Hello from Zurich,

I hope you had an interesting and productive week.

A sign on a well placed park bench in Vancouver reads; “Ask Why, Figure Out How”

May this simple recipe serve you plenty of food for thought

1 Getting Visual

2 If You Read One Thing Today - Make Sure it is This

3 Consequential Thinking about Consequential Matters

4 Big Ideas

5 Big thinking

6 Vision

1 Getting Visual

Big Picture: Global GDP Energy Input - Significant changes: The oil intensity of global GDP has plummeted since its 1970’s peak and is only half the level it was at the time of the Gulf War in 1990. Even the natural gas intensity of GDP has declined since 1980 despite natural gas consumption tripling since then. The primary drivers of these declines are shown on the right: substantial improvements in energy efficiency, a shift from coal to more efficient combined cycle gas turbines and renewable energy used for industrial power needs, building HVAC and transport via EVs and biofuels.

Key Trend - US constraints & The shifting realities of conflict: “While it’s impossible to know exactly what munitions have been expended so far, some foreign policy institutes and research centers are making some educated guesses based on satellite imagery. Note in the first table that the US may have expended in just 6 days more of certain munitions than will be produced in 2026. On the right: an estimate of the number of days that munitions could be sustained at same pace as the first 96 hours. Obviously the pace of munitions deployment has slowed since the first 96 hours, but the table is illustrative with regards to the amount of munitions expended and how the US could only sustain it for roughly a month.The byproduct of rapid munitions use is the need to replace the critical minerals embedded in them. The next table estimated the mass of critical minerals consumed within the first 96 hours of the war. Such amounts are small shares of annual US consumption; the greater risk is that for some of them, China is the largest single critical mineral counterparty.” - JPM

Building the infrastructure of the Future & Strategic Relationships: Reports of the end of China’s One Belt One Road lending were very premature - 2025 was a record year for new commitments.

Spotlight: Bullish CATL - A winner of the global shift to the Electrostate model:

“CATL, the Chinese battery group that commands roughly 40% of the global EV battery market, posted Q1 revenue of Rmb129.1bn ($18.9bn), up 52%, with net profit rising 48.5% to Rmb20.7bn. The revenue surprise, 40% above the median forecast of 12 brokerages, suggests the sell-side was still modelling a battery supplier. The results read more like those of an energy conglomerate in formation. Two fronts drove the beat. Domestic EV battery share crossed 50% for the first time in 5 years. Energy storage, where margins run about 3 percentage points above those on EV cells, rose to roughly a quarter of Q1 cell shipments; April production schedules show storage climbing to 41.3% of cell output, up from less than 20% a year earlier. The Hungary plant, a €7.3bn investmentserving Mercedes-Benz, BMW, Stellantis, and Volkswagen, began mass production.About a third of revenue now comes from outside China...” - Hello China Tech Blog

Spotlight - Transportation Trends: Creative Destruction is Real: “If the European car industry thinks it has a China problem, spare a thought for Japan! Nissan, once the sixth-largest carmaker in the world by sales, is entering the second year of a brutal restructuring, with seven factory closures planned by 2028. A 25% tariff on cars imported into America has bitten into the industry’s profits. Yet it is the blistering rise of Chinese competitors that has hit hardest. In 2019 Japanese carmakers accounted for 31% of sales globally; by last year their share had fallen to 26%. The shock has been greatest in Asia. In China itself, sales of Japanese cars have slumped by a third since 2019. In South-East Asia, once a stronghold, their share of the market was 57% in 2025, down from 68% just two years earlier. Japanese carmakers once seemed unstoppable. How did it go so wrong for them? The heart of the problem is that, even more so than their Western counterparts, Japanese carmakers have struggled with electrification. Many have been sceptical of the staying power of electric vehicles (EVs), which account for a vanishingly small share of their sales (see chart). Conventional petrol vehicles make up more than half of sales for all Japanese carmakers; at beleaguered Nissan it is 80%. Rather than plug-in cars, most have opted instead to emphasise conventional hybrids, which rely on the engine and regenerative braking to power the battery, as the assembly of these fits more easily into a production line built for internal-combustion engines. Japan’s carmakers have expressed interest in alternative technologies such as hydrogen-powered cars for much the same reason.” - The Economist

AI - The CFO Perspective: “We are in the midst of an AI revolution, and the CFO Survey below shows the many channels through which business managers expect AI to have an impact on firm outcomes. CFOs expect the biggest impacts to be on decision speed, output per worker and time spent on high value–added tasks, with employment essentially unchanged.” - Apollo Research

AI - A force multiplier or destructor of employment - It’s early but so far no sign of the doom case scenario: “The first chart below compares the unemployment rate for the entire US population with the unemployment rate for people ages 20 to 24. It does not show any sign that unemployment among younger workers is structurally higher because of AI. Similarly, the second chart shows the unemployment rate for US college graduates ages 22 to 27. The unemployment rate has increased for men, but it has recently converged toward the unemployment rate for women. For women, since ChatGPT was released, the unemployment rate has been moving lower, and more recently it has increased slightly again. The bottom line is that there is no sign that AI is increasing unemployment among younger workers, and there is also no sign that young people or recent college graduates are having a harder time finding jobs at the moment than other demographics.

