But this revenue is almost entirely attributable to two American companies: OpenAI and Anthropic. Mistral reported earlier this year that its annual revenues had reached $400 million, around the same time that Anthropic’s annual revenue reached $9 billion. Since then, Anthropic’s revenue has grown fivefold, to $47 billion; Mistral has not yet released new figures.

Whether this distribution persists matters a lot. It determines who will make money off AI, and who can afford to stay at the frontier. AI progress is expensive. Whether privately or publicly funded, AI projects that cannot generate revenue will fall behind.

In this post, we’ll look at two questions:

1. Why is AI revenue so concentrated?

2. Will frontier companies stay ahead?

Why is revenue so concentrated?

In many industries, the best workers make much more than normal workers: the best football players and musicians make thousands of times the industry average; in AI research, the best engineers earn tens of millions of dollars.

These sectors are governed by what Sherwin Rosen called the economics of superstars, which apply if two conditions are true: imperfect substitution and joint consumption. Imperfect substitution means that quantity cannot replace quality: several bad violinists cannot substitute for one great violinist, and a thousand bad lawyers are not equivalent to one good lawyer. Joint consumption means that the same product can be consumed by everyone at once: a single producer can serve the entire market cheaply (think recordings, software, or Substack posts). If both conditions hold, you get superstar returns.

Both conditions appear to apply to AI. You cannot use multiple instances of a weaker model to approximate the output of a stronger model. New models can also solve new problems, because they can put together hundreds of individually correct decisions. Worse models are more error-prone, which puts hard problems entirely out of reach for them. It took GPT 5.6 to solve a fresh Erdős problem; no amount of GPT-4o instances would have been able to do the same.

New models can also do existing tasks to a standard that no combination of weaker models can match, similarly to how many people can write books or compose music, but no committee could write War and Peace or come up with Bach’s cantatas.

For such tasks, people appear to prefer the better model. Consider emails or job applications. We do not know any believer in the commoditization thesis who is using Kimi 2.5 or Le Chat to help with these simple administrative tasks, despite both being capable enough to do them.

One explanation is switching costs: the effort of choosing different models for different tasks is not worth it when the cost of prompting a more expensive model under a subscription is zero. For companies with more sophisticated processes (and usage-based API bills), switching costs will feature less heavily.

But such businesses might also prefer better models for simple tasks for a different reason. Many things in business and life have nonlinear outcomes: a slightly better sales pitch can win a contract; a slightly better trade is the difference between making and losing millions; and a slightly better job application is the difference between getting and not getting a job. In areas where a small (relative) edge in performance makes the difference between winning and losing, the better model or worker is worth all of the upside it creates.

Sherwin Rosen’s joint consumption condition is a bit trickier. In traditional software, one person using a computer program has no effect on someone else using that program, so everyone can use the best. AI is somewhat different. One person using a model’s weights (its design) has no effect on someone else using those weights, but someone using an AI company’s API does exclude others from using it, because the compute the model runs on is itself scarce even when the model is not. The result we’d expect to see is fragmentation on the compute layer, with many different providers hosting models, and concentration on the model layer, as everyone switches to hosting the best model at a given time.

Subscribe now

Will frontier companies stay ahead?

For the economics of superstars to persist, Anthropic and OpenAI must remain better than the competition. Their models currently being better gives them much deeper pockets: Anthropic makes $47 billion per year and recently raised another $65 billion to train future models. Many companies experience economies of scale when spending money on research: a large company can spread fixed research costs over a much bigger revenue base, while an upstart has to cover those same costs with far less. This was one source of Intel’s long-term dominance, and it is what underpins ASML’s monopoly on extreme ultraviolet (EUV) lithography machines.

But eventually, each additional dollar spent on research should achieve less, allowing other companies to catch up – as happened to Intel, and as may happen to ASML.

ASML and Intel were limited, however, by what kinds of increasing returns they had. EUV machines make them lots of money, but the act of owning EUV machines does not itself make ASML better at discovering better EUV machines. In this sense, AI is different: it can write code, come up with experiments, and implement them. It is uniquely capable of improving itself. (The only other example we can think of is this lathe, which is necessary for making lathes.)

Anthropic is able to use its models for internal AI research where its competitors cannot. One of the reasons we have heard from a friend at DeepMind for the talent flight to Anthropic is precisely this: that DeepMind researchers feel stunted without access to a good internal model. When Anthropic did finally release their strongest model to the public, they made it useless for AI research.

Today, the best AI companies are pulling away from their competitors. xAI is leasing its data centers to Anthropic rather than using them itself, Google DeepMind is losing its superstars to Anthropic, and Meta is spending tens of billions and struggling to produce a good model.

The same is true for cheap Chinese models. They have stayed somewhat close to the frontrunner but are falling behind over time. Last year, the best Chinese companies were only six months behind the best American companies; now, that gap has grown to eight to ten months.1

This is a consequence of how they are made. Many Chinese models are distilled from American models, trained on input-output combinations of Anthropic and OpenAI products. GLM-5.2, the best Chinese model, was so heavily trained on Claude Opus output that it frequently believes it is Claude. But distillation necessarily involves some loss along the way, meaning you cannot surpass the frontrunner. Distillation is also becoming less effective, as Anthropic and OpenAI clamp down on distillation attacks and the underlying intelligence of good models transfers increasingly poorly. (For European companies, distillation probably isn’t even politically viable, given that American companies treat it as a deeply adversarial action.)

The other problem for open-source progress is that open-source companies don’t make much money. Companies that serve open-source models, like Baseten and Fireworks AI, are making money. But the companies that created the models don’t capture any of that value. Making better AI models is expensive, and it is not clear why any provider beyond Meta and TikTok, which have a different channel to make money from models, would perpetually provide them for free, let alone surpass the frontrunners.

OpenAI and Anthropic might not be able to sustain the current speed of improvement. But a slowdown doesn’t seem like the likeliest outcome, given the enormous amounts of additional compute soon coming online, and the fact that the leading labs are only now beginning to effectively use internal models to accelerate AI research.

Cheap models might provide an alternative for easy or unimportant tasks. But even that wouldn’t change much about the revenue distribution between AI companies: once a task can be done by a commodity model, its price collapses. The revenue from this kind of work is small even if it creates lots of the value. In this world, the AI market would resemble something like the market for lawyers, where much work is routine and modestly paid, while a few superstars work on high-value problems and make enormous returns.

