Now to an affair that's shaken the tech world.

And now Donald Trump says the FBI is investigating.

NVIDIA CEO Jensen Wang has admitted to having an affair with an alleged Chinese spy.

Now his company designs computer chips, software and AI, earning the favor of the U.S.

president.

With Trump and Chinese President Xi expected to meet next week, professor and author Chris Miller joins Walter Isaacson to discuss the NVIDIA scandal and the implications it might have on U.S.

security amidst the AI boom.

Chris Miller, welcome to the show.

Thank you for having me.

In your book, Chip War, you call it the world's most critical technology.

The big news now in the chip world is of course the investigation of Jensen Huang, the head of NVIDIA, the most valuable chip-making company in the world, by his own board for an alleged relationship with somebody who might be a Chinese spy.

What is the significance of that to the chip world and to the U.S.

relationship with China?

Because it's an American-based company that sells chips to China.

Well, it's a shocking allegation.

It certainly shows the lengths at which China is willing to go to get access to these ultra-critical semiconductors.

But I think it's also going to spark a rethinking in the Trump administration as well as at NVIDIA of their approach to China.

For the last couple of years, NVIDIA has been lobbying for more access to the Chinese market, and Trump's been very open to giving the company more access.

And I think the new Congress in particular is going to call that into question and push for tougher rules on the export of chips to other countries, including to China.

And Jensen Huang has been one of those tech leaders that threw himself very close to Donald Trump.

He's been around Donald Trump, donated to things.

To what extent do you think Donald Trump is going to defend Jensen Huang?

Well, we've seen some early defense from the president, but I think we should wait and see how the investigation develops and also how the rest of Washington approaches this case.

It's already been reported that Congress wants the NVIDIA CEO to come testify to it, and I think we should wait and see what new news actually emerges.

Secretary of State Marco Rubio oversaw the drafting of a classified memo about Jensen Huang.

What do you think Rubio's role will be in terms of trying to push Trump on what should be done on the Jensen Huang and NVIDIA, and for that matter, the whole China situation?

There's been an active debate in the White House about whether Trump's China strategy has been too lenient when it comes to artificial intelligence.

There was a lot of disagreement in the White House when Trump began loosening restrictions over AI chips.

And I think this news combined with the midterm loss is likely to push Trump in a less lenient direction going forward.

And these sort of tremors are happening right when President Trump is preparing to meet this month with President Xi.

To what extent do you think that summit is going to revolve around issues such as microchip access to microchips and AI.

And what would they talk about?

I think this scandal is only going to further limit Trump's ability to offer concessions to Xi on the issue of chips.

Trump has been trying to do so for the last couple of years.

He's offered some concessions.

He's been hemmed in by Congress in offering others.

But now I think Trump's hands will be largely tied on this issue.

But it's worth noting Congress has been advancing several pieces of legislation that would further limit the President's discretion in terms of offering waivers or improved access to advanced semiconductors.

So I expect that across Washington, including in the White House, this will lead to a tougher line when it comes to China and AI.

The Chinese Foreign Ministry has denied that this was an espionage case.

I don't know how much inside information you have, but to what extent do you think it's clear that this was Chinese espionage?

I don't have any inside information, but media reports suggest that it wasn't only U.S.

intelligence, that Taiwanese and Singaporean intelligence were also tracking her, which suggests that there was a substantial base of evidence that this conclusion was based on.

Well, let me go back to the basics here, because we've been talking about microchips, the most critical technology.

Explain to me exactly what a microchip is and why it's so critical, especially in this era of AI.

A chip is a piece of silicon that has millions or now billions of tiny circuits on it that enable all of the ones and zeros that undergird all digital computing.

For AI, you need a lot of math, a lot of ones and zeros, and that's exactly what NVIDIA's chips are perfectly primed to do.

It's already been a decade that NVIDIA's been producing chips specially tuned to AI, and so the company was perfectly placed when the release of ChachiBT led to a surge of investment in data centers full of the world's most advanced chips.

That's why NVIDIA has grown to become a $4 trillion or $5 trillion company as all of the world's leading tech firms spend tens of billions of dollars buying the company's chips.

So when we talk about these chips, which are really just on/off switches, transistors, etched on a piece of silicon, you say, which is a semiconducting material, meaning it can be doped to be on or off, so it undergirds.

It's like the dynamo of this new revolution, the digital revolution.

But tell me, why is it that chips are so hard to make, and why aren't they made in the United States more?