US Macro Picture - Big Boy Rolls at the Treasury for as far as the Eye can See…

2 If You Read One Thing Today - Make Sure it is This

ChinaTalk dropped a couple of interesting explorations on matters that are very relevant to consider as you seek to understand the paths ahead - the first focuses on Ukraine’s scaling of drone production and their application in warfare and the second dives into Mythos’ applications in cyber and it’s potential impact on national power. These two technologies have the potential to be the “Howitzer cannons” to the established “fortifications” and related strategies of of our time…Go explore it all in more detail here:

ChinaTalk

How Ukraine Scaled to Millions of Drones

Ukrainian drone manufacturing. How has the country been able to scale from thousands to millions of drones over the past four years? What dependencies does its industrial base still have on China? And what lessons does its rapid scaling offer for the US…

6 months ago · 51 likes · 4 comments · Jordan Schneider, Chris Miller, and Phoebe Chow

ChinaTalk

Mythos and National Power

Anthropic’s new model found decades-old vulnerabilities in foundational open-source code that millions of automated tests and countless human experts had missed, presaging a potentially revolutionary moment in cyber…

6 months ago · 82 likes · 5 comments · Jordan Schneider

Some Takeaways

“Ukrainian drone manufacturing. How has the country been able to scale from thousands to millions of drones over the past four years? What dependencies does its industrial base still have on China? And what lessons does its rapid scaling offer for the US?”

“In February of 2022, we had about 3,000 drones total being produced in Ukraine — FPV (First-Person View), UGV (Unmanned Ground Vehicles), sea drone, anything of the sort. Ninety-nine percent of them were imported as entire systems from China.

In 2026, we basically have 99% being assembled in Ukraine. Now, just the FPV industry alone is cited to be able to produce up to 5 million FPV drones per year. That doesn’t include our massive industry of heavy bomber drones, ISR (Intelligence, Surveillance, and Reconnaissance), loitering munitions, or UGVs, which is now a booming industry in Ukraine as well.

But the most impressive thing isn’t necessarily just those numbers — we went from about 3,000 systems being made in February ’22 to 4 million FPVs alone. It’s the actual localization of that final assembly and the way that Ukraine has been able to completely transform its drone manufacturing industry.

Now we’re at a point where 99% of the systems are final assembly in Ukraine with a lot of components being imported, but basically no final systems being imported from China anymore, which is a massive accomplishment.”

“By 2023, when we were thinking about potentially another wave of counteroffensive, Ukrainian soldiers and volunteer networks started buying up more and more drones initially to perform ISR functions. Then pretty quickly — if you’re a soldier fighting in an existential battle, you’re going to do absolutely everything you can with the tools at your disposal. People realized that they could strap explosives onto these things and just fly them directly into the enemy, which was huge.

What was huge here for Ukraine was the asymmetry of using drones. First, because we were strapped for cash. The cost asymmetry of being able to put a payload and ammunition onto a drone that would only cost you a few thousand dollars was huge for a country that’s at economic disadvantage and fighting against Russia, which has one of the largest military industrial complexes and military budgets in the world.

Second, the asymmetry of being able to protect our soldiers and pilot these drones remotely was also huge because you’re never really going to be able to go person for person with the Russian army. They’re always going to have more people. In a war of attrition, which we pretty quickly realized it was going to be, we weren’t going to be able to hold the line with as many infantry and as many soldiers as Russia would. Being able to send remotely controlled tools to perform certain functions instead of putting human life at risk was another huge benefit for the Ukrainian side.

Pretty quickly we went into overdrive to produce these drones just by the sheer necessity of not having as much money to buy different systems and not wanting to put our people’s lives at risk. From 2023 to now, it was just a huge industry boom. We got to where we are today because we realized that was one of the major things keeping us in the fight — our ability to leverage unmanned systems as opposed to putting our capital and our people’s lives at risk.”

“Although we had the history, and Ukraine has historically been full of engineering talent with a lot of that knowledge, the manufacturing was not maintained to the extent that it should have been. Most of our legacy exquisite systems were completely out of date, in need of repairs, and basically unusable. One of the huge reasons that we had to start using USVs and sea drones was because our fleet was in complete shambles and complete disrepair. Even though we had some ships, it just wasn’t realistic to use them in a wartime scenario at all.

A lot of the tech talent in Ukraine wasn’t actually working in the defense industrial base at the time. Ukraine was famous for its IT industry, software, and computer science. When the full-scale invasion began, harnessing civilian talent was one of the big things that kept us in this fight. Many people who were previously working in the software industry, in consumer goods and technologies, completely shifted.

It was similar to what happened in the US during World War II, where you tapped into this massive civilian talent and massive civilian production lines and directed them to contribute to the war effort. The tapping in of the civilian industries, which was supported in large part by our government and its state policies to encourage more companies to direct their efforts into defense, was what kept us afloat.”

“Can you tell us about the typical entrepreneur who started up a new drone company or runs a production line? Who are these people and what makes them good at their jobs?

They come from completely different backgrounds, which is super interesting. You have some people who, in their past lives, used to be top software engineers at B2B SaaS companies. You have one of the biggest defense tech VCs right now supporting the entire industry, who used to be the chief marketing officer at a workflow automation company. Some people weren’t in tech at all and became CEOs, stepping into it from working at video game companies.

The video game overlap is actually quite real — that pipeline exists.

I actually used to play a lot of video games and learned drone operating from that. Many of them were working across the industry at places like Uber Ukraine or other rideshare companies. There are a few examples of that.