The consequences of revenue concentration

At some point in the future, a French or German company may release a good AI model that is only slightly worse than what is made by Anthropic or OpenAI. This model would provide temporary independence.

But a model that is close to the frontier one month will be far behind it six months later. To stay a fixed distance behind the frontier, the model would need constant improving. To fund that improvement, the model would have to make money.

Quite a bit of capital is being invested in companies that are developing or serving models behind the frontier: Mistral has raised at an €11.7 billion valuation, Baseten at $13 billion, Fireworks at $4 billion, and AMI Labs at €900 million. It is fine for private investors to take these risks. It is different when European governments, or those advising them, assume that there is a great opportunity being left on the table here.

To justify government investment, there must be some benefit that the market is not pricing in. That benefit may be national security. But if it is national security, then we should be clear-eyed about the fact that it is governments who will pick up most of the tab. Without financial returns and especially without distillation, staying even a fixed distance behind the best models will be achievable only with exponentially growing subsidies.

• Cyber offense: Anthropic’s Mythos, an AI model that has already identified 10,000 high-risk cyber vulnerabilities, is used by the National Security Agency (NSA) to conduct cyberattacks.

• Cyber defense: The US government is using the same AI model to identify and patch vulnerabilities within its own systems.

• Warfare: Frontier models are integrated into Palantir’s Project Maven and were used for target prioritization in the US airstrike campaign on Iran.

Last week, the White House ordered all national-security agencies to accelerate AI adoption further. Already, new AI models are available to the Department of Defense upon release, and, as per DoD, half of its staff (more than 1 million people) use AI systems in their work.

What about Europe? There is no public record of any European military or security agency using frontier AI in its operations. The French government intends to adopt AI—but the models used would be Mistral’s systems, not OpenAI’s or Anthropic’s.

European AI systems are less capable than American frontier models. Using them might be acceptable for writing emails or reviewing planning applications. But in competitive domains, relative ability makes all the difference: a small but persistent edge between rivals in territorial conflict or cyber warfare will ultimately tilt the balance toward the stronger side. Just imagine you had to improve a server’s security setup with the two-year-old GPT-4o, while your adversary could prod the server using Mythos.

Right now, many still argue we should rely on lagging European models for sovereignty’s sake. But this is quickly becoming untenable. Soon enough, using laggard AI systems will be equivalent to staffing our governments with middling interns, while America and China deploy millions of ever-improving AI agents across their militaries and security services.

Incorporating American frontier AI systems across Europe’s security agencies and militaries will be complex, bureaucratic, and politically fraught. Yet the institutions that could liaise between European governments and American frontier companies are currently being set up—just yesterday, Germany’s National Security Council announced it would launch an AI Security Institute. Making sure these institutes can secure frontier AI access for their governments is getting more important by the day.

The final procurement policy

Over the past year, prominent European institutions have floated the idea of requiring governments to procure European AI systems. Arthur Mensch, the CEO of Mistral, recently argued for such mandates. Similarly, the Commission’s Apply AI Strategy calls for adopting European AI systems in public administrations.

Going through with these proposals would see European governments opt for worse, more expensive models instead of better, cheaper alternatives: Mistral’s Medium 3.5 performs worse than GPT-5.4 nano, while costing fourteen times as much. The revenue thus generated will not be enough to allow European companies to become competitive: large European governments spend around $5 billion per year on IT1. If three such governments together paid Mistral $3 billion a year for model access, the resulting revenue would represent six percent of Anthropic’s annual revenues.

Procurement mandates are bad industrial policy. But they might prove far worse than simply wasting public monies: they would prevent European governments from using frontier AI systems for cyber defense, while others are free to use frontier AI systems to scour our governments’ systems for vulnerabilities.

Since OpenAI and Anthropic launched their cyber defense programs, they’ve discovered thousands of software vulnerabilities. Even among applications that are deemed highly secure, vulnerabilities were flagged that escaped humans for years, including in the operating systems of Apple, and in the open-source OpenBSD (commonly used for running critical infrastructure). It is fair to assume that many European government systems are nowhere near as well secured as the systems of Apple, a company that can pay cybersecurity professionals hundreds of thousands of dollars per year, and OpenBSD, whose code is publicly available and has been pored over by thousands of people.

Now, some European institutions are actually using American AI systems. But, based on private conversations, both their usage policies and the models in use are inadequate.

The EU Commission was still using GPT-4o late last year, well after Claude Opus 4.5 was available, and has only recently moved to GPT-5.1. Closed-source models like GPT-5.1, however, cannot be used on private materials, so staff must rely on open-weight models instead. But again, rather than using reasonably good Chinese models, they are served outdated Llama models hosted on Commission services.

To informed citizens, the EU’s AI use policies should be unnerving. But aside from the performance hit the Commission is taking, interacting with outdated models also makes staff blind to the sheer ability of AI models. Using Anthropic’s latest AI model, Fable, in a coding harness is vastly more impressive than Llama writing mediocre emails, and makes you much more inclined to take cybersecurity risks seriously.

AI Security Institutes could prove central to European frontier access

European AI policy discourse is only slowly taking security issues more seriously. But I think this might change very soon. Prompted by the Mythos release, more governments are studying AI’s security risks. Most importantly, they are launching AI Security Institutes, as the German government did on June 8.

One of the shortest paths to frontier AI adoption in European governments goes through such institutes, both because they engage directly with frontier companies, and because they are (ideally) staffed by individuals who understand frontier AI well. But for such institutes to perform well, they need to be set up well. In thinking about how to do this, we can learn from the world’s most performant such institute: UK AISI. They played a large role in publicizing Mythos’ cybersecurity potential, and can be a template for other nations.

What will matter most for attracting top AI experts into government is pay, mission focus, and fast, flexible hiring. UK AISI has done this reasonably well—they can pay more than the UK government usually would and are free to employ non-UK nationals. Germany should follow this model. It can also learn from the setup of SPRIND, Germany’s ARPA equivalent, which was given more flexible hiring and financing rules through a SPRIND-specific law. A more cautionary example is the EU’s AI Office. While staffed with some excellent mission-driven people, its pay scales are rigid, and months-long hiring processes make it hard to routinely attract top talent. It also follows the same type of country-proportionate hiring that degrades the effectiveness of many other parts of the Commission.

