing do you work in?
Hit the reply button and I might have a resource or two I can point you in the right direction.
P.S. If you like what I write about, forward my emails to your colleagues or friends .... it helps me a lot :)