Subscribe

Subscribe to our newsletter and be kept up to date on upcoming Aventine projects

Dan Wang spent six years as a technology analyst tracking China’s rise as a manufacturing and technology superpower while living in Hong Kong, Beijing and Shanghai. The experience became “Breakneck: China’s Quest to Engineer the Future,” a New York Times bestseller published in 2025.

The book argues that China is primarily an engineering state, fixated on building factories, power plants and transport infrastructure, while America, in Wang’s view, is a lawyerly society that “brings a gavel to block almost everything, good and bad.” These divergent histories, Wang argues, influence almost every aspect of how progress evolves in each country, whether it's developing AI models or building aircraft.

Wang, who was born in the Chinese province of Yunnan and grew up in Canada, is now a research fellow at the Hoover Institution at Stanford University, a move he made partly to be closer to the AI industry. He spoke to Aventine about how the ideas in “Breakneck” extend to AI, whether America’s data center boom complicates his book’s thesis, and why China still lags in some important areas of advanced technology. He also explained how an artificial intelligence crisis in China might finally bring Beijing and Washington together to collaborate on AI safety. 

This interview was conducted before reports that Jensen Huang, Nvidia’s CEO, had an affair with an alleged Chinese spy, so that topic is not covered. What follows has been condensed and edited for clarity.  

You argue that the United States, built and largely run by people with law degrees, is essentially a legalistic society. China, meanwhile, was built by engineers and has become a society that builds. What can each learn from the other? 

I think it’s easier for the US to learn engineering than for the Chinese to learn law. The Chinese have never really had much respect for individual rights or pluralism. The state has always monopolized the intellectuals in its own service. But the US built a lot of canals and railways and major cities and subways. It has the engineering heritage, the heritage of going to the moon and the Manhattan Project. So this is partially why I think that the US has greater capacity for self-reform.

The US feels like it has become an engineering state again through AI: It's building data centers and developing new models at lightning speed. Does that complicate your framework?

It’s too soon to tell. I think data centers have been the only thing that Americans have been able to build [in recent time]. Yes, [with data centers] there has been some greater physical construction. That's definitely the case. But that might slow, in part because there hasn't been enough construction of all the supporting elements [such as power generation]. When I look at the AI buildout in China, I think it has better joined up thinking about trying to build, for example, more electrical power. In the US, the emphasis still is too much on the software world. 

Historically, people have argued that the West does “innovation” and China does “scaling.” You’ve argued against that. Why are people wrong?

It is true that China has not produced blockbuster innovations like large language models or new sorts of drugs. But [China excels] when it comes to basically any physical product. It's China making the electric vehicle batteries, most of the world's iPhones, all of these really important electronics products. China is in some sense building the future, because it is where the factories are. Factory workers and their managers are solving three new problems a day before breakfast; people in the West don't even have a sense of what the problems are [in manufacturing]. And China is not just doing production. It is getting better at design, it's doing a lot more top scientific research. This idea that China was not doing much innovation may have been defensible up until, let's say, 10 years ago, but I don't find it very defensible today.

You say China is substantially behind in only two fields, semiconductors and aviation. Why the struggle in those areas?

China is relatively weaker in the complex integration of different scientific disciplines. Aviation is the integration of aerodynamics and mechanical engineering and material science; semiconductors is the integration of computer science and electrical engineering and chemistry. This is still a Western strength — to be able to work at the frontier of a lot of different scientific disciplines and integrate them. But China is getting stronger. It is doing better science, and I would say that the US and Europe are not improving their manufacturing at a comparable rate. So China is doing pretty well at patching up its own weaknesses, but the US and Europe are not patching up their own weaknesses.

China seems to have accelerated its biotech abilities dramatically in recent years, moving from being a generics manufacturer to a fast follower. Now licensing deals are surging too. Have you been surprised by how quickly that's happened?