If Germany’s AI Security Institute and similar organizations are modeled after UK AISI, being staffed with individuals who are competent and aware of AI’s direction of travel, they could play a key role in establishing long-run relationships with frontier AI companies. Once such relationships are established, they could provide their respective governments’ security agencies and militaries with reliable, secure frontier AI access.

How to deploy frontier AI without surrendering sovereignty

Now, I understand why European governments would be wary of using American AI systems, given how unreliable the United States has become. But there are ways to integrate AI into our governments that would minimize downsides, while giving us the large benefits of using the world’s best AI systems.

For general public service use, governments should simply act like any other market participant: purchasing different AI systems and thus reducing the dependency on any one specific company. There are European companies like Langdock that allow users to access many different AI systems through one portal, even ensuring European hosting of some AI systems. At the same time, having parallel bespoke agreements to host compute for public sector AI is commendable—the OpenAI for Germany program is one such example.

For national security use, I expect we want solutions that are even more secure, just like the American government wants compute reserved for national security applications. Comparing artificial intelligence to prior dual-use technologies is somewhat treacherous, given how different AI is. But we can still learn from how the West shared dual-use technologies among Allies in the past.

One example is the United Kingdom’s Trident program. Under it, the UK has full control over its own nuclear weapons, though it requires the United States for maintenance. Similarly, American frontier companies could work with AISIs to provide individual European governments with secure frontier access, using air-gapped, domestic AI clusters that run the same AI models accessed by the United States government.

Such models would require frequent updating. But even if updates happened a couple weeks after model release, this would be vastly preferable to relying on systems that lag the frontier by a year or more.

There are surely better mechanisms than the one I briefly sketched here. But almost anything beats the status quo: in competitive domains, laggard systems are entirely inadequate. AI Security Institutes are best positioned to make their governments understand this. And while the subsequent work of providing European governments with secure frontier AI access will be hard, it will prove crucial to maintaining Europe’s security.

Compare this with the United States. The scale of its privately-funded data center buildout is staggering. Amazon’s Project Rainier has already brought online 750 MW, with plans to scale to over 1200 MW by June this year. Meta’s Prometheus is at 700 MW, scaling to 1275 MW by October. xAI’s new Colossus 2 data center will likely come online in May with 1500 MW. In sum, Prometheus, Colossus 2, and Project Rainier would consume all power generated by the UK’s Hinkley Point C nuclear power plant, planned to come online in 2030.

Together, these three US projects absorb around 70 billion dollars—three times as much as the European Union wants to put toward AI Gigafactories (which are yet to be announced). On current trajectories, the United States is projected to have around 24 GW of capacity by 2029. The UAE adds another 1.5 GW. Ignoring projects whose fate is uncertain, the share of European capacity will be a fraction of America’s.

The Commission’s goal of having more data center capacity in Europe is a good one. But data center construction is a poor fit for governments, and a particularly bad fit for the European Commission: data centers aren’t a commodity like oil, which you can purchase and store for local use. Instead, they are incredibly complex infrastructure projects and depend tightly on private investments and subsequent private demand to be financially viable. To go ahead in the first place, they often need financial backstops from trillion-dollar companies such as Google or Amazon.

The demand for more compute is only increasing. The price for renting high-end AI chips is rising quickly, because of the many individuals who want to access ever-improving American AI models. To alleviate this shortage, AI companies continue to scout for locations where they can transform hundreds of billions of dollars into compute.

This shortage could, in part, be addressed by European-American collaborations that build new data centers in Europe. Such buildouts are possible. Nscale, a British company, is building a 230MW facility in Norway for use by OpenAI, which might be expanded by an additional 290MW. A 720MW project in Northumberland, UK, funded with £13 billion by Blackstone, is currently undergoing site preparation.

Europe would benefit from such investment, even if data centers on European ground will first run American AI models. Yes, European citizens and companies could also use AI hosted in the US. But even if financed by American companies, building AI infrastructure in Europe creates the pre-conditions for new sources of European growth: Inventive Bavarians could launch robot manufacturing companies by accessing low-latency frontier AI. The French could develop new nuclear power plants that power data centers over the long haul. Domestic expertise in building large data centers could allow the construction of clusters used by researchers at ETH or the Max-Planck Institutes. Ultimately, the interests of European nations and (most) American AI companies align in important respects.

Europe could also reject such investment. But in my mind, there is no clear alternative—the domestic funding for new data centers simply isn’t there. And I worry that the Commission will not realize this anytime soon, as it is invested in its own data center projects.

Compared to Brussels, capitals like Berlin, Paris, and Oslo are more likely to take the kinds of high-upside, unilateral decisions that enable privately-run compute buildouts: turning power plants back on, prioritizing grid connections, and providing expedited permitting. Ultimately they could even lobby the Commission to help in the build-out by removing barriers to data center construction1. Engaging with them, not Brussels, is the key to compute in Europe.

The perils of compute poverty

AI infrastructure projects resemble a game of Factorio—a highly addictive infrastructure game where one smart strategy involves generating power with hundreds of steam engines. In this vein, one Utah AI infrastructure project has ordered 700 Caterpillar G3520 engines: Weighing 19 tons, each of these piston-driven, gas-fueled generators produces 2.5 MW, yielding 1.75 GW in total. Using a small number of large-scale gas turbines would be far more efficient—but those are sold out for years.

No project of similar size is undergoing construction in Europe. Though there is talk of a 1GW project in France, one of the involved companies—Fluidstack—recently pulled out of the deal. As of now, Mistral (the leading European company) is targeting a mere 200 MW by 2027. If no other projects begin construction soon, the continent will end up with a minuscule share of global compute by 2030.

European compute poverty could be harmful. For instance, German companies might be able to build technologies complementary to AI, such as advanced robotics. Tom Davidson has previously laid out how such complements could help non-US nations experience AI-enabled growth:

“The US will have abundant cognitive labour. That cognitive labour will be strongly complementary with other economic inputs like human manual labour, physical capital, and raw materials. Those other physical inputs are very much distributed worldwide. So getting the most economic value out of AGI will involve significant trading with other countries. [...]

