Supermicro co-founder Wally Liaw got arrested this week.

He’s accused of smuggling $2.5 billion in Nvidia GPU servers to China through fake purchase orders routed through a Southeast Asian shell company.

They allegedly used hair dryers to swap serial number stickers to fool compliance audits.

$SMCI dropped 33%.

Smart investors should be asking the "what happens next" questions:

Who supplies Supermicro?
Who buys from them?
How concentrated is this supply chain?
If they get sanctioned or lose their Nvidia allocation, who else gets hit?

AI can now do a great job at these sorts of tasks.

I wrote a single prompt, fed it to Claude Code, and had a fully sourced supply chain risk map back in under an hour with 9 suppliers, 8 customers, revenue exposure percentages, concentration risk flags, and 23 cited sources.

Here’s the plan:

Write a prompt pulling supplier/customer data from 10-K filings and earnings calls
Calculate revenue exposure and concentration percentages
Generate a visual map with SMCI at center, suppliers left, customers right
Package into a single HTML report I can print to PDF

I did this while eating a chicken burrito with one hand.

The Prompt

I ran this on Claude with a single prompt, no chaining needed.

The prompt instructs Claude to act as a sell-side equity research analyst and produce a Bloomberg SPLC-style supply chain map for any ticker you give it.

If you want to customize this for your coverage universe, adjust the parameters.

If you want deeper research with more context, try running this on Claude Code.

Prompt: https://www.finprompter.com/share/e565cb1d-f9d8-47b7-bb77-ea09a96b6b5e

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:

Claude Code for Finance Bootcamp Waitlist

The Result

Full output here: https://davidwang95.github.io/smci-supply-chain-risk/

Claude Chat: Link

Here are the numbers that jumped out.

NVIDIA is 64.4 percent of SMCI’s total purchases. Roughly $12.6 billion. That is disclosed in the 10-K as “Supplier A.”

That’s not a supplier relationship. That’s a dependency.

On the customer side, it’s just as concentrated.

One customer (likely CoreWeave)hit 63 percent of quarterly revenue in Q4 FY2025
Four customers individually exceeded 10 percent of annual revenue
CoreWeave’s own concentration risk cascades: Microsoft was 62 percent of CoreWeave’s 2024 revenue

Then there’s the related-party structure.

~9% of COGS flows through two Taiwan entities (Ablecom, Compuware) run by the CEO’s family. Both export 99%+ exclusively to SMCI. Hindenburg flagged this in August 2024.

I want to note that some of these figures are estimates. The report flags each data point as “Disclosed,” “Estimated,” or “Inferred” so you know what’s from the 10K versus what’s derived from industry analysis.

Here’s what matters more than the SMCI analysis itself.

Bloomberg has a function called SPLC that does this. Suppliers on one side, customers on the other, revenue flows mapped visually. It’s better right now because they have teams manually scrubbing and verifying each data point.

But most of this data is sitting in public filings and earnings transcripts. It’s scrapable. And AI just made scraping it fast and cheap.

This type of analysis wasn’t accessible to people without a Bloomberg terminal. Now it is.

The public data layer is becoming table stakes.

When anyone with a prompt can build a supply chain map from the same underlying filings, the data itself stops being the edge.

The edge is private data.

Say you did an expert network call with a former Nvidia account manager who covered Supermicro. Or you have shipping data showing server volumes through Southeast Asian ports.

Layer that onto a dashboard like the one I just built. Now you’re looking at something no other fund can see.

I think large funds start building dedicated data teams over the next year or two.

AI makes it viable to scrape and standardize alt data at scale now. Before, it didn’t make sense to try to compete with Bloomberg’s data infrastructure. That’s less true now.

The funds that pair commodity data with proprietary sourcing will have something the terminal can’t give you.

Try this prompt on your own ticker. Hit reply let me know if you tried this out!

Personal

Last month, I was appointed to the advisory board of the Harvard Data Science Initiative, where I will be advising on AI applications in financial services.

This is an area I care deeply about. There is still a meaningful gap between what academia is researching and what practitioners actually need, and I am excited to help narrow that divide.

If your company is looking to hire AI talent from Harvard, or if you are interested in contributing to academic research in this area, feel free to reply and I will connect you with the right people.

Get one high impact AI prompt every week