It became unimaginable for most people in Ukraine after February 24th to work on anything except this. It’s something that you really can’t replicate unless your country is at war — and not only at war, but in an existential one. It’s extremely difficult for any other country to imagine.”

“The process from parts to combat-ready drone involves assembling components in the shop, shipping to the unit, where they’re disassembled and reassembled in their own production lines before deployment. This is something many Western countries don’t comprehend — it’s almost impossible to ship a finished system that flies straight out of the box. R&D shops and assembly lines operate across locations closer to the front lines, run by the military doing their own assembly work.”

“You have this line in one of your reports from late last year that “Ukrainian startups can assemble and ruggedize, but they cannot easily reproduce decades of specialized chemical material or electronic expertise.” Before we get to the second half of that statement, let’s explore the assembly and ruggedization aspect. What has that unlocked for Ukraine? Why was it important to have that domestic capacity developed in the first place?

Cat Buchatskiy: The ruggedization was crucial because Ukraine is fighting a war of attrition. Modularity is incredibly important, which is why in-house assembly matters so much.

As I mentioned earlier, when systems get built in the factory and sent to military R&D labs, they essentially disassemble and reassemble them. The reason is that manufacturers can’t predict which features the frontline will need by the time products ship out.”

“When people talk about short innovation cycles in Ukraine, this is mostly what they mean — the ability to have continuous R&D and for soldiers to get hands-on and adapt modular systems to their needs. We’ve nailed this down, and it’s been incredibly important.”

China’s Calculated Neutrality

Jordan Schneider: Nice drone industry you have there — shame if some export controls were to happen to it. I want to read in full this opener from a Financial Times piece from about a month ago, which featured one of your interviewees —

“On his numerous visits to the factories of southern China, Oleksandr Yakovenko finds that his hosts increasingly plan his arrivals and departures down to minutes and seconds. They sometimes ask him to wait nearby for a while or usher him through side doors, down service corridors, or into empty conference rooms. It took the founder of TAF Industries, now one of Ukraine’s biggest drone producers, a while to realize why his arrival at the head office of a camera developer or battery maker required such opaque rituals of schedule juggling and extreme punctuality. It was because the Russians had just been there, or they were on their way, or both.

‘Our suppliers make an effort to manage the Ukrainian and Russian customers. They try to make it so we don’t have to be in the factory at the same time,’ he told theFinancial Times. ‘They invite us for one time slot, but they invite the Russians for a different time. As soon as the car with the Russians drives away, the car with the Ukrainians goes in,’ he adds.”

What an unbelievable situation we’re in. It’s truly surreal. There have been other times in history where arms manufacturers sold to both sides — actually, the more I think about it, it’s not that uncommon. But the fact that we’re having this iterative technological race, as opposed to just selling some AKs to this side and some AKs to that side with a shrug, is really weird.

Cat, can you start by telling the story from the Russian side as well? How do both sides of this war have significant drone dependencies on what comes out of factories in China?

Cat Buchatskiy: It’s definitely a very bizarre scenario, especially for our manufacturers dealing with this. Both sides have a dependency because most of the critical components for the drone industry are based in China.

The whole world really has this dependency.

For both sides to produce the unmanned systems we need at the scale we’re going through them, it’s impossible to do without China.”

“My conspiracy question here for you — do they also not want the Ukrainians to lose? Selling drone parts to Ukraine is not a central pillar of the Chinese economy. There has to be some larger strategic calculus going on to allow this number of parts to continue to flow to the Ukrainian drone base.

Cat Buchatskiy: I don’t think that China wants Ukraine to lose. The reason being that I don’t think China and Russia are real friends, and I don’t think China minds depleting Russia’s arsenal. China doesn’t really need Russia for the most part. While there’s a lot more economic interaction with Russia now, in terms of defense and the role that Russia plays in the world, China sees it as a defeated, has-been power.

China understands that, frankly, the US also doesn’t see Russia as its greatest threat. Read the recent national security strategy — it’s barely in there. The US is all focused on China.

I don’t think Xi really cares if Ukraine is able to continue to attrit the Russian defense industry. For them, playing both sides is a win-win scenario because they keep their biggest ally dependent on them. The Chinese defense industry is going to be stronger than the Russian defense industry, and Russia is going to continue to need to buy parts from China.

Frankly, they’re exacerbating the divide. If you’re thinking about great power politics, China’s only getting stronger, Russia’s only getting weaker, and it’s not going to be a tripolar world between Russia, the US, and China. They’re going to want to make it a bipolar world, and Russia is going to be dragged into that orbit as long as Ukraine continues to weaken its global position and sanctions continue to be held.”

Mythos and National Power

Has the A-bomb of cyber just been discovered?

Anthropic’s new model found decades-old vulnerabilities in foundational open-source code that millions of automated tests and countless human experts had missed, presaging a potentially revolutionary moment in cyber.”

“So how big a deal is Claude Mythos?

Ben Buchanan: This is a big one. I’ve been thinking about cybersecurity and AI for more than a decade. I think a lot of us who were thinking about AI and cyber back then imagined that a day like this might come where you could see automated vulnerability discovery. It does feel like something that had long been imagined is actually now finally here, and it’s up to all of us to figure out what that means.

Jordan Schneider: So what can the model do?