In practice, US-built superintelligent AI systems will instruct human workers in (e.g.) Germany on how to use existing German factories and machines to build new and improved physical technologies like robots. If there’s just one US AI company, they might sell cognitive labour at monopoly prices and extract most of the gains from trade. But if there’s multiple competing AI companies, the German companies providing the physical inputs might capture most of the gains from trade. Either way, the manufacturing will take place on German soil.”

This is one small but positive example of how economies without frontier AI companies can capture AI-driven growth.

But if German companies want to focus on robotics, they might need access to very low latency compute. Agents interacting with the physical world will need to perform continuous, seamless inference calls. If those calls need to run across the Atlantic seabed, German robotic systems’ performance might be low. Compare this to China: They are capable of rapidly building large amounts of energy and data centers, the resulting efficiency of which might compensate for having slightly worse AI systems. China might thus outcompete the West’s industrial sector in one of this century’s most important industries.

German robotics is of course just one example of how Europe could profit from AI adoption. The larger point is simply that local compute, even if initially used for American AI systems, will improve European productivity. Such gains from trade alone might motivate American companies to place more compute in Europe. But enabling rapid data center buildouts will require action by European capitals.

A German example

Working on tech policy in Europe can make one frustrated with the European Union’s many regulatory barriers. But all too often, member states are to blame too.

A prime example is the EU’s “Energy Efficiency Directive”. Because it’s a directive, the precise implementation is left up to national governments. For this directive, Germany’s previous government engaged in severe gold-plating—the practice of adding things on top of European regulation. The original directive only contained reporting requirements, but Germany’s law (Energieeffizienzgesetz) went further.

First, Germany’s efficiency law captured more facilities, moving the cutoff from 500 kW down to 300 kW. They also added binding power usage effectiveness (PUE) rules, i.e., mandating facilities to spend little energy on anything beyond their IT equipment. Finally, they added obligations on waste heat reuse e.g., via municipal heating networks. None of this the EU asked for.

Many other European nations have committed similar mistakes in the past. But with targeted policy engagement, they might be fixable. Reviewing the German government’s current work, there are plenty of valuable policy windows one could use to improve things.

Right now the new German coalition is rewriting the Energieeffizienzgesetz. They already propose loosening the data center power effectiveness requirements so that they always exempt AI data centers2. They will also amend the heat reuse obligations to make them not apply to AI data centers in the first place.

Additionally, the government has recently passed its data center strategy. It is not yet ambitious enough: they plan on quadrupling AI capacity by 2030, which would amount to adding 1.5GW—as much as the aforementioned 700-piece fleet of Caterpillar engines in Utah will produce. Still, the strategy contains smart proposals: instead of taxing data centers by the number of people that work there (of which there are few), they might tie taxation to the amount of capital deployed, or revenues created. Attaching tax income to the actual economic value of data centers will make municipalities far more likely to welcome their construction.

Data center realism

The politics of enabling foreign data center investments remains very tricky. It runs counter to many political instincts, particularly in Brussels. But the German government’s approach points at some partial solutions. As you get closer and closer to the economic activity generated by data center investments—foreign or domestic—the likelier you are to get buy-in from political stakeholders. Governments are interested in hosting hundreds of well-paid AI engineers. A municipality might welcome funding with which they can improve local roads and schools. Domestic companies would be eager to partner with the world’s best AI companies. Hence, it is at the national and state level where the path to Europe-based compute is easiest.

Right now, European governments are not close to the AI ambitions of the United States. In part, this can be explained by policymakers not having heard good explanations of the strategic importance of AI. But that’s exactly why nation-based policy engagement is so valuable. Germany and other governments are already moving in the right direction, they just need to be pushed to act more ambitiously. One such policy priority will involve attracting and expediting data center investment.

Most likely, a successful AI infrastructure buildout will be a precondition for staying relevant in the 21st century. But if you want to make this buildout happen, don’t focus on the Commission. Nation states are far more likely to appreciate that only the private sector is ambitious, pragmatic, and risk-tolerant enough to take on infrastructure projects requiring tens of billions of capital. It is on us to argue why they should enable the private sector’s ambition.

Europe’s lack of compute is concerning. The US is receiving investments, growth, and jobs through its AI infrastructure build-out, while few such build-outs happen in Europe. As a result, European companies and governments also miss out on the tacit knowledge gained by building large-scale datacenters. Compute itself has both economic and strategic importance: economies and militaries that run on top of frontier AI systems will be far more efficient, making restrictions on frontier AI system use devastating. But such restrictions might become harder once several companies host AI infrastructure in European nations.

Hence, Europe should prioritize attracting compute build-outs. Some governments and parts of the European Commission have begun trying to build local data centers, but many of these efforts are much too statist. The European Union frequently talks about funding AI infrastructure, but a multilateral, consensus-based institution is not set up to orchestrate the construction of massive infrastructure projects: You need to decide on one site, which creates competition among member states, and the amount of capital required creates financial risk that governments are unlikely to take on. Most importantly, there is no European frontier lab acting as the final customer: the Commission’s “AI Gigafactories” are not moving forward because companies like Siemens and SAP simply do not require that much compute.

But there are entities out there who share classic European values, and who need compute: American AI companies.

Convincing European governments to host American companies is not particularly easy. But European nations need frontier AI systems. The more components underlying those systems are based in Europe, the better. And American companies might themselves have an interest in hosting some of their compute outside the US. Their own government is not following basic tenets of liberalism, such as respecting Anthropic’s private property. Building compute in the UAE is looking more precarious, now that Iranian drones hit AWS data centers. In comparison to its competitors, Europe is looking better each passing day.

Liberal compute

After the US executive branch crackdown on Anthropic, there has been talk of getting Anthropic to move to Europe. I believe this is very unlikely: Right now, Anthropic would be leaving its entire infrastructure behind, which the USG could possibly seize. Most importantly, it’s politically untenable, both for Anthropic, and for European governments1.

But a more pragmatic long-term plan would be to get AI companies to build more of their compute in Europe. Given the few good AI policy options European governments are faced with, this is one of the highest upside actions they can take.

