Speaker A
These are the chips powering a transformational technology reshaping how we live, work, and communicate.
Explores the black market for US-designed AI chips, US export controls, and the tech rivalry between the US and China in AI development.
Key Takeaways
- US export controls aim to maintain AI technological leadership by blocking chip sales to China.
- A significant black market exists to smuggle restricted AI chips into China, undermining export controls.
- Taiwan’s semiconductor industry is geopolitically critical in the US-China tech rivalry.
- China is using multiple tactics, including smuggling and legal loopholes, to access advanced AI chips.
- The AI chip race is a core element of broader geopolitical competition between the US and China.
Summary
- The video investigates the critical role of advanced AI chips designed by US companies and their importance in the global AI race.
- US export controls restrict the sale of these chips to China to maintain technological and geopolitical superiority.
- A thriving black market has emerged to smuggle these chips into China, involving both small-scale and large criminal operations.
- Taiwan is a crucial hub in the semiconductor supply chain, adding geopolitical complexity due to China's ambitions toward the island.
- Nvidia dominates the AI chip market, with its GPUs essential for AI model training and inference, but US sanctions have cut off sales to China.
- China is aggressively pursuing ways to circumvent US restrictions, including smuggling, exploiting legal loopholes, and state-backed diversion campaigns.
- The US export control regime has historic roots in Cold War policies aimed at limiting adversaries’ access to critical technology.
- Recent legal cases highlight vulnerabilities in the export control system and the scale of illicit chip diversion.
- China’s AI development is being forced to innovate domestically due to restricted access to US technology.
- The AI competition is accelerating rapidly, with the US currently holding a significant lead, but ecosystems in the US and China are diverging.
Chapters
- 00:00Introduction to AI Chips and Geopolitical Stakes
- 00:47US Export Controls and the Black Market Emergence
- 01:53Why China Desperately Wants US AI Chips
- 02:51Historical Context of Export Controls
- 04:05Taiwan's Role in the Semiconductor Supply Chain
- 05:24Nvidia's Dominance in AI Chips
- 07:04Impact of Export Controls on Market and China
- 07:51Chinese Tactics to Circumvent Export Controls
- 08:54Scale and Nature of AI Chip Smuggling
- 10:15Legal Cases and Export Control Challenges
Full Transcript — Download SRT & Markdown
Speaker A
They sit at the heart of a new race between Washington and Beijing. The outcome will help determine each superpower's status in a changing world.
Speaker A
And right now, the most advanced semiconductors are designed by US companies. They are also highly sought after by China.
Speaker A
The chips that go into training and running models are the most complicated things that humans make. These are maybe the most valuable goods in the world.
Speaker A
And US export controls are meant to prevent their sale to China. But a thriving market for smuggled silicon has emerged. The stakes are enormous. China cares both from a technology point of view, a business point of view, and a
Speaker A
political point of view. We're going to show you how this illegal trade works. This guy is claiming that he's just taken stock of a big load of H200s that's come through Hong Kong and is going to be delivered to Shenzhen.
Speaker A
We will also reveal how the controls are tearing Chinese and US technology companies apart and forcing Chinese chip makers to innovate. It has also opened a new sphere of competition between the two superpowers.
Speaker A
But first, let's explain what export controls are and why they're being used by Washington.
Speaker A
Why is China so desperate to acquire US-designated chips? The reason is obvious. AI is truly transformative technology, and it's already changing how we fight wars, run our government, and operate companies. It's essential for the United States to maintain a decisive lead in
Speaker A
the AI race. In 2022, the US government introduced blanket bans on the sale of the most advanced chips to China.
Speaker A
These chips were invented in America. Let's get that straight. They were invented in America.
Speaker A
And we used to make 40% of the world's chips. Export controls were born in the aftermath of the Second World War when the world was divided along ideological lines. The US and its allies on one side, the Soviet Union on the other.
Speaker A
Both sides were locked in a cold war and a nuclear arms race. Washington knew that technological supremacy would ensure military dominance. So in 1949, it created rules to stop Moscow from obtaining critical Western technology.
Speaker A
But the fall of the Berlin Wall and the dawn of the 21st century created a new strategic rivalry, this time between the US and a rising China.
Speaker A
Many in Washington see Beijing as a threat to the US-led world order and China's AI development as something that must be curtailed. It has turned to this cold war era rule book to stop its new rival.
Speaker A
Kevin Kurlin spent 28 years at the Department of Commerce. Most advanced chips and certainly AI chips that are designed by US companies and many Western companies are virtually all fabbed or produced in Taiwan, usually at TSMC, and then those
Speaker A
products either get packaged in Taiwan or in third countries and then from the packagers they're directed then to either send the items back to the designer or many times they're just directly sent to an integrator, if you will, or a distribution chain. And so
Speaker A
in order to harden these supply chains, you really need to have an export compliance program that not only addresses the activities that are happening, for example, in the United States, but their distribution networks and who their customers are.