Well, these tiny transistors that you reference are today microscopic in size, measured in just a handful of nanometers.

That's billionths of a meter in scale, yet they're produced by the trillions and trillions.

There's nothing that comes close to the complexity of manufacturing of these chips.

And today, there's just one company in the world that actually does the manufacturing.

NVIDIA itself, like many chip firms, only actually designs the semiconductors.

And that's a very hard thing to do on its own, but it works closely hand in hand with the Taiwan Semiconductor Manufacturing Corporation, which manufactures almost all of the world's AI semiconductors.

And so we've got a situation in which NVIDIA is the dominant designer of AI chips, and TSMC in Taiwan produces almost all of the world's chips based off these designs.

So it's an extraordinarily concentrated supply chain that links up Silicon Valley with Taiwan.

And that, of course, adds a complexity, meaning the Taiwan question with China.

The fact that most of these chips are done in Taiwan, to what extent does that play into this meeting President Trump is going to have with Xi to try to figure out how to calm down the issue of whether Taiwan has its autonomy retained from China?

What's a hugely important question that's only gotten more important as the US becomes more dependent on Taiwan for AI chips.

This year, the United States is on track to spend more money importing goods from Taiwan than it imports from China.

An extraordinary data point.

And it's all about these critical AI semiconductors.

And that's why I think it's so important that President Trump be clear to China's leadership and Xi Jinping that the security of Taiwan, the defense of Taiwan is critical not just for Taiwan, but also for the United States.

Well, let me drill down on this question of why we can't build them in the United States.

The Taiwan semiconductor company that you referred to is moving some operations, I think, to Arizona.

Do you think they will have more production capacity in the United States real soon?

TSMC is certainly building capacity in the United States.

They have one factory up and running in Arizona, producing close to the leading edge of semiconductors, and they're building more as we speak.

By the middle of the 2030s, they might have a quarter of their global production in the United States.

And that's a substantial change from just a couple of years ago when we had 0% of advanced ship production in the United States.

But because the manufacturing is so complex, because it requires so many advanced materials, and specialized tools, this is an ecosystem that just can't change overnight.

It requires billions of dollars of investment, unique types of talent that just can't be straightforwardly replicated.

And so it's going to be a long time in which the United States and Taiwan are deeply intertwined in producing these advanced semiconductors.

We've just gone through midterm elections where somewhat in a real wave a resistance to AI, a resistance to data centers, and perhaps even a resistance in Texas now that it went Democrat, Democratic leaders across the board to building things like a TeraFab factory.

How do you explain this resistance to things like data centers, and do you think it will affect even the fabrication of chips in the U.S.?

Well, I think data centers and chip plants are in many ways quite different.

There's a large number of data centers being built.

There's a small number of chip plants by comparison.

Chip plants provide jobs over the long run.

They build local ecosystems.

And so you've seen in Arizona, in Texas, and other places, a lot of local support for building more chip plants.

That's not universally the case, but that's been the trend.

The challenge, though, for the chip industry is that they need data center growth to buy all the chips they're producing.

And if local political leaders in Texas or elsewhere slow the rate of data center construction, that will have inevitably an impact on the number of chips that can be profitably produced.

We talk about the US tried to get into the game of building chips, Taiwan dominating it.

Why doesn't China have the ability to build more of these chips?

China wants to China's been investing tens of billions of dollars each year for now, well over a decade, but it can't figure out the manufacturing of the chips themselves or the ultra complex tools that are used to make these chips.

And if you go inside of one of TSMC's factories in Taiwan, you'll find that they import equipment from all over the world, specialized machines from Japan, from the United States, and most importantly from the Netherlands, where one company called ASML produces almost all of the world's advanced lithography tools, which are critical in making advanced chips.

And China can't make these tools at home, and the US and Netherlands won't let them buy these tools either.

And that's why China's chip manufacturing is a tiny fraction of Taiwan's, despite this huge investment.

Just one data point.

This year, best estimates are that China will produce something like 2 or 3 percent of global AI chips.

Taiwan's going to produce well over 95 percent of the world's AI chips.

So this difficulty that China is having in terms of fabricating chips, to what extent is that caused by the success of restrictions that the U.S.

has tried to put on advanced technology going to China?

Well, I think that's exactly it.

If China had been able to buy these ultra-advanced chipmaking tools when they started to try to in 2018 or 2019, I think we should expect that Chinese firms would be among the best in the world.