Ben Buchanan: What this system does at its core is it takes a general-purpose capability — it is not a cyber-specific model — and applies it to the business of vulnerability discovery and exploit development. As Michael can attest very well, these are fundamental tasks in cybersecurity: finding a weakness in a piece of computer code and then figuring out how to exploit that weakness to do something as an attacker that you’re not allowed to do.

The evidence is very clear that Claude Mythos is by far the best automated system in the world ever to do this, and is better than even some of the best expert humans — or close to some of the absolute top-tier expert humans — at this task of vulnerability discovery and exploit development. The proof is in the pudding. It found vulnerabilities in code that all of our operating systems and all of our browsers are running. Those vulnerabilities in some cases had lurked there for multiple decades. In some instances, we thought that code was secure.

Millions of automated tests had been run on it, and yet Mythos found ways to exploit it. There is a real raw capability here that is vital.”

“The core credo of the open-source software movement, which I should be clear I totally support, is: with enough eyeballs, all bugs are shallow. Basically, if enough smart people are looking, they will find everything that is to be found.

I think the answer for this moment is we need to have machines look too — or at least, a machine of this capability level can find things that a lot of good humans looking for a long time didn’t find.”

“What I think remains the same is that success in cyberspace generally has come down to a race — a race from when the offense or the exploiters know about a problem and how fast they can get at it, compared to how fast the defenders can identify, fix, and then disseminate the fix as broadly as possible. So part of the answer is: if you’ve got the offense, you’re the only one, and defense doesn’t know, it’s pretty open season.”

“The conclusion we came to in 2019 and 2020 was that at least theoretically, at each step of that offensive operation process, AI could help. Now I think with something like Mythos, that conclusion is just far more robust. We saw the glimmers of it in 2019 and 2020, but Mythos is really doing it — not just in vulnerability discovery, though that’s a key part of it, but throughout the process. There’s something in the system card for Mythos where it carried out a simulated network exploitation that would have taken a human 10 hours. So there really is evidence now that what cyber operators call the kill chain can be transformed by AI capabilities.”

“The whole process from discovery of the bug to development of the patch to deployment of the patch — that’s going to have to go so much faster in a post-Mythos era, because stuff like this will proliferate and folks will be looking for these things and maybe they can reverse-engineer patches. The IT industry and the backbone of critical infrastructure is going to have to level up in speed because of Mythos.

That probably is a harbinger of what’s going to come in AI — that where we have the things for societal resilience, we’re going to have to get more resilient faster for individual cycles because AI is going to accelerate the offensive side of the ball.”

“One lesson we should take away from Mythos is not “wow, this means AI is really good at cyber” — it’s that AI is really good.

This is a general-purpose system that happens to be good at cyber.

If you read the Anthropic system card for Mythos, it’s also really good at bio. I imagine the next version is going to be even better. There’s been a lot of debate for the last five years about how good AI systems are going to be. Obviously folks like me have argued for a very long time that they’re going to be very good, faster than people think.

I’m biased here, but this feels like a pretty big piece of evidence that should update us towards taking AI risks seriously — in cyber, yes, but also in things like bio, because those are not going to be far behind.”

3 Consequential Thinking about Consequential Matters

Dwarkesh has Jensen Huang on the pod for a conversation on TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat and much more - Plenty of Consequential Thinking about Consequential Matters…Go listen here:

Dwarkesh Podcast

Jensen Huang – TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat

Listen now

6 months ago · 320 likes · 56 comments · Dwarkesh Patel

Some Takeaways

“We’ve seen the valuations of a bunch of software companies crash because people are expecting AI to commoditize software. There’s a potentially naive way of thinking about things, which is: look, Nvidia sends a GDS2 file to TSMC. TSMC builds the logic dies, it builds the switches, then it packages them with the HBM that SK Hynix, Micron, andSamsung make. Then it sends it to an ODM in Taiwan where they assemble the racks. Nvidia is fundamentally making software that other people are manufacturing, and if software gets commoditized, does Nvidia get commoditized?

Jensen Huang:

In the end, something has to transform electrons to tokens.

The transformation of electrons to tokens and making those tokens more valuable over time is hard to completely commoditize. The transformation from electrons to tokens is such an incredible journey.

Making that token is like making one molecule more valuable than another molecule, making one token more valuable than another. The amount of artistry, engineering, science, and invention that goes into making that token valuable, obviously we’re watching it happen in real time.

The transformation, the manufacturing, all of the science that goes in there is far from deeply understood and the journey is far from over. I doubt that it will happen.

We’re going to make it more efficient, of course.

The way that you framed the question is my mental model of our company.

The input is electrons, the output is tokens.

In the middle is Nvidia.

Our job is to do as much as necessary and as little as possible to enable that transformation to be done at incredible capabilities. What I mean by “as little as possible,” whatever I don’t need to do, I partner with somebody and make it part of my ecosystem.

If you look at Nvidia today, we probably have the largest ecosystem of partners, both in the supply chain upstream and downstream, all of the computer companies, application developers, and model makers.

AI is a five-layer cake, if you will. We have ecosystems across the entire five layers. We try to do as little as possible, but the part that we have to do, as it turns out, is insanely hard. I don’t think that gets commoditized.

In fact, I also don’t think the enterprise software companies, the tools makers… Most software companies today are tool makers. Some of them are not. Some of them are workflow codification systems. But for a lot of companies, they’re tool makers. For example, Excel is a tool, PowerPoint is a tool, Cadence makes tools, Synopsys makes tools. I actually see the opposite of what people see. I think the number of agents is going to grow exponentially, and the number of tool users is going to grow exponentially. It’s very likely that the number of instances of all these tools is going to skyrocket.