Speaker A
In this tech war between China and the US, Taiwan plays a critical role. The island has emerged as an essential node in the global semiconductor supply chain. That Beijing wants to reunify with Taiwan has only added to the
Speaker A
geopolitical significance of this story. I've witnessed Taiwan's growth into an AI powerhouse. First as a student and later as a reporter, I've returned to Taipei to attend Computex, the world's most important chip conference.
Speaker A
What is on display here today is the most leading-edge AI products which are not allowed for export in China. These AI chips have become more and more powerful, more and more capable of training bigger, more powerful AI models,
Speaker A
and the American government has stopped Chinese companies from getting their hands on these chips.
Speaker A
So this is one of the racks that go into the large data center product. It's full of Nvidia GPUs that are essential for model training and inference, i.e., the process of running AI models, and these chips are among the series that are
Speaker A
restricted for shipment to China. Chris Maguire has advised both the Biden and Trump administrations on China and emerging technologies.
Speaker A
Computing power is the core input into AI. It is the reason that training and developing these models is the most expensive thing that humans are doing right now. Nvidia is just the example frequently because they are so significant. They are somewhere between
Speaker A
80 to 90% of the AI chip market. The reason for that is that their products are extremely good. They're very, very good chips. They're very, very powerful.
Speaker A
Nvidia has very, very robust supply chains where they can ensure they can really kind of maximize their output and ensure delivery on time to customers as well as they can given the kind of supply constraints that they have.
Speaker A
The world's most valuable company is now at the heart of geopolitics. The United States is ahead in the AI race and we have to stay ahead in the race. So we've taken China out of our forecast. We're currently
Speaker A
shipping zero dollars. Our market share has gone from 95% to zero. We're currently assuming China is going to be 0% of our business.
Speaker A
But winning the AI race has become a priority for Beijing. President Xi Jinping is marshalling both the state and private sectors to catch up with the US.
Speaker A
The Chinese Ministry of State Security has recently said that circumventing the US technology blockade is one of its top priorities. It's certain that there are diversion campaigns that are being facilitated by, you know, the Chinese intelligence services in order
Speaker A
to support their technology companies. This also is not surprising. This is an asset that they absolutely need to stay competitive that they cannot make themselves. And the only way that they can actually remain competitive is by either
Speaker A
convincing the United States to give it to them, smuggling it, or exploiting loopholes in the export controls that allow them to continue to have access to them. Without those things, there would be a very significant drop-off for Chinese AI
Speaker A
capabilities. And I think the Chinese government recognizes that and is pulling out all the stops in order to prevent that from happening. Shares of Super Micro Computer are losing about one-third of their value. It's because of an international smuggling scandal.
Speaker A
In March this year, the US Department of Justice targeted employees at Super Micro, a Taiwanese company and Nvidia partner. They stand accused of smuggling $2.5 billion worth of Nvidia chips and servers into China. The US alleges that a front company was used to divert the
Speaker A
chips. Super Micro was not named as a defendant in the indictment and has denied any wrongdoing, as have the named defendants. Still, analysts say that the case highlights historic failings of the export control regime.
Speaker A
The smuggling really ranges from kind of mom and pop smuggling operations to very large systematic criminal organizations. Nvidia makes somewhere between 5 and 10 million of these chips per year. If there's three or four that go out in a backpack, obviously you want
Speaker A
to stop that. But if you have organizations that are funneling tens of thousands of chips, that's a significant problem. And I think that we've seen organizations are committed to that because this is an extremely valuable industry.
Speaker A
So, I've been reporting on the various tactics Chinese companies have used to get their hands on American AI chips for about four years now. It's been an evolving story and the companies have been using ever more creative loopholes, some
Speaker A
legal, some illegal. Last year, we published an investigation that provided a snapshot into how these large-scale
Speaker A
into China. The estimated number of of chips sold here on the black market is far higher than that. But really this story provided an insight into how the larger scale operation works.
Speaker A
So I'm on Xiaong Shu here which is a Chinese social media platform. And if you type in Nvidia or AMD, you'll come up with a lot of these kinds of posts by middlemen selling these smuggled chips.
Speaker A
Here we have H200's. There's also someone selling B300's, which is a more advanced Nvidia series. And if you go onto their profiles, these guys are very clearly advertising their sales. This guy is claiming that he's just taken stock of a big load of H200's that's
Speaker A
come through Hong Kong and is going to be delivered to Shenzhen. Shenzhen is important for the chip smuggling story for a variety of different reasons. The first of which is it's a port city. It's close to Hong Kong which is a thorough affair for a
Speaker A
lot of these smuggled ships. It is also a place where there's a lot of AI hardware. So, server builders, rack builders. It's somewhere where the chips are just being sold and repackaged to the rest of China. There are a lot of
Speaker A
tech salesmen based in Shenzhen. It's a hardware hub in China. And so, if you've got requirements for new computers, new servers, you go down to Shenzhen, you speak to these middlemen, and they will kind of create you a custom solution.