I haven't looked at biotech extensively. It's pretty orthogonal [to other areas of technology]. But I’ve heard arguments for why this is the case. First, there's no such thing as a rare disease in China. There are hundreds of thousands of patients for whatever disease you care to mention. Second, around the time of Covid, China's regulatory system had a couple of reforms that just made it much easier to do clinical trials, which created this surge of experimentation. This is one area where you could potentially say that the Western world is holding itself back. Plus, China has a lot of talent, and there's a lot of manufacturing capability. It is perhaps surprising that China has not done as well as it has until recently. We should be primed for exciting new Chinese drugs to come out.

To what extent does China’s top-down governance impede its technological potential? What's the balance that China needs to strike between regulating and encouraging industry? 

What it has struck is a state that has been pushing on the gas and the brakes sometimes simultaneously. And there is a very robust, dynamic set of entrepreneurs who try to do their own thing and will not really listen to the government. The government has been trying to discipline these unruly people, which it has often not been able to. I think that is part of the magic of China: a strong state as well as strong entrepreneurs. The balance, I think, over the past 10 years, has swung a little too hard toward the state, especially as Xi Jinping delivered regulatory thunderclaps against China's most dynamic entrepreneurs, people like Jack Ma. Alibaba [Jack Ma’s company, which the Chinese government hit with antitrust fines, investigations and enforced governance changes] is still around one-half its market value off its peak five years ago. These things have not instilled confidence among Chinese entrepreneurs, and the state has grown much more frightening over the last couple of years.

The power dynamics between the government and the AI companies are very different in the US and China. What can the US learn from China in this regard?

The contrast is stark. In the US, people like [Nvidia CEO] Jensen Huang have been able to stand at President Trump's side and sort of dictate policy. It is impossible to imagine anything equivalent in China. Xi and the rest of the Politburo do meet with entrepreneurs, but these are always settings where the leaders are on a podium, facing the entrepreneurs. It feels much more like the leaders are lecturing to the entrepreneurs. They're not co-equal. I'm not sure what is right here. Probably these entrepreneurs should not be treated like children, as they understand the market and technology much better than the leadership, and better than the regulators. But it is probably not right, either, that anyone can reach Trump.

So a middle ground might be nice?

In principle, yes. Although we often treat the UK or Europe as some sort of happy middle ground with policies that don’t look so batshit crazy. And yet it is the US and China that lead the world. 

You’ve written that it's unbecoming for the US, as the world’s largest economy, to be so leveraged on just one technology, AI. Why the word “unbecoming”?

It's undignified. An American superpower needs to have many strengths. It cannot be so narrowly focused.

You touched on the idea that AI scale is increasingly becoming an energy problem. Given China’s massive rollout of nuclear, solar and industrial power infrastructure, could the US lose the race because it doesn’t have enough power?

Well, China lacks compute, and that is by American design. So China doesn't have enough chips, and the US doesn't have enough energy, also by US design. In the US, if you want to build a data center as well as [power generation], you run into these hyper-local permitting issues where local residents have a pretty defensible position and just want to sue your pants off. It is just difficult to build in the US. So I would say that both are capacity constrained. But in principle, the US doesn't have to be so capacity constrained. 

There’s a growing US populist backlash against AI. Why is the same not true in China?

In part it’s because people in China don't really protest that much against any sort of physical infrastructure, even pretty disruptive projects. It’s also in part because people are pretty proud of these companies. In China, companies like Alibaba and DeepSeek are viewed as national champions, while in the US the AI giants are pretty broadly hated. 

Dario Amodei and others want to slow the frontier, keep chips from China, and get China’s cooperation on pacing. Can the US ask a country to cooperate on safety while also actively hindering its progress?

I can imagine that. If these tools are as dangerous as Silicon Valley claims, and if certain Chinese models have fewer safeguards, as Anthropic claims, then China is going to run into its own safety incidents. Maybe its state-owned enterprises get hacked; maybe someone in China uses AI to spy on the Politburo. My expectation of China's reaction is that, if there is something that truly frightens the regime, they're going to crack down faster than you can say “Jack Ma.” The Chinese are masters of paranoid freakouts. Over the summer, a small propeller plane flew into Beijing's tallest building. It wasn't a 9/11-style attack, but the debris did cause some casualties and injuries among bystanders. What did Beijing do? It shut down all private aviation in an Italy-size zone across the city. When there's an incident, they will crack down pretty quickly, and that is the basis of pretty good negotiations, I think, between the US and China. So I'm an optimist here. There will be a big crisis. After that, you should expect some cooperation.