And it's worth noting, if you separate the processor chips that NVIDIA produces from the memory chips, what you find is that China has made a lot more progress in catching up in memory chips precisely because the leading Chinese producer of memory chips, a firm called CXMT, is allowed to buy many of these advanced tools.

So where the U.S.

has restricted tool sales to China, China has struggled to reach the cutting edge of manufacturing.

The U.S.

has a great advantage in chip design.

That's what NVIDIA and other companies, including Intel, was all about.

Taiwan has this huge advantage in being able to fabricate them.

But China has certain advantages too, including the processing of rare earth materials.

To what extent can we really decouple this complicated supply chain that, as you say, goes all the way from the Netherlands to rare earth minerals in China to fat plants in Taiwan to design in the United States.

I think a complete decoupling or separation is not going to happen, but we've now got a decade of active competition where Beijing and Washington and Tokyo and Seoul and Taipei are all trying to improve their competitive positions in these supply chains.

So you mentioned rare earths.

These are critical materials in making the tools that make chips.

And China's been restricting the flow of these rare earths, especially to Japan, which is a key node in the supply chain for processing these advanced materials.

And so one of the key questions that I think Trump will face as he meets Xi is, can he actually get China to sell more of the rare earths that it's promised to sell?

Right now, China is not following through on the deal that was struck last year, whereby it would sell more rare earths.

And if that doesn't change, we might face supply chain difficulties across the manufacturing early as next year.

Well, we always talk about rare earth minerals, but help explain it to me, which is what minerals are these and why can't we be doing them in Colorado and Arizona?

Well, there's an array of different materials that are used for different things.

Gallium and germanium used in manufacturing certain semiconductor devices, but the most important today is called yttrium.

China's the leading producer, almost the sole producer globally, and it's not really because of mining capacity, it's because of processing capacity.

China's, for the last several decades, invested in this processing capacity.

It's subsidized companies to do so in ways that might not have been profitable outside of China.

And today it has a dominant position in this material, which is critical both for chip making, for aviation, and also for certain medical devices.

And so there's a race right now in Japan and the United States and other countries to bring online new processing capacity faster than we run out of our existing stockpiles of these materials.

This whole conversation has really revolved around the difficulty of supply chains.

And do you think all of this is going to help push countries to onshore things more, to manufacture more things in their own country, to process more minerals in their own country, instead of relying on global supply chains?

I think to some extent that will be the case, but I think this also shows one has to be very specific about which node in the supply chain you invest in.

You can't do everything domestically.

There's just no chance that's going to work.

Why not?

It's just too expensive.

It's too hard.

The cutting edge of manufacturing requires bringing together the most advanced capabilities from Japan and from Europe and from Taiwan and from the United States.

And so self-sufficiency is a pathway towards technological backwardness.

And we're better off if we're able to work with allies in Europe, in Japan, in Taiwan, and in Korea to manufacture even better products taking advantage of that R&D.

In a recent Washington Post column, you warned about what's called the Dutch disease when it came to AI.

And the Dutch disease, I think, is when they had a boom in natural gas and they got all excited and they neglected other industries.

Explain to me why you think that's a problem right now with the AI boom.

We've seen this surge of investment into artificial intelligence, building data centers, buying chips, and that's driven up the price of electricians' wages as they work more on data centers, as well as chips.

And so if you want to buy an iPhone right now, it's more expensive, largely because of the AI boom driving up the price of components inside of it.

And I think a lot of the backlash against AI is partly explained by the fact that people are paying higher prices for certain types of components and the products that use them without yet really feeling the benefits of this technology.

Do you think there is an existential danger, as some of these tech leaders say, of putting the brakes on data centers, putting some guardrails around AI, and thus allowing us potentially to fall behind China, as President Trump has warned?

Or do you think we really should now be figuring out how to get our own microchip house in order, get our own AI with the right guardrails, and doing this in a more prudent fashion?

I think guardrails are a reasonable thing to pursue.

The question is, what type of guardrails are we looking for?

It seems to me that if you're going to talk about making AI systems that are higher quality, that aren't going to do things you don't want them to do, that's a better AI system, not a worse AI system.

When AI systems go out and hack autonomously when they're not supposed to, that's a system that malfunctions.

And so I don't think it's right to see better controlled systems and progress as dichotomies.

What we want are AI systems that do what they're told and don't do what they're told not to do.

Thank you so much for joining us.

Appreciate it.