Ideally we’d have a European company with demand for such compute. Alas, that’s not the world we’re in.

Instead we could either choose to remain stuck in a situation where American companies are running models on clusters based in the US, or one in which more AI clusters are constructed in European nations. As a European, the latter world still seems far preferable to me. Though I think relations between the US and Europe have a chance of becoming calmer, and the US might become more liberal again, it remains possible that the US government tries to cut off European access, especially once the power of AI systems becomes obvious to all. This will be harder if European economies are working with several AI companies at the same time, all of whom have their inference infrastructure hosted in Europe.

There is a further, much more speculative point here. The type of government that would make the US try to cut off Europe might also be staffed by people adversarial toward some companies like Anthropic. In such a world, AI will be highly advanced, so those companies themselves will be powerful and might resist government coercion, especially when this government is not aligned with their priorities. This scenario is unlikely, but if it were to occur, giving companies an escape valve where they retain power by using European compute would be highly valuable. For this, companies will require having such compute available in the first place.

European governments are currently not considering such outcomes. Unlike the leaders of some Gulf states, few European heads of state are cognizant of the massive impacts of AI. But we still need to make plans for the moment they become aware.

The perks of Europe

At first glance, building infrastructure in Europe is difficult—its environmental laws are strict and its energy prices ($90/MWh) are often twice as high as in the US ($48/MWh). But some of these problems are smaller than they appear at first glance.

In Germany, construction has been getting faster: Germany’s first LNG terminal was planned and constructed in 10 months. Tesla’s first European factory in Brandenburg went from announcement to construction in 2.5 years. Unlike the US or the UK, the country is further simplifying its permitting rules, labeling a lot of infrastructure as being in the “supreme public interest”.

In Norway, construction of a large data center is underway: OpenAI is building Stargate Norway, with a final capacity of 520 MW. This is in part enabled by hydropower plants owned by Nordkraft, which can offer low energy prices because it sits in an area with limited demand and poor transmission capacity.

Now, one should stay clear-eyed. Stargate Norway is nowhere near the scale of projects abroad. Meta’s Hyperion is planned as a 2GW campus, with a final capacity of 5GW. Stargate UAE’s final size will be 5GW. Training models in Europe is made difficult by strict data laws and copyright protections (which make demand-side AI measures by the Commission ineffectual, given that their supply-side restrictions negate any stimulus and innovation).

But attracting compute to Europe remains a great strategic option. Unlike many other measures the Commission is pursuing, easing American compute buildouts works along the desire of markets, not against them. While training models in Europe is difficult, inference demand will only increase. As of mid-2025, OpenAI uses half of its compute for inference, and this share will likely rise further. And while companies currently serve European AI demand from the US (because latency doesn’t matter with slow-running frontier models), this might not be tenable for long, as interactions with AI become more seamless (e.g., when using voice interaction).

Laying the ground work

There is still a lot of intellectual work to be done on what a realistic pathway to more European compute could look like. But several actions look pretty good already.

First, national capitals need to take AI seriously as fast as possible. Right now, many of them simply lack the knowledge required to understand what is coming. When doing such advocacy work, one should focus on countries that fulfill criteria required for quickly building large data centers: fast permitting, local taxation that minimizes NIMBYism, government structures that allow rapid action, and pragmatic attitudes toward American AI companies. That rules out some countries2, but leaves a number of viable candidates.

We also need to be more forthright about why existing sovereignty initiatives can’t be the answer to being behind—though when pointing this out, we should ideally point to constructive alternatives. The government cannot coordinate the buildout of compute infrastructure, but large companies that promise to spend multi-billion foreign investment might have more success. Not all companies involved in such build-outs must be foreign: there are hyperscalers with European roots, such as nscale or Nebius.

Finally, we need more work pushing the European Commission toward a supply-side approach to AI. Instead of regulating the use of AI (see previous piece), slowing down Germany’s pivot to gas power plants, mandating energy-efficient data centers, and hampering frontier AI training, they could help by identifying the largest regulatory bottlenecks to AI development and deployment, and unblocking those.

Giving European nations a stake in the future of artificial intelligence is crucial. We can get closer to this world by moving more data centers to Europe. There is still vast strategic complexity for how to best achieve this, but thinking about which actions to take sooner rather than later is one of our better bets.

Thanks to Dominik Hermle for reviewing a draft of this post.

Such a tool can be built today—I created a version of it in 20 minutes. To deploy it in Europe, I would first need to work through a thick layer of regulation. As per the AI Act, Annex III, point 3b, such a system would soon1 be considered “high-risk” because it is

“intended to be used to evaluate learning outcomes, including when those outcomes are used to steer the learning process of natural persons in educational [...] institutions [...]”

What obligations would I face? First I’d need to review guidance documents to understand if the system is high-risk (Article 6). If it is, I’d need to create a risk management system (Article 9), implement data governance measures (Article 10), list the system on an EU database (Article 49), and—if I teach at a public school—provide a fundamental rights impact assessment (Article 27). There are further obligations not listed here.

Now, the AI Act contains various exceptions and qualifications (though I doubt they apply, see footnote2). But the process of understanding if something is considered high-risk or not already creates large amounts of work and, depending on how you decide, legal risk.

Given the state of AI today, these rules are already unworkable. Tasks that can take humans hours, take AI a few minutes. Building something far more advanced than the math tool, it took a Polish doctor only one week to create an AI platform that helps patients navigate their condition after leaving the hospital. This too, might be considered high-risk3. The rules’ effects will become worse as AI improves further. As per research by METR, the complexity of tasks that AI can accomplish is rising rapidly, and both AI company CEOs and forecasters are predicting human-level AI systems to come online in only a few years.

Predicting all the future use cases for AI is a hopeless task—it is a bit like asking someone to predict how electricity might be used in 1882, the year when Edison built Pearl Street Station, the world’s first power plant. But this also makes highly detailed, use-case based AI regulation untenable. Once AI is fully integrated into most people’s everyday lives across the entirety of Europe, the AI Act’s high-risk regulations might get triggered thousands of times each week.