It seems like Chinese engineers and officials see AI as a tool to deploy, while in the US there is more of a focus on building AGI. Why does that difference exist?

I think there is that difference, although that has softened a little bit in the last couple of months. There are more Chinese entrepreneurs who are now also excited to build a Machine God. But for the most part, I think Chinese leadership has spoken about AI as a technology to be harnessed, one that is controllable. I think they take the precedent of the earlier days [of attempting to censor and control] the internet — which Bill Clinton scoffed at and said, "You're trying to nail jello to the wall.” Well, they nailed it pretty firmly to the wall. Initially, it was very plausible that the internet was going to be a political threat to the regime. In fact, they managed to pretty much enjoy the upsides as well as control the downsides. So perhaps from that basis, they have this confidence that they can pull another rabbit out of the hat. 

You’ve said America doesn’t need to become China but should study its playbook. Which parts should the US choose?

Some degree of resilience — energy resilience, food resilience, building up a manufacturing base so that it's not just-in-time. Having an idea that manufacturing and technology are important. Building, a lot. And finishing infrastructure projects roughly on time.

When outsiders admire China’s ability to build, what do they miss? What are the costs of that breakneck progress?

Respect, and individual desire not to be displaced or bothered. The pulverization of different ecologies. Focusing so much on building infrastructure rather than allowing a greater degree of creative flourishing. Concentrating so many resources in the hands of the state. And fortifying the country to prepare for a soft cold war against the United States, rather than allowing people to engage more with you folks overseas.

Is it hard to know which side to prefer?

I like pluralism. It is really great to have too many homes, more parks, more subway stations and cleaner air. These are all real achievements of the engineering state. But I also like debate and pluralism and discussion. And I think that it is possible for the US to build [while maintaining those qualities]. Denmark, France and Japan have all built excellent infrastructure without running roughshod [over society]. The problems are fixable.

Inside Onkalo, Finland’s nuclear waste repository on the island of Olkiluoto Kallerna / Wikimedia Commons (CC BY-SA 4.0)

New burial chambers will hold nuclear waste for 100,000 years. The world’s renewed interest in nuclear power resurfaces a problem from the last nuclear boom: Where to put the waste? In the past, sites for burying nuclear waste have sometimes failed due to water ingress or subsidence, but a series of new facilities — including Onkalo, in Finland, which last month secured an operating licence — hope to lock waste away for good. Nuclear waste loses potency slowly over decades, centuries or even millennia depending on the material in question. Some of the waste that needs to be secured will remain hazardous for as long as 100,000 years. Today, most nuclear waste is stored above-ground at reactor sites, waiting for the construction of safe and secure below-ground disposal locations. The challenge is that engineering such a site is a tall order, given that geological movement, cracked rock and water ingress can all wreak havoc. At Onkalo, on the Finnish island of Olkiluoto, the solution is a sort of Russian doll of safety measures, New Scientist reports. Deep tunnels have been drilled into dense and impermeable metamorphic rock called gneiss. Nuclear waste is loaded into a cast iron container, which is then sealed inside a two-inch thick copper canister. That is lowered into a hole bored into the tunnel floor and surrounded by a type of clay called bentonite, which, when damp, swells to create a nearly watertight barrier. Finland’s nuclear waste management company, Posiva, has been building Onkalo since 2004, and believes its multilayered approach should keep the waste sealed away for 100,000 years — a period during which, as New Scientist puts it, ice sheets may come and go, and civilizations rise and fall. Sweden is already building a similar repository, and Canada and the UK plan to follow suit.