Speaker A
We are here at Huatang Bay in Shenzhen, a famous electronics market. You can find any electronics component here, including lots of people selling secondhand chips. So, we're going to go and see if we can find anyone selling Nvidia GPUs in here today. The market is
Speaker A
a labyrinth of retailers selling all kinds of consumer electronics and the components that go into them. I'm here to talk with a chip reseller. He's agreed to be interviewed, but not on camera given the sensitivity of the topic.
Speaker A
Until recently, it was also possible to buy smuggled AI chips uh from Nvidia and AMD. But in recent months, it's become incredibly difficult to get access to these because the Chinese authorities have restricted imports of these chips in order to support local competitors.
Speaker A
That means that the prices of these have gone absolutely crazy. We've spoken to a number of sellers in this building and they said that prices have gone absolutely crazy, exacerbated also by the memory chip shortage. But for example, one uh seller said that one
Speaker A
B300 machine that is one of the leading server machines has almost doubled in price in recent months. Chinese companies are also exploiting loopholes in the US export controls by remotely accessing restricted chips.
Speaker A
Right now, the administration has not articulated a coherent strategy for controlling advanced compute and AI chips. The most advanced NVIDIA chip, say a Blackwell that is located in a in a data center in Malaysia, there's no restriction on any Chinese company from
Speaker A
accessing the compute power on that item. So, there's an inconsistency there. There's also a loophole in the current regulations that do not restrict a Chinese front company in a third country or a Chinese-owned data center in a third country from buying the same
Speaker A
chips or servers that are restricted for sale directly to China. So on the one hand you have the Chinese AI labs who want to build the best models and it just so happens that building the best model means access to
Speaker A
Nvidia GPUs. But on the other hand, you also have Chinese tech companies that are trying to support the growth of domestic AI chips like Huawei and Capricorn. And their argument is that we should be supporting the growth of these
Speaker A
domestic chips. And if that means relinquishing some progress or lead on AI in the short term, that's okay. We should be doing it for the long-term goal of Chinese AI independence.
Speaker A
Fore I don't buy that uh export controls have are are driving China to self-sufficiency. They had already put out a policy to do that. To me, again, export controls are meant to maintain US leadership by delaying their ability to
Speaker A
achieve the same goals or or or levels of innovation um that allow the United States uh to continue to be a leader both from an economic and a national security perspective.
Speaker A
The question is ultimately will China be able to reach a level where Huawei or one of the other Chinese AI companies is equally or even better than the American companies. And I think that's still a massive open question. It looks like a
Speaker A
no for now, but you know, China has proven time and time again that it's very capable of building cutting edge technology.
Speaker A
Right now, the US leads by about 8 months in AI. That's per the latest US government estimate. It also means is AI is doubling in capability about every four months. So, an 8-month lead means about a 4x capability difference. That's
Speaker A
significant, but it's also not so significant that I think the Chinese say, "Okay, well, we're down and out, right? They're they're looking at this like, we have an opportunity to catch up. We're behind, but we think we can
Speaker A
close the gap." We certainly have become obsessed with this idea of a time lag between the US and China. To me, I don't think it matters a huge amount right now. What certainly does matter is whether or not,
Speaker A
you know, the US is the first to build this fated future of super general intelligence. And if and when we get to that stage, it certainly looks like America is going to be the first to get there given the trajectory of the model
Speaker A
development at the moment. The United States just needs to recognize that the US and Chinese AI ecosystems are fundamentally diverging.
Speaker A
That is exactly how it was in the Cold War. The US and Soviet nuclear programs in no way helped each other. They were completely separate. Uh and that is undoubtedly in the national security interests of the United States. that
Speaker A
also did not preclude cooperation between the United States and the Soviet Union on certain discreet areas where we had shared interests. And I think in AI we probably need to think about cooperation with China more in that sense.
Speaker A
For now, the US dominance in AI hardware is clear. There are several choke holds that Chinese companies have not yet broken. It will take time and a lot of money to build a rival ecosystem to the one Nvidia and its partners have
Speaker A
established. But what we have seen time and time again is China build technology that the world said was outside of its grasp. We know that Beijing is pouring investment into becoming a self-sufficient AI state. Can it succeed? This is a story that will
Speaker A
define the world's future.
Topics:AI chipsexport controlsblack marketsemiconductorsNvidiaTaiwanUS-China rivalrytechnology smugglingAI racechip supply chain