It’s very likely that the number of instances of Synopsys Design Compiler is going to skyrocket, along with the number of agents using the floor planners, our layout tools, and our design rule checkers.

Today we’re limited by the number of engineers. Tomorrow, those engineers are going to be supported by a bunch of agents. We’re going to be exploring the design space like you’ve never seen before, and we’re going to use the tools that we use today.

I think tool use is going to cause the software companies to skyrocket. The reason why it hasn’t happened yet is because the agents aren’t good enough at using their tools yet.

Either these companies are going to build the agents themselves, or agents are going to get good enough to be able to use those tools. I think it’s going to be a combination of both.”

“This is one of the concerns that I have about the doomers describing the end of work and killing of jobs. If we discourage people from being software engineers, we’re going to run out of software engineers.

The same prediction happened ten years ago. Some of the doomers were telling people, “Whatever you do, don’t be a radiologist.” You might hear some of those videos still on the web saying radiology is going to be the first career to go and the world is not going to need any more radiologists. Guess what we’re short of? Radiologists.”

“We’re developing all kinds of new techniques so that we drive efficiency in addition to increasing capacity. None of those things worry me. It’s the stuff that’s downstream from us. Energy policies that prevent energy from… You can’t create an industry without energy.

You can’t create a whole new manufacturing industry without energy.

We want to reindustrialize the United States. We want to bring back chip manufacturing, computer manufacturing, and packaging. We want to build new things like EVs and robots. We want to build AI factories. You can’t build any of these things without energy, and those things take a long time. More chip capacity, that’s a 2-3 year problem. More CoWoS capacity, 2-3 year problem.”

“Should we be selling AI chips to China?

Dwarkesh Patel

Okay. I want to ask about China. I actually don’t know what I think about whether it’s good to sell chips to China or not, but I like to play devil’s advocate against my guests. So when Dario was on, who supports export controls, I asked him, why can’t America and China both have a country of geniuses in the datacenter? But since you’re on the opposite side, I’ll ask you in the opposite way.

One way to think about it is, Anthropic actually announced a couple days ago Mythos Preview. This model Mythos, they’re not even releasing publicly because they say it has such cyber-offensive capabilities that we don’t think the world is ready until we make sure these zero-days are patched up. But they say it found thousands of high-severity vulnerabilities across every major operating system, every browser. It found one in OpenBSD, which is this operating system that’s been specifically designed to not have zero days. It found one that’s existed for 27 years.

So if Chinese companies and Chinese labs and the Chinese government had access to the AI chips to train a model like Claude Mythos with these cyber-offensive capabilities and run millions of instances of it with more compute, the question is, is that a threat to American companies, to American national security?

Jensen Huang

First of all, Mythos was trained on fairly mundane capacity, and a fairly mundane amount of it. By an extraordinary company. The amount of capacity and the type of compute it was trained on is abundantly available in China. So you just have to first realize that chips exist in China.

They manufacture 60% of the world’s mainstream chips, maybe more. It’s a very large industry for them. They have some of the world’s greatest computer scientists. As you know, most of the AI researchers in all of these AI labs are Chinese. They have 50% of the world’s AI researchers.

So the question is, considering all the assets they already have—they have an abundance of energy, they have plenty of chips, they’ve got most of the AI researchers—if you’re worried about them, what is the best way to create a safe world?

Victimizing them, turning them into an enemy, likely isn’t the best answer. They are an adversary. We want the United States to win. But I think having a dialogue and having research dialogue is probably the safest thing to do.

This is an area that is glaringly missing because of our current attitude about China as an adversary. It is essential that our AI researchers and their AI researchers are actually talking. It is essential that we try to both agree on what not to use the AI for.

With respect to finding bugs in software, of course, that’s what AI is supposed to do. Is it going to find bugs in a lot of software? Of course. There are lots and lots of bugs. There are lots of bugs in the AI software. That’s what AI is supposed to do, and I’m delighted that AI has reached a level where it could help us be so much more productive.

One of the things that is underemphasized is the richness of the ecosystem around cybersecurity, AI cybersecurity and AI security and AI privacy and AI safety.

There’s a whole ecosystem of AI startups that are trying to create this future for us, where you have one AI agent that’s incredible, surrounded by thousands of AI agents, keeping it safe, keeping it secure. That future surely is going to happen.

The idea that you’re going to have an AI agent running around with nobody watching after it is kind of insane. We know very well that this ecosystem needs to thrive. It turns out this ecosystem needs open source.

This ecosystem needs open models. They need open stacks so that all of these AI researchers and all these great computer scientists can go build AI systems that are as formidable and can keep AI safe. So one of the things that we need to make sure that we do is we keep the open source ecosystem vibrant. That can’t be ignored.

A lot of that is coming out of China. We ought to not suffocate that.

With respect to China, of course we want the United States to have as much computing as possible.

We’re limited by energy, but we’ve got a lot of people working on that. We’ve got to not make energy a bottleneck for our country. But what we also want is to make sure that all the AI developers in the world are developing on the American tech stack, and making the contributions, the advancements of AI—especially when it’s open source—available to the American ecosystem.