New techniques could help us clean up “forever chemicals.” Polyfluoroalkyl substances, or PFAS, are referred to as “forever chemicals” because the carbon-fluorine bonds within them are stubbornly resistant to breaking down. This feature makes them highly attractive for some applications, like Teflon coatings and specialized firefighting foams. But PFAS have also been linked to higher cholesterol, weaker vaccine responses and higher cancer risk, and their environmental ubiquity means they show up in drinking water, food, soil and the blood of almost all people. PFAS readily dissolve in water, which is why they are so widely spread. So the first step in getting rid of them is filtering them out of what we drink, cook and bathe in. The good news is that this step is pretty much solved from a technological standpoint. The challenge is what to do with the PFAS afterward. Historically, concentrated PFAS waste was incinerated, but some incinerators released fragments of PFAS back into the environment. (The hottest hazardous waste incinerators can destroy PFAS almost completely, but they are scarce.) Science reports that newer, portable techniques to permanently get rid of PFAS offer an alternative. Some expose concentrated PFAS waste to intense temperatures and pressures to rip apart the carbon-fluorine bonds. Others use sound waves, ultraviolet light or electricity to do the same thing. Comparing the effectiveness of competing approaches is difficult, as the companies and researchers developing them measure success differently. And these new approaches will have to demonstrate that they destroy PFAS without creating other harmful compounds. Ridding the world entirely of PFAS is likely a pipedream given the cost, but the new techniques could make a difference by focusing on cleaning up PFAS hotspots.

Tiny silicon probes are revealing the inner workings of the brain. How can we better observe subtle brain activity in animals and humans, preferably when they’re awake? For decades, it’s been impossible, but Nature reports that a system known as Neuropixels is changing that. The latest version is built around hair-thin strands of silicon dotted with 5,120 electrodes, each of which can measure the electrical activity of a single brain cell. Data can be recorded from more than a quarter of the electrodes simultaneously, allowing researchers to choose which signals to measure along the 10-mm length of the device. Insertion requires that a hole be drilled into the skull, which helps explain why early volunteers for the device are people already undergoing brain surgery. (It has been used in animals since 2017.) The system is already enabling new insights into how brains work — helping, for instance, to map cognitive and motor function in mice, and providing what one neurosurgeon has called “the most high-resolution data that we’ve seen” to understand speech and language. The developers are currently testing versions that use optogenetics — a technique that uses genetic engineering to control brain cells with light — that not only measures activity but can also switch neurons on and off in live animals. A planned next-generation device is designed to record signals from tens of thousands of neurons at once. 

AI has enabled Anthropic to double the pace at which its own AI models are advancing. The news comes from a November 5th report by the independent evaluator METR, though there remains uncertainty around the results and what they mean for the future. (METR, short for Model Evaluation and Threat Research, is a nonprofit that assesses risks and capabilities of advanced AI systems.) The finding is the result of a monthslong study by METR staff who had access to Anthropic’s internal data. Its central conclusion is that Anthropic is now improving its models on “well-specified tasks” about twice as fast as it did between March 2024 and October 2025. METR also wrote that perhaps as much as 80 percent of that acceleration is “likely attributable to Anthropic’s extensive use of AI to do R&D,” and that it found no obvious bottlenecks that would prevent further such progress in the coming 12 months. One of METR’s key concerns is that advancement like this could lead to recursive self-improvement, or RSI, the process through which AI develops better versions of itself, potentially leading to an explosive growth in AI capability. METR wrote that “AI is not yet able to make highly significant and fully autonomous contributions to the most important parts of the model training process,” but added that staff at Anthropic “described significant ‘signs of life’ implying such contributions could occur soon.” Days before the report was published, Anthropic updated its IPO filing to warn that it “may imminently meet” its definition of “dramatically accelerated progress attributable to AI.” Yet Anthropic redacted most of the key evidence behind the METR report, citing fears that sharing too many details would help other companies accelerate their AI progress. That means METR cannot fully justify many of its claims, admitting that, “as a consequence, the public has to rely on METR’s judgment much more than would be ideal, given the weight of the questions answered here.” Some in the AI safety community are, perhaps predictably, terrified by the report. But the lack of full transparency from Anthropic makes it almost impossible to form a complete picture of how close to RSI the company is.

