Supply chains don’t care about press releases. That’s the idea running through The Verge AI’s new Decoder interview with Willy Shih, a Harvard Business School professor and the author of the upcoming book Where Does My Stuff Come From? The Powerful Economics Behind Global Supply Chains.
The timing matters. As The Verge AI puts it, tariffs keep changing, “the factories that were supposed to reshore in America mostly haven’t materialized,” and “AI is devouring the planet’s computing resources” during an unprecedented infrastructure buildout. Shih’s argument is that economics sits underneath all of it, and that’s where anyone betting on AI infrastructure should start looking.
Why Shih is worth listening to
In 2019, Foxconn promised to build an LCD factory in Wisconsin. Shih didn’t buy it, and his reasoning was refreshingly physical. A panel factory needs a glass plant nearby, so once he knew there wouldn’t be one, he says, “there’s no way there’s going to be a factory.”
Then locals sent him the actual plans from their zoning hearings. His verdict: “there’s not enough concrete to make this a panel factory.” A panel plant needs stability, which means lots of structural steel and concrete. The factory never happened.
One good call doesn’t make anyone a prophet, but this one came from checking what was physically on the ground instead of listening to the announcement. That method travels well.
The fight over deglobalization
There are two competing stories here.
- The deglobalization camp: several Decoder guests have said flatly that the economy is deglobalizing. The pandemic broke just-in-time manufacturing, tariffs finished the job, and companies can’t rely on global supply chains anymore.
- Shih’s camp: global supply chains exist because they make economic sense, and the market keeps responding to that logic.
His examples are great. Cashews labeled as Vietnamese were actually grown in the Ivory Coast or Ghana. Fish caught off Alaska go to Qingdao, China, for processing before coming back. Both sound absurd until you see the cost math.
So when the US puts up a trade barrier, the first reaction tends to be, in his words, “How can I circumvent that?” Manufacturers shift assembly to Vietnam or Mexico. The chain bends around the wall, like water finding a crack.
What this means for AI
This next part is my read, not Shih’s. Apply his concrete test to AI and the headlines start to look different.
An AI data center announcement is a promise, just like Foxconn’s was. Dollar figures and gigawatt targets tell you what a company wants. They don’t tell you whether the supporting pieces exist, like power hookups, cooling, chips, memory and the people who can build it all. Those are the glass plant across the street.
The other lesson is about policy. If chip tariffs or export rules hit AI hardware, expect rerouting before you expect reshoring. A rule can change where the last assembly step happens without changing who controls the hard parts. That’s why China’s manufacturing position is so sticky, even though, as The Verge AI notes, every other country considers the dependence risky.
Shih learned one thing after moving from engineering into product businesses: “if you don’t get the economics right, then it doesn’t matter for science and technology.” AI labs, take note. Great models still sit on top of a physical business with physical costs.
What to do about it
If you buy AI capacity or build on top of it, a few habits will help:
- Run your own concrete test. When a vendor or cloud provider promises future capacity, ask what’s already permitted, powered and under construction.
- Map your dependencies past the first layer. Your model provider depends on a cloud provider, which depends on chips, power and a lot of steel.
- Treat low prices as a snapshot. Tight compute supply can change your costs fast, so keep a second model or provider tested and ready.
- Read tariff news skeptically. A new label on a product doesn’t always mean the supply chain moved.
What comes next
The AI buildout will get its own Foxconn test as today’s announcements turn into construction sites, or don’t. My strategic advice is simple: plan around what you can verify on the ground, and treat everything else as a forecast. The full conversation, including why so much LCD manufacturing ended up in China, is at The Verge AI.