It would be extremely foolish to create two ecosystems: the open source ecosystem, and it only runs on a foreign tech stack, and a closed ecosystem that runs on the American tech stack. I think that would be a horrible outcome for the United States.”

“The amount of energy they have is incredible. Isn’t that right? AI is a parallel computing problem, isn’t it? Why can’t they just put 4x, 10x, as many chips together because energy’s free? They have so much energy. They have datacenters that are sitting completely empty, fully powered. You know they have ghost cities, they have ghost datacenters too. They have so much infrastructure capacity. If they wanted to, they just gang up more chips, even if they’re 7nm.

Their capacity of building chips is one of the largest in the world. The semiconductor industry knows that they monopolize mainstream chips. They have over-capacity, they have too much capacity. So the idea that China won’t be able to have AI chips is completely nonsense.

Now, of course, if you ask me, would the United States be further ahead if the entire world had no compute at all? But that’s just not an outcome. That’s not a scenario that’s true. They have plenty of compute already. The amount of threshold they need for the concern you’re worried about, they’ve already reached that threshold and beyond.

So I think you misunderstand that AI is a five-layer cake, and at the lowest layer is energy.

When you have an abundance of energy, it makes up for chips.

If you have an abundance of chips, it makes up for energy.

For example, the United States is scarce on energy, which is the reason why Nvidia has to keep advancing our architecture and do this extreme co-design so that with the few chips that we ship—with the few chips, because the amount of energy is so limited—our throughput per watt is off the charts.

But if your amount of watts is completely abundant, it’s free, what do you care about performance per watt for? You get plenty. You can use old chips to do. So 7nm chips are essentially Hopper. The ability for Hopper… I’ve got to tell you, today’s models are largely trained on Hopper, Hopper generation. So 7nm chips are plenty good.

The abundance of energy is their advantage.”

“The fact of the matter is, their AI development is going just fine. The best AI researchers in the world, because they’re limited in compute, they also come up with extremely smart algorithms. Remember, I just said that Moore’s law is advancing about 25% per year. However, through great computer science, we could still improve algorithm performance by 10x. What I’m saying is that great computer science is where the lever is.

There is no question, MoE is a great invention. There’s no question, all the incredible attention mechanisms reduce the amount of compute. We have got to acknowledge that most of the advances in AI came out of algorithm advances, not just the raw hardware. Now, if most advances came from algorithms and computer science and programming, tell me that their army of AI researchers is not their fundamental advantage. We see it. DeepSeek is not an inconsequential advance.

The day that DeepSeek comes out on Huawei first, that is a horrible outcome for our nation.”

4 Big Ideas

Jacob Stern takes a trip to China to explore their clinical trial abundance and to seek cancer care for his business partner - it’s a personal and granular journey into the future…Plenty of Big Ideas and important perspectives - read it here in full:

ChinaTalk

Clinical Trial Abundance, Made in China

A guest post by Jacob Stern…

6 months ago · 81 likes · 3 comments · Jacob Stern

Some Takeaways

“I work with Sid Sijbrandij, a technology entrepreneur who has taken a radically personalized and high-agency approach to fighting his osteosarcoma (bone cancer). Before meeting Sid, I was a product lead at 10x Genomics, a sequencing technology company developing novel tools for understanding biology. Sid was the first person I met who had used 10x tools to inform their care. I now run the enterprise of Sid’s care, pursuing a strategy of maximal diagnostics, making personalized therapeutics, and doing treatments in parallel rather than one at a time. Against the odds, Sid has had no evidence of disease for almost a year now. We are scaling this approach for others, both bystarting companies and through philanthropic efforts.

Elliot Herschberg wrote an excellent and approachable post on Sid going “Founder Mode” on his cancer on his blog, Century of Biology. We recently gave a talk at the OpenAI forum on Sid’s journey and our approach. More details can be found atsytse.com/cancer, and 25TB of data and Sid’s treatment timeline are available open source at osteosarc.com.

Last August, Sid Sijbrandij and I traveled to Beijing for an experimental scan to look at a biomarker that’s specifically upregulated in his cancer. At that time, the only place we could do this was in China, using a molecule developed by Yang Zhi (杨志)’s group at Beijing Cancer Hospital. So that’s where we went.

We were stunned. The whole experience — from international patient check-in, to preparation of the radiotracer, to injection, to imaging, to discussing the result with the physician, to leaving with a glossy printout of the whole-body scan — took two hours.

Even in Germany, where clinics are experienced in using developmental tracers, this process would take most of a day. Beijing broadly and the hospital specifically were surprisingly straightforward to navigate for foreigners such as us who speak no Chinese.

This experience inspired me to return to China in search of a deeper understanding of what is happening at the forefront of biotech and medicine. I often read and hear that it is becoming more difficult for American biotech to compete with what’s happening in China. I wanted to understand specifically what was going on, and what the implications were for a patient seeking the world’s most innovative care.”

“I came away impressed. Medical tourism is likely to invert, with patients flying to China to seek cutting-edge care. And I hope that we in America can learn from the sensible steps the Chinese ecosystem has taken and speed up our own innovation cycle. Patients deserve it.”

“…the US system emphasizes uniform standards and upfront rigor, while China’s IIT model pushes decision-making closer to the doctor and the patient, making it easier to start trials quickly and iterate as data comes in.”