The European Union is currently debating how to change the AI Act, as part of a larger simplification law, the “EU Digital Omnibus”. Looking at the Commission’s proposal for how to amend the AI Act, it contains some obvious improvements: In the original law, Article 6(4) would have required individuals to document why they considered a tool non high-risk and register it on an EU database. However, neither of the high-risk categories (Chapter III, Sections 1–2), nor the actual obligations (Articles 9–15) are simplified or removed.

The economics of EU digital regulation

An obvious rejoinder to the above might be that most individuals and many companies who accidentally use AI for “high-risk” use cases will not know about the AI Act (even though they might face fines). At the same time, the member states tasked with enforcing the AI Act’s high-risk regulations might not have the capacity to track and fine each breach of the law.

This points to a more general rule about regulations: they rarely get enforced. Instead they often end up hitting companies that do not generate much sympathy anyway, from American tech firms that provide much of our digital services, to Chinese e-commerce companies that threaten European retailers. Based on the political economy of the day, rules are used selectively.

I’m not particularly sympathetic to Chinese e-commerce firms. But while European companies might not always be affected by EU regulation, the risk of a multi-million or billion Euro fine is still looming in the background. This depresses investment and economic activity. After GDPR passed, the Economist mostly deemed it a fine law, but US investment into EU ventures then dropped substantially, while no such drop was observed for US ventures. Existing businesses too were harmed by the law: for GDPR, compliance costs were substantial, ranging from $1.7 million for SMEs up to $70 million for large companies. Seeing similar costs for the AI Act is plausible.

Europe could grow by applying AI

Many Europeans are already using AI enthusiastically, with many of them using AI about as much as Americans. Company adoption has been slower, for which there are many different reasons4. The high-risk regulations will further hinder adoption.

Realizing the bureaucratic burden their citizens and companies will soon face, governments like Germany or Denmark are lobbying the Commission to simplify the AI Act’s high-risk regulations. Germany’s new Minister for Digital Affairs, Karsten Wildberger5, has asked to delay the most onerous high-risk regulations. Friedrich Merz talks about reducing EU regulatory burdens all the time. But the same Commission which is taking the lead on amending the AI Act has come up with all these rules in the first place. The Commission’s incentives run against its taking ambitious actions to reduce regulatory overhead.

If you review the commission’s most recent simplification draft, it makes no changes on which use cases are considered high-risk, or on the specific obligations those use cases face. The member states themselves aren’t doing much better—the European Council plans to reinstate Article 6(4) (the requirement for individuals and companies to register systems they considered not being high risk).

The European Union’s inaction should concern those who want a stronger Europe, but also everyone who likes the AI Act’s sensible parts. Its rules for “general-purpose AI systems with systemic risks” are worth preserving: they apply only to the most powerful systems and try to reduce CBRN, cyber, and loss of control risks. Taking the positive potential of AI seriously should both make one wary of the Act’s high-risk rules and favor its measures on preventing worst-case risks. But, if European nations can’t apply AI in the first place, they will continue to fall behind other countries, some of which are adversaries. Frustration with regulations like the AI Act will only grow, making it more likely that both its worst and best parts will be cut.

The AI Act’s high-risk regulations sit at an awkward place in the discourse. AI safety advocates don’t talk about their harm, as they possibly fear any change to them will also introduce changes to GPAI rules. Traditional industry groups dislike them somewhat, but their opposition isn’t strong enough, both because they do not yet grasp the profound economic impact AI might soon have, and because some of them worked to have their sectors not be considered high-risk. The Commission drafted the original AI Act and its recent draft amendment, putting them at an advantage in understanding the law’s complicated rules vis-a-vis the EU Parliament and the Council.

The potential damage of fumbling AI adoption from the very beginning should focus people’s minds. Instead of covering all of Europe with fuzzy, hard-to-understand rules, it seems much more straightforward for European governments and their citizens to experiment with AI, observe the effects, and then, if need be, intervene at a national level in cases where the harms outweigh the benefits. Experimentation across Europe will also make it more obvious which rules are sensible, and which aren’t, whereas decreeing the same guidelines across the continent will only allow us to compare ourselves with the United States or China. Prior European rules have already reduced innovation and growth across most of Europe, which we were only able to realize after years and years of lower rates of innovation compared to less regulated, faster-growing countries like the United States or Switzerland.

I believe that both those in favor of making the AI transition go well, and those in favor of European growth (and the many people who hold both priorities) could help reduce the harm of the AI Act’s high-risk regulations. The many groups that advocate for keeping GPAI should keep doing so, but they can increase their credibility with non-safety actors by advocating for easing and removing the AI Act’s many other unworkable sections. Meanwhile, the growing ranks of those who want to see Europe prosper should simply put more focus on the AI Act’s worst components. AI could be crucial for advancing European growth and security. Squandering this opportunity from the start would be a painful oversight.

Statistics support first impressions: Switzerland is far wealthier than most of Europe.

The country has a median household income higher than that of the United States, a life expectancy similar to Japan, and the world’s highest share of Fortune 500 companies relative to its GDP. All this wealth creates plenty of surplus: a strong welfare state, decent public schools, and some of the most advanced animal welfare laws in the world.

Switzerland’s success should be interesting to anyone who cares about European prosperity. But the country is rarely discussed; I see an in-depth piece on Switzerland maybe once every two years. Given Europe’s stagnant economies and increasing debt burden, this seems like an oversight—we urgently need ideas for increasing European wealth, if only to fund our defense.

Switzerland could serve as an example that pro-growth advocates can point to. But to do so, we need an explanation of how the country got so affluent in the first place.

Explanations for Swiss wealth

When I returned from Switzerland, I asked people about the origins of the nation’s wealth. A small set of explanations frequently come up. Many note the country’s banks. But while the financial sector profited from the country’s previously lax financial regulation, banking and insurance make up only 9% of its GDP.

Some mention neutrality: ever since the Vienna Congress in 1815, Switzerland has not participated in any foreign wars, sparing its destruction in both World Wars and plausibly increasing investor confidence. Others note the country’s stable institutions—the country’s modern nation-state has existed since 1848 and has seen no major political upheaval since then.

Neutrality and institutional stability surely explain part of Switzerland’s long-run growth. But there is another axis along which Switzerland differs from most of Europe: its economic governance is far more liberal. All else being equal, this should explain a large part of Switzerland’s economic success. Now, all things are very much not equal, but Switzerland is really very liberal!

