China and the U.S. Are in a Race for AI Supremacy

A new Cold War has broken out between the U.S. and China over artificial intelligence. The U.S. and China are locked in a competition, but there are some really striking similarities to the tech races between the U.S. and Soviet Union that we saw during the Cold War. That's my colleague Josh Chin. The sense that you have two rival superpowers who are competing in technologies with really, really broad potential applications. During the Cold War, the Soviet Union and the U.S. were the two strongest countries in the world, and they were racing against each other to create transformational technologies like nuclear weapons, rockets and satellites, and advanced computing. Josh says that history is repeating itself with AI. You know, this is the first general-use technology we've seen come along since the Internet, and so it affects potentially everything. There are people in both Washington and Beijing who see this competition over AI having similar consequences that, you know, whichever country manages to run away with a lead stands to have reached just humongous advantages in economic and military and scientific power as well as like global influence. And like with the Cold War, the AI race also has the potential to accelerate a technology that many fear could have dire consequences. What happens if we create a superintelligence that doesn't have humanity's best interests at heart? There's a lot of people in government, in industry, in the military who worry that this race dynamic is actually leading both sides to sort of downplay safety concerns, which are really significant with the technology that's powerful. Welcome to The Journal, our show about money, business and power. I'm Ryan Knudsen. It's Tuesday, December 2. Coming up on the show, the AI Cold War. In 2017, Beijing announced that it planned to lead the world in AI by 2030. But back then, the country was focused on a specific aspect of AI, something called computer vision. So, computer vision basically is exactly what it sounds like. It's a technique in which machines teach themselves how to recognize objects, you know, from coffee mugs, cats, dogs, airplanes and people's faces. It is what's being used like when your phone is like categorizing your photos for you. That's computer vision. Why was Beijing investing so much in facial recognition technology? That's an area that China, because it has this huge surveillance state, was really interested in because it's just a much faster way to kind of track down people of interest as they call them, which would be, you know, criminals, political dissidents, that sort of thing. But in November of 2022, the U.S. jumped ahead when open AI released Chad GPT. The AI landscape, as we know it, was rocked and forever changed, I think you can say, by a single chat box. That GPT, which stands for Generative Pre-Train Transformer. Essays, philosophical questions, even therapy. Chad GPT is a computer program that will write whatever you want quickly and convincingly. Chad GPT is a large language model, or LLM, that has much broader applications than computer vision. And since then, U.S. companies have been pouring tens of billions of dollars into AI development, and specifically into LLMs. And after Chad GPT, the advancements kept coming. After open AI launched Chad GPT, Google started promoting its model, Facebook launched LLMA, Anthropa came out with Claude. It was basically all American companies across the board. There were really no major competitors anywhere else. When the Soviet Union launched the Sputnik satellite in the 1950s, it jolted the U.S. to speed up its own work on the space race. And the debut of Chad GPT was a similar wake-up call for China. That's when top Communist Party officials really started to ramp up pressure on Chinese tech companies to do more. One major AI company told us that it received calls from 10 different government agencies in the space of a month as the Communist Party was trying to catch up. Saying what? Like get to work on AI, please? Yeah, basically. Beijing strategy is to throw everything it has at the problem. That includes promoting AI in schools and at private companies, and offering generous subsidies to industries that power AI development, like infrastructure and cloud computing. So it's building computing clusters in these remote places with lots of cheap renewable energy like Inner Mongolia, and eventually wants to connect like hundreds of data centers nationwide into what calls a national cloud for training and running models. It's putting like billions of dollars into expanding its power grid. Beijing has also been mobilizing local governments, which act essentially as the Communist Party's middle management. And it's basically signaling to local governments, you know, that this is the big priority. So the local governments are going to the companies in their areas and really trying to get them going on AI efforts. This isn't the first time China has done something like this. When the country wanted to accelerate development of electric vehicles and renewable energy, it adopted a similar strategy. I mean, a lot of economists will say, will tell you that like there's there's a ton of waste in this approach, in part because you have all these local governments all trying to compete with each other for Beijing's favorite. But in some industries like EVs, for example, or renewable power, it does have the ultimate effect of making China very competitive, right? And also making it really hard in the global marketplace for companies from other countries to compete just because other countries don't have the level of state support for companies. And earlier this year, finally, there was a breakthrough. A new China-based artificial intelligence startup is shaking up an industry known for its rapid innovation. It's called DeepSeq. DeepSeq, the Chinese artificial intelligence upstart that rattled the American AI industry and the markets this week. DeepSeq, the Chinese artificial intelligence startup, dethroned ChadGPT on Apple's AppStore to become the store's top amp. A Chinese startup named DeepSeq released a large language model. It managed to build this model that was basically almost as good as the best model that open AI had at the time. DeepSeq hadn't initially benefited much from China's centralized AI efforts. It had largely been bankrolled by its founder's hedge fund and marched to its own beat. Still, what was notable about DeepSeq was that it managed to build a model using much less computing power and a fraction of the money compared to the open AI and other U.S. companies. There's