“China’s State Council has recently adopted Decree 818 (国务院令第818号) to streamline IITs for cell and gene therapies. Prior to this regulation, IITs were popping up everywhere (particularly around regenerative cell therapies), leading to uneven data quality. With the goal of making data quality more systemically robust, 818 restricts the authority to run IITs to a pre-selected set of Tier 3 hospitals and requires Good Clinical Practice (GCP) certification for investigators. Interestingly, 818 opens the door to bring therapies to market very very quickly. Once ~10-15 patients have been treated with a therapy at a given hospital, that hospital can apply for the right to charge patients for access to that therapy. Essentially, the combination of therapy and institution is being approved. Data across institutions can also be leveraged for national approval down the line.

All of this makes sense! The system leans on the reputational sensitivity and naturally risk-averse incentive structure of academic medicine to regulate which medicines move forward to human trials. By putting trust in clinicians’ and hospitals’ judgment, the system is able to bring therapies to patients quickly.”

“Labor costs, medical costs, and infrastructure costs are lower compared to the US and Europe. There’s apparently a local discount, too — I heard from one company with operations in both the US and China that the Chinese operation’s quotes from local providers are half what gets quoted to American companies (this gives them an advantage in capital efficiency). But the most striking dynamic I observed was the speed. For companies that know how to navigate (read: have relationships, know whom to trust, and possess pre-built trust with those people), there’s a vibrant, redundant, end-to-end supply chain that can be tapped on demand with a high degree of responsiveness.

The ability to go from zero to patient data in 18 months is an advantage that will compound, as companies and the ecosystem writ large will be able to get to the real learning (testing drugs in patients) faster and iterate. Many of the companies I talked to were primarily (if not solely) funded by local capital markets and domestic government support. But the local Chinese pharmaceutical market is not enormous, with prices substantially lower than in Western markets and many patients paying for drugs out of pocket. I got the sense that the ecosystem sees preclinical development and clinical data generation as an important export market, with China serving as the innovation and proof of concept generator for medicines that will help patients around the world. Capital is already starting to flow to support this vision. At multiple stops, my visit was preceded or followed by VC firms and multinationals looking to feed and tap this innovation engine.

This is good! Humanity needs more medicines — and right now, the path of least resistance to generate more medicines seems to lead through China.”

“One of the smartest, most innovative scientists I spoke with on my trip noted that the process China currently follows is inspired by the one America once used for cell and gene therapy. As Dr. Ruxandra Teslo has very cogently laid out in her work on Clinical Trial Abundance, the early work on CAR-T’s in labs like Carl June’s at the University of Pennsylvania followed a playbook that seems similar to what’s happening in China now. Meanwhile, efforts to extend the efforts that led to the Baby KJ gene editing triumph seem to be running into stringent standards that are unrealistic for academic groups or narrowly scoped efforts to overcome. Dr. Teslo has put forward a number of specific policy proposals that would help America return to agility in early-stage clinical trials.

I suspect we could lean more on a marketplace of reputation to keep clinical research in check in the US.

The Clinical Trial Notification Pathway recently proposed by the FDA would be a good step in this direction. Some states, including New Hampshire, Montana, and elsewhere, are also moving in this direction; if not precluded federally, state-level innovation could serve as a laboratory of governance to test different versions of reform prior to wider-scale implementation.

It was interesting to observe that the expanded access/single-patient IND pathway is more suited to flexibly get an individual access to a potentially important treatment than anything I heard about in China. It was encouraging to seeDr. Marty Makary say recently on X that he’s signed every compassionate use request that’s crossed his desk. This is great! We should continue leaning in on single-patient INDs, with situation-appropriate standards that reflect the risk of inaction for a patient in dire straits.

How can America go faster?

The default is to leverage the CRO/CDMO infrastructure that’s available in China to develop Western IP, which is clearly happening. What about parallel infrastructure? My mind goes to companies like Plasmidsaurus, Adaptyv, and Aequita, which are building highly automated, fast, pay-by-credit-card offerings for specific high-volume assays, and earlier-stage analogs in manufacturing like Nature’s Toolbox, Harton, and Exthymic. I’d love to know what else is out there.

It is ironic to me that the ‘marketplace of reputation’ that seems to govern China’s IIT ecosystem is more market-oriented than the regulatory apparatus we use to govern early-stage trials in the US.

Every system has its strengths and drawbacks, China’s included. The parts I saw up close show how the Chinese ecosystem is leaning into its strengths — velocity of science and engineering, urgency, close-knit relationships within the ecosystem, compassion for patients. I’m hopeful that, as a country, we can reflect on and actively lean into our strengths as an ecosystem too.”

5 Big thinking

Brendan McCord of the Cosmos Institute explores “Why the Race to Stay Useful is a Trap” - it’s an interesting lens for assessing the state of play in the world today…Go explore it in full here:

Cosmos Institute

You Are Not a Function

In the autumn of 1809, Prussia was a country that no longer knew what it was for. Three years earlier, Napoleon had destroyed its army in an afternoon and walked into Berlin without resistance. The king fled. Half the territory was gone, the treasury empty. French soldiers were still…

6 months ago · 217 likes · 13 comments · Brendan McCord

Some Takeaways

“In the autumn of 1809, Prussia was a country that no longer knew what it was for. Three years earlier, Napoleon had destroyed its army in an afternoon and walked into Berlin without resistance. The king fled. Half the territory was gone, the treasury empty. French soldiers were still garrisoned in the capital.