The country has low corporate taxes, flexible labor laws, and a large number of free trade agreements1. Unlike France or Germany, income taxes are mostly levied on the state level, incentivizing local administrations to run efficiently for fear that citizens will move to low-tax states. This provides the country both with a debt-to-GDP-ratio of only 38% and some of the lowest income taxes in Europe2:

Is Switzerland just an economic free-rider, harming its neighbors by pulling in wealthy foreigners and multinational companies? Swiss policies do attract plenty of foreign companies and workers (25% of its population is foreign-born). But I don’t think they are free-riding: workers in Switzerland put in plenty of hours—around 200 more than workers in Germany—and Swiss companies are productive: across various innovation rankings, Switzerland often ranks first. I think an easy heuristic on free-riding goes as follows: Do you think Europe would lose out on growth and innovation if Swiss taxation and regulation were raised to the European average? I’d certainly think so.

Ultimately responsible for the liberal policies outlined above is a populace that is majority pro-market. Relatively liberal parties always make up a majority of the government, which regularly passes free trade agreements, e.g., with China (2013), Indonesia (2021), and India (2024). Voters reject nearly all ballot initiatives that might threaten Switzerland’s economic position: a 2012 referendum increasing the number of mandatory holidays was rejected by 66% of voters.

Explaining a country’s wealth by pointing to its people being pro-market seems ideologically convenient. But economics is a science with many clear findings: high taxes can reduce government income; making it harder to fire people increases unemployment; tariffs are inefficient and lead to cronyism. It wouldn’t be surprising if a country following economists’ recommendations turns out to be fairly prosperous.

However, it’s still somewhat surprising that the Swiss hold these views. Many other countries could enact more liberal policies, but they do not. What made Switzerland so liberal in the first place?

1848: Liberal lock-in

To understand the onset of liberalism, we can review past Swiss elections. Regular elections date back to 1848; reviewing vote shares from then on, Liberal parties (“Freisinn”) dominated the legislative and executive for the first 70 years of the country’s existence.

Why were the Liberals in such a dominant position from the get-go? Without going too deep into 19th-century Swiss politics3, here’s a rough sketch.

Prior to 1848, Liberals in Switzerland already had broader support than in most other countries. Luckily, they also didn’t face a monarch with centralized power and a large army. From this position of relative strength, the Liberals won a brief civil war (with only 82 dead) against the Conservatives in 1847, giving them near-complete control of the country’s political fortunes. With this power, they gave Switzerland a government modeled after the United States: two chambers, one representing the population, the other the states, with both using first-past-the-post voting.

To my eyes, 1848 therefore looks like an important moment of political and institutional lock-in: The Liberals won a civil war and controlled the country for decades through their absolute majority, while also putting in place a fairly limited central government.

This lock-in wouldn’t matter if Switzerland’s current level of wealth were due to policies enacted in the past 70 years. But Switzerland started to outgrow Germany quickly after 1880, suggesting that Switzerland’s initial institutions and policies set the foundations for strong long-run growth.

Starting the growth flywheel

What did the Liberal government do to get Swiss growth started? Pointing to the definite sources of long-term economic growth is tricky, but we can at least make some guesses. And post-1848 events provide some good candidates. First of all, Liberals enacted the kinds of policies required to have a functioning economy by abolishing internal tariffs, setting up a nationwide currency, and enabling relatively free trade with its neighbors.

In parallel, the decades after 1848 saw a frenzy of institution-building, many downstream of the actions of Alfred Escher4, a major Liberal statesman/industrialist. Among many other things (he launched ETH Zurich and spearheaded the construction of the Gotthard tunnel), he strenuously lobbied for a private rather than public buildout of Swiss railways. Winning the argument, Escher himself founded a railway company, realized he needed large amounts of capital, and also created the “Schweizerische Kreditanstalt”, the progenitor of Credit Suisse. The subsequent railway buildout created one of the densest transit networks in the world.

With basic policies and institutions in place, and the subsequent adoption of hydroelectric power, the Swiss economy was able to power a rapid shift from agriculture toward industry, launching both electronics (e.g., Brown Boveri, later ABB) and chemical industries (e.g., Roche). By 1913, Swiss GDP per capita was already one of the highest in Europe. Maintaining its industrial heft until after the 2nd World War, the country then switched more and more toward services and is now a highly advanced, diversified economy.

So what should Europe do?

Throughout its modern history, Switzerland was lucky. A commitment to liberalism is easier without war—economic policies needed during crises, such as high taxes or price controls, can stick around long after, depressing long-run economic performance. Similarly, lower rates of growth can increase demands for higher taxes, which lowers future growth even further. All of this makes Switzerland an imperfect template. I certainly don’t want the rest of Europe to become neutral!

But overall, Switzerland still seems like a good example of what a century-long attachment to liberalism can yield. In the absence of catastrophe, growth has become self-reinforcing. By now, a central goal of Swiss politics is to keep the golden goose alive. While it would take a lot longer to identify all the other factors that underpin Swiss growth, liberal governance seems among the most obvious candidates.

This is useful: when arguing for liberal economic policy-making, you could now point to Switzerland, not the United States. The US has lots of dysfunctions that make it a tough sell. On the contrary, there’s much to like about Switzerland for people of most political stripes. It is economically liberal, while also having high trust in government and a working welfare state. After my visit this summer, I’ve started pointing to Switzerland as a template for a European rebound far more often. Maybe you want to do the same?

Thanks for reading! If you have thoughts, shoot me an email at simonleandergrimm@gmail.com.

• German academia doesn’t have world-class universities and is self-avowedly egalitarian.

• Without a clear top university, many talented students instead enter highly competitive medical schools to prove their ability.

• But, as argued here, medical school is a bad default choice for these students if you care about accelerated scientific, material, and moral progress. This is for four reasons:

  1. Entering many different universities instead of one top college, talented students do not generate and thus do not profit from local agglomeration effects.

  2. Medical students aren’t allowed the intellectual flexibility to explore ideas and projects independently.

  3. Medical school takes six years, offering no intermediate degree. This locks in students' choice of study, even if they change their minds.

  4. Lastly, practicing medicine offers small impact at the margin (i.e., talented medical students can’t add much to an already highly advanced medical system).

