What happens to AI progress if China invades Taiwan? A chat with ‘Chip War’ expert Chris Miller.

There’s a lot about the chip industry that is evolving by the day. And some things stay the same.

What’s changing is the progress AI chips continue to make training ever-larger models on more and more data.

The stasis: The worrying fragility of the semiconductor supply chain.

Apple, Nvidia, Qualcomm, AMD, and most other chip designers rely on Taiwan Semiconductor Manufacturing Company to make these crucial components. TSMC is the most advanced manufacturing company. And its location on the island of Taiwan, which China claims, puts the tech industry and AI’s future in a precarious geopolitical position.

Chris Miller is the author of “Chip War: The Fight for the World’s Most Critical Technology.” He spoke to us about what’s happened since his 2022 book came out, and how to think about the forces that threaten AI progress.

This Q&A has been edited for clarity and length.

Is the return on investment of AI and what GPUs enable important over the long arc of history?

The level of investment will only be sustainable if there is monetization of the products that are produced with all this compute. I personally get a bit worried by the scale of investment relative to what is possible to monetize today. And I think the big tech companies investing $10 billion a quarter in AI infrastructure — they’re going to have at least another year, maybe a bit longer, before their investors start asking, ‘What’s the return on all this CapEx?’ But at some point, that question is gonna be asked in a way that they’ll have to answer or they’ll have to slow the CapEx.

So, the best case scenario is that it’s not actually a problem. What do you see as the worst case scenario?

I have no doubt that we’re going to be using lots and lots of AI in lots of use cases over the coming decade. I think the realistic downside scenario is just that we’ve done a bit of over investment and we’re probably going to have to wait a cycle or two before we build our next round of data centers, and that will look pretty ugly for chipmakers. And I’m not predicting this will materialize, but that’s what I worry about — that the slope of the line of investment is just too high.

That reminds me of one of the major takeaways from your book, which was how precarious the chip industry is.

Yeah, it’s always cyclical. So the question is, when is the cycle going to move up and down? Right now, it’s in a really steep upward period and it always feels good when you are there. But it’s a cycle.

Let’s stick with worst-case scenarios. Given the rise of TSMC and Nvidia, do you think it’s more or less likely that China invades Taiwan now than it was 18 months ago?

I don’t think it’s less likely. I think that the international situation broadly is a lot darker with Russia, Ukraine, with everything happening in the Middle East. If you look at what China is doing right now in the South China Sea, with regard to the Philippines, they’re doing that in no small part because they know they can get away with it because the US is overstretched. And I don’t see a dynamic in which the fact that the US is overstretched changes in the short run. And it’s, I think, obvious to everyone, including to people in Beijing. This is a reason for serious concern.

What would happen to TSMC in that scenario?

The moment conflict starts in the Taiwan Strait, you have to assume that TSMC shuts down very, very quickly regardless of what any of the players decide to do — regardless of whether anyone decides to disrupt the supply chain or destroy this or that or not.

Taiwan imports a big chunk of its energy and chip factories need energy. And there are a bunch of critical chemicals and materials that are imported into Taiwan, and those would stop. What’s more, you couldn’t get the ships out of Taiwan if there was a shooting war going on. And so your incentive to produce a lot also declines very rapidly if you can’t actually sell chips or get them off-island.

People like to speculate who would blow up what. That’s probably a lot less relevant than it seems, because the supply chain would end up being a lot more fragile than a lot of people realize.

Can you say a little bit more about that?

Taiwan imports a lot of liquid natural gas. Its power grid depends on it. Their publicly reported data says they’ve got 10 or so days of LNG storage on-island, which means all you have to do is turn on a couple of chip factories and suddenly Taiwan is in a power crisis. And then what do you shut down first if there is a limited supply of power? Even if there’s no shooting — if there’s just Chinese block tankers entering the harbor — that alone could be enough of a catalyst to force some really tough decisions about who gets energy or who doesn’t get it in Taiwan. And chipmakers are a major user of power.

On the chipmaking material side, there are a lot of key materials used in making chips, photoresist for example, and many others are produced by Japanese suppliers in particular. Having a steady supply of those is critical for chipmaking and those aren’t going to be shipped if there’s a war going on.

When people understand the true significance of TSMC, they’re often pretty shocked at how precarious the chip situation is. Do get that reaction a lot?

Yea. I mean this is why it’s so problematic that we’ve found ourselves in this situation because you can’t really sugarcoat it. The more you learn, the more worried you are. The more you realize that everything depends on peace being preserved in the Taiwan Strait — it just illustrates the stakes. This is a dispute that implicates every aspect of the American economy in a potentially catastrophic way. There’s not a silver lining — that’s the reality. It’s dark and it’s high stakes.

Nvidia CEO Jensen Huang is paranoid about competition. Do you think companies in that position should be worried about geopolitics, too?

My sense in talking to a lot of people in the tech sector is that they believe the likelihood of a military escalation or outright war is lower than is commonly believed among people who spend their time thinking about these issues in places like Washington, Tokyo and other capitals. So there is a mismatch in the predictions.

If you’re in the Silicon Valley consensus and tell yourself ‘Oh, China’s going to be GDP-maximizing. They’re focused on their economy, they’re not going to do it,’ then you worry less, because you just think it’s not going to happen.

But suppose you think there’s some chance that it happens. What do you do about it? For some companies, there are good options to diversify. But for a lot of companies, there really haven’t been great options in the past. Over the next couple of years, as more high-end chipmaking capacity is built in the US in particular, it will be interesting to watch companies like that and see whether their decision-making changes.