As Prussia began rebuilding from the wreckage, most people assumed it needed more officers, administrators, and engineers. People who could do things. The task of designing the new system of education fell to a thirty-two-year-old diplomat named Wilhelm von Humboldt. He gave them something else entirely.

In a series of memoranda written over the next year, he laid out a vision for a new university in Berlin organized aroundBildung. The word has no English equivalent. “Education” is too narrow, “self-improvement” too thin. “Formation” gets closest but still misses its moral weight.

Humboldt’s Bildung means the free, harmonious development of a human being’s powers into a complete and consistent whole, through encounter with the world in its variety and resistance.

Mill, who took the idea from Humboldt, put it more simply: a human being is more like a tree than a steam engine.

Humboldt proposed a university where professors and students would be joined in the pursuit of knowledge, unconstrained by political demands.

In a defeated nation hungry for officers and administrators, he was arguing for formation before function.

The ideal of the modern research university, with its union of teaching and inquiry, its seminar culture, and its commitment to academic freedom, descends from what Humboldt designed in those desperate months.”

“Credentialism twisted the university into a vendor of certificates, and the formation of the student as a complete human being came to seem anachronistic.

The cathedrals remain, but not the faith.”

“The loudest responses to the crisis have come from outside the university. Alex Karp tells young people to skip college and learn a trade. Marc Andreessen argues the university is a credentialing middleman and should be disintermediated.

Both are right that the university is failing.

But if the answer to a broken formation system is to skip formation altogether, you have already conceded that education is justified only by utility.

Neither is asking the question Humboldt asked: What is a human being, that education should serve it?”

“Technical skill on a foundation of general cultivation is more resilient and more humane than technical skill resting on nothing.”

“A tree does not exist in order to produce lumber. You can make lumber from it, and good lumber is nothing to sneer at.

But if you look at a tree and see only lumber, you have missed what is standing in front of you.

Something is growing there under its own power, toward its own form, and the growing is not a means to some further end.

Humboldt’s claim about human beings is the same shape. A person is a self-developing being whose worth is not exhausted by function.”

“You have had the experience even if you never had a word for it. Real engagement with something that has its own demands—a hard problem, a serious book, a gifted teacher—changes who you are. You could not have planned the person you became.

Such formation is not the property of any particular university department. This is not a “save the humanities” argument. A coder who, after tackling a hard systems-design problem, comes out thinking differently about complexity, tradeoffs, and the limits of formal reasoning has undergone a kind of Bildung—but only if the encounter changed who they are, not just what they can do.”

“…if the response to being replaceable is always to train for a different function, you have entered a race you structurally cannot win. The principle that makes your education valuable is the same principle that makes you disposable the moment the function migrates.

The scramble into computer science was an early sign of the trap: students rushed toward the field that seemed safest, and then AI began destabilizing the very functions it trained them to perform.

The flight to function looks rational from inside it; that is what makes it a trap.”

“Humboldt’s solution was to design an environment rather than a curriculum.”

“Aristotle called it scholé. Humboldt had a related word for it: Muße. Both named a kind of structured freedom for the work of becoming, and for most of history that freedom was radically exclusive.

Aristotle could imagine the highest forms of human flourishing only for those relieved of labor by wealth and the work of subordinates and slaves. The good life required freedom from necessity, and in his world only a few could have it. But in the first book of the Politics he imagined something stranger: that if shuttles could weave by themselves and picks could play the lyre, craftsmen would need no subordinates and masters would need no slaves. The “self-guided machine” would mean that the material basis for leisure no longer depended on the unfreedom of others. It is one of the oldest thought experiments in Western philosophy, and we are now enacting it.

Aristotle did not celebrate the prospect.

He understood that freedom from necessity does not automatically yield the pursuits that make such freedom worth having. In his account, those with wealth and leisure often turned to unlimited acquisition or bodily gratification rather than to the activities that justify leisure in the first place.

With AI, we are building something like self-guided machines. Whether these systems liberate or merely displace is not settled. But the possibility of leisure at scale is real enough to become a serious question.

If AI can compress parts of instruction, it may deepen learning where it is used and clear ground for formation where it gives time back. But only if it preserves productive struggle rather than bypassing it.

The alternative is already visible: autocomplete for life. Not just help with expression, but the slow outsourcing of judgment itself. That is Bildung’s antithesis.

Worse, the same technological society enabling leisure is also shaping the desires of the people who receive it. If our dispositions have already been trained toward optimization and outsourced judgment, the freed hours may arrive in hands that no longer know what to do with them.

For most of history, the conditions of formation were reserved for the few. The capacity for it was not.

If scholé at scale is now possible, refusing to pursue it ratifies a world in which full human development remains the privilege of those who can afford time.”

“Bildung is, at its normative core, anti-servility: the effort to form people who cannot be reduced to instruments of external authority, whether state, market, or algorithm.”

“The people who use AI well right now are drawing on judgment they formed before these tools became ambient.”

“The kind of judgment this essay is defending may be a similar afterglow, formed in a world before AI mediated everything.

Without that judgment, you get agency without autonomy.

If the capacities required for non-servile life in an AI world were all formed in a pre-AI world, what happens when that formation stops? You can live on an inheritance for a while.

You cannot educate a civilization on inherited judgment forever.”

6 Vision

Have a Great weekend when You get to that stage,

Sune

Sune Hojgaard Sorensen

Weekend Food For Thought WFFT