some debate about some of their claims in terms of how they did this, but the bottom line is they figured out how to engineer chips and algorithms in a way that just massively improved the efficiency of training a large language model. Mark Andreessen, who's arguably Silicon Valley's most visible venture capitalist, comes out and calls it one of the most impressive accomplishments he's ever seen. In the aftermath of DeepSeq, U.S. investors freaked out. The Nasdaq lost 3% of its value as DeepSeq's model cast out on whether all the spending by U.S. tech companies was necessary. What did the DeepSeq moments say about the status of the AI race between the U.S. and China? Well, I mean, after DeepSeq, it basically, to a lot of people, the gap had closed. It had almost closed completely. Some people, even what's of ours to say China was on the verge of pulling ahead. The lead was basically in the sort of months, not years category. Now that it's developed a proven model, China has put state power and support behind DeepSeq. Does China have everything it needs now to get ahead in the AI race? China has a lot of what it needs. I mean, it has tons of engineering talent. You know, AI relies a lot on data and China has huge amounts of that just because it's a massive country and has pretty plentiful power. Those are all working in its favor, but the one thing it really doesn't have is chips. What China is trying to do about computer chips is next. Advanced computer chips are the secret sauce when it comes to the U.S.'s lead in AI development. So the best models were trained on huge numbers of very expensive cutting-edge chips. And American companies made the best chips, and American investors had the financial power to buy them. Back in 2022, under the Biden administration, the U.S. decided to cut China off from those advanced chips. The government restricted exports, hoping to hobble China's ability to catch up. Most of those restrictions are still in place. So what is China doing about the chips problem? Yeah, so the biggest burden for all of this for China right now falls on one company, which is Huawei. Huawei is a giant Chinese tech company that makes things like smartphones and wireless equipment. It also makes the best chips among Chinese companies. And Huawei has a plan. What Huawei is doing is trying to link together as many as a million of its chips to produce these systems that, according to Huawei, offer similar levels of computing power, which is a strategy that some industry in China refer to as swarms to beat the titan. Hm, swarms to beat the titan, basically just like mass produces many lower quality chips, stitch them all together, and that'll compete with the big ones the U.S. has. Yeah, I think it's, you know, it's a strategy that China applies in a lot of ways, right? It's just trying to take advantage of its capacity to produce relatively good quality things in huge quantities in order to sort of compete with the more exquisite, but less numerous American technologies. In response to these moves from China and to stay ahead in the AI race, the U.S. government has taken a page out of China's playbook and adopted some elements of state-run capitalism. I mean, traditionally, the U.S. has kind of let the market take the lead, right, in private enterprise. Although, you know, increasingly, you're starting to see the U.S. kind of adopt elements of the Chinese sort of state capitalist model, you know, the Trump administration, the House was taking a stake at Intel, one of the big chip companies in the U.S. It's also under the Biden administration, we had the CHIPS Act, right, which pumped a lot of money into building chip factories in the U.S. So the models are sort of converging somewhat, although this still is a much bigger role, obviously, for private enterprise and private capital in the U.S. One consequence of this race is that in order to move faster, both countries are throwing a bit more caution to the wind and placing less emphasis on AI safety. For Silicon Valley's part, their argument, which has been pretty successful is that, you know, the threat of competition from China means that the government should take a pretty hands-off approach regulation-wise and kind of let them do what they're going to do, not slow them down, and that has generally been what's happened. The Chinese government has also relaxed some regulations. At one point, Chinese tech companies are required to prepare as many as 70,000 questions to test their models' answers for safety, though when China, these rules are mostly focused on content, including political speech. But now that testing process has been simplified, including allowing Chinese companies with a good track record to skip reviews of training data, according to the people involved in the process. The goal, of course, is to accelerate the arrival of all the potential benefits that AI could offer. Just like with the Cold War, this competition, you know, it feels likely to fuel a lot of innovation on both sides, and that could be a good thing, right, in terms of making the technology better. The risk, though, is that all this competition could instead accelerate the arrival of all the harms people worry AI could cause. I mean, you have the potential, for example, for non-state actors to ask AI to help them build bio weapons, right, or maybe more immediately, like, there's this fear of how AI could make warfare faster and deadlier. Or maybe an AI superintelligence turns the entire earth into paperclips. Sorry, that was an AI joke, if you know, you know. A lot of AI experts, especially AI safety experts, are still worried that that race dynamic will effectively make it very difficult to counteract some of the risks associated with this technology. So in this AI race between the US and China, who do you think is going to win? The most likely scenario is that the US maintains its lead for the foreseeable future. But you know, if China figures out how to close the chip gap or finds new ways to do more with less the way that DeepSeek did, you know, to build bigger, better models with fewer chips, or it finds, like, a path to new forms of AI that are more powerful than what we have now, then it's totally plausible the US could fall behind. I think closing the chip gap is a major hurdle, and I think there are a lot of people who are skeptical that China will be able to do that. But you know, like we said, China has surprised people before. That's all for today, Tuesday, December 2nd. The journal is a co-production of Spotify and The Wall Street Journal, additional reporting this episode by Rafael Flan. Thanks for listening. See you tomorrow.