Women travel thousands of miles for a reproductive moonshot. Scientists say they’ve been misled, from STAT
9,300 words, or about 37 minutes

A reproductive technique known as mitochondrial donation can enable mothers with mutations in the DNA of their mitochondria — which provide cells with energy — to have children who are genetically related to them but at much lower risk of developing mitochondrial disease. But the technique, also referred to as three-parent IVF because the child inherits DNA from three people, has also been controversially marketed as a solution to age-related infertility. This investigation by STAT dives into the murky world of that application. Mitochondrial replacement therapy, or MRT, is effectively illegal in the US and is permitted in the UK only to prevent disease from being passed on. Against that backdrop, medical tourism facilitated by organizations such as the International Fertility Group of North America enables women to receive the treatment in countries like Albania, where it is unregulated. Such treatments cost tens of thousands of dollars, and hundreds of women have sought them out. But, as STAT explains, the evidence for MRT’s benefits in age-related infertility is slim. The only relevant peer-reviewed study did not include patients over the age of 39, and other research used to support the approach claims success rates that experts deem too good to be true. Meanwhile, unpublished work suggests that age-related infertility is tied to the nuclear, not mitochondrial, DNA, which would make MRT ineffective. During the course of STAT’s investigation, the International Fertility Group paused taking on new MRT patients, and just ahead of the story’s publication, took all mention of MRT off its website.

When humans disappear, biodiversity takes a surprising turn, from Nature
2,400 words, or about 10 minutes

Conventional wisdom suggests that where there are humans, plant and animal life suffer. And while it’s certainly true that biodiversity falls as development spreads, it can also decrease when human populations decline, especially in rural areas. In Japan, forests that are no longer used as a source of firewood and compost by humans become clogged, squeezing out plants and insects that thrived in more open woodland. In sites across Albania, Bulgaria, Croatia and Greece, human abandonment of farmlands has led to a decrease in bird populations. Across numerous sites, it seems that modest disturbance of nature by humans — through things such as livestock grazing, seasonal flooding, clearing of vegetation or burning of small areas of land — enhances diversity of wildlife. To be clear, this isn’t true of all species, or in all locations. Apex predators like wolves, for example, thrive without humans around. Still, as a result of urbanization, just 19 percent of the world’s population now live in rural areas, down from 40 percent in 1950. In some parts of the world, this new perspective suggests that conservation efforts may require active human involvement in rural settings through activities like small-scale farming, to safeguard biodiversity.

The agentic web is here, from MIT Technology Review
4,000 words, or about 16 minutes

Take your pick from Meta’s Muse, OpenAI’s Dots, Anthropic’s Claude Duet or upstarts like the much-hyped Instinct — AI agents are here, and people are using them to book flights, shop for food and snag dinner and other reservations. This piece looks at how agents have already changed the web: AI search results have reshaped web traffic, AI shopping agents are starting to affect how purchases get made and advertisers and retailers are adjusting to a world in which they may have to sell products to bots rather than humans. What happens next is less clear. Some people imagine that agents will deal directly with sellers so that middlemen like DoorDash or Airbnb will be cut out. This could mean savings for consumers and more control for small businesses. Others think that the self-serving practices of existing tech companies could become even more entrenched  — with Meta, say, designing Muse to steer users toward its own services and businesses that have paid to be featured. As for the web that humans interact with? One view is that it will change much as it did when the smartphone arrived, becoming more multifaceted. As the process unfolds we can expect some businesses built on the human web to struggle or die as norms shift and for new ones to emerge built only for agents. In the process, the human web may become less a platform for commerce and more a playground for fun. 

More stories worth knowing about: 

China may put guard rails in place around advanced AI. A new gene therapy appears to cure epileptic seizures. AI agents can run an automated lab that sequences DNA. China’s clean-tech exports in 2026 will have been responsible for preventing as much C0₂ from being released as Brazil emits in a year. Supermarket and blood bank freezers have been hacked in Europe while in the US, the Pentagon is investigating whether some of its freezers may have been targeted, too. Meta is investing in low-carbon concrete to offset the environmental impact of data centers. Google’s first experimental data center satellite is in orbit and operational. AI detectors are changing the way we write. And, how a brain injury made one Tennessee man a musical genius.