• Instead, talented individuals could study subjects and enter jobs that allow them to do much more good.

• Changing this status quo is difficult, as i) strong competition between universities is probably disliked by university administrations and ii) reforming existing universities is famously hard through entrenched bureaucratic decision-making and ensuing vetocracy. Thus, change might only be possible through affluent outsiders who launch a new, better university.

Epistemics

I’m uncertain how big the problem of bad talent allocation in Germany really is. However, given that it reduces the likelihood of careers with an outsized positive impact, the damage could be severe. 

I would change my mind on the importance of this problem if i) elite colleges only add credentials without increasing students’ human capital, or if ii) research breakthroughs, new companies, or social movements aren’t made more likely by improved talent allocation. 

Introduction

German undergraduate education is largely not considered world-class. Defining and ranking a “world-class university” is fuzzy and unsatisfying, but many still know which universities have the most prestige. For the sake of being more concrete, let’s defer to Matt Yglesias to pick a criterion. He notes:

“Boris Johnson’s government recently rolled out an initiative to let recent graduates of the world’s top universities move to the UK. That required them to come up with a definition of the world’s top universities. Of the 38 on the list, a staggering 20 are American. And while it’s true most of those are private [...], there are six US public universities on the list versus five European universities. There are five California universities on the list (Stanford, Caltech, and three UCs) versus one from Germany [LMU Munich].“

Hence, every year American high school students jostle for a rare spot at one of the country’s Ivy League universities. While students will also take other factors into account, they will want to go to a university that provides a credible signal of their ability. This might be even more true for ambitious or talented students, as the mere signaling value of being admitted to an elite university is often enough to guarantee a stellar career. This explains why Oxford graduates with a language degree can join DeepMind right after university. 

In contrast, the absence of a clear top university leaves German high-school students with no obvious university to go to. (We can test this by asking German and US-American students where they want to go to college after high school. Predicting the answer in Germany will be tricky.) This is partly by design, stemming from a stronger egalitarian streak in German academia. 

A troubling century

Where does this stronger desire for egalitarianism come from? Looking one century into the past, German universities were renowned throughout the West. Various American universities that are by now world-leading explicitly took the “Humboldtian” model of higher education as a gold standard worth emulating. As detailed by William Kirby’s recent book “Empires of Ideas”, German universities lost much of their standing through the upheaval and destruction of World War II. A figure in Miguel Urquiola’s ‘Markets, Minds, and Money’ plots this decline, showing which universities Nobel prize winners attended before receiving their Nobel awards.

Subsequent reforms made many universities increasingly ungovernable through exaggerated democratization of university decision-making. Finally, the horrors of Nazi ideology made explicit talk of nurturing an elite unpalatable. 

The cultural effect of this past manifested in a recent conversation with a German professor: When I pointed out the potential value of building a German Ivy-league type university, he was concerned mainly by how this would make all other German universities look second class. 

What do German undergraduates study?

Faced with a university landscape with no clear hierarchy, students disperse among many universities that do not differ much in their prestige. 

Students still want to distinguish themselves through their ambition and intellect. Thus, instead of competing to get into a high-status university, many instead vie for the most prestigious course of study, which in Germany is still, by and large, medicine. 

The source of this prestige is legible: spots for studying desired fields in popular universities are often mostly allocated by GPA; medicine is the only field requiring perfect GPAs (1.0; equalling all A or A* GPAs). Hence, not wanting to waste their perfect GPAs, many students enter medical school at various universities. Below are some above-average universities and their required GPAs for competitive fields. As we see, medicine is the only field that requires all A*:

You might wonder why students' love for medical school should be questioned, given the respect owed to medical doctors. While not disregarding medical doctors' self-sacrifice, there are two important reasons why medical school shouldn’t be talented students’ default choice. First, students (and their society) would gain more by studying flexible subjects, at one location, with other like-minded students. Second, the actual impact of additional medical doctors in Germany is fairly small. Let’s take a look at these considerations in turn.

Empowering students can have great value

The obvious choice for talented students in the UK and the US are universities like Oxbridge or Ivy League colleges. Though these universities have their own issues, they serve as an alternative that highlights the downsides of German undergraduate education.

To begin with, instead of going to medical school, some A* students could live up to their strengths more flexibly by entering US-style undergraduate programs. Equipped with a broader set of skills and the signal of an Ivy-league, graduates can enter a wide variety of companies, work their way into prestigious publications, or push the technological frontier at start-ups or well-funded research institutes.

Furthermore, lacking a national Schelling point for talented students leads to weaker agglomeration effects. Though economists typically use such effects to explain the productivity of London or Silicon Valley, bunching bright students together can have similar outsized returns. Innovations achieved by outfits like MIT’s Building 20 (home of the first LIGO experiments) and Bell Labs (development of the transistor, the photovoltaic cell, and information theory) illustrate the value of small but strong intellectual hubs.

Though undergraduate students at Ivy League-type universities only spend 3-4 years together, their interactions likely last long after graduation: Investigating hyperlocal agglomeration effects, Matt Clancy of New Things Under the Sun shows that once innovators find each other, they will continue to work and collaborate with one another even after geographic separation.

Instead, without a clear intellectual hub to move to, many talented German students choose to enter medical studies at whatever university instead of all entering the same institution. This limits students’ opportunity to build numerous, long-lasting relationships with similarly gifted students.

More doctors, little value

But so what? After all, many Western economies complain of acute shortages of medical personnel (though in many cases, this shortage is largely self-inflicted). Practicing medicine is also still widely respected, ranking as the 2nd most trusted profession (right after nursing) in a 2021 Gallup poll:

Even so; although medical doctors do good (adding approximately 4 DALYs for every year of work), there are many other ways to do vastly more good than working as a doctor (for instance, by donating $3,500 to the Against Malaria Foundation, thus saving around one entire life). 

In economic terms adding physicians to a highly advanced medical system offers comparably small impact at the margin. The reasons for this are manifold. As noted in this far more extensive article by Gregory Lewis, Western medical systems are already filled with well-trained specialists who are aiming to rid us of remaining non-communicable diseases. All the while, (most) low-hanging fruit that extend life span have already been plucked by curbing endemic infectious disease, fighting air pollution, and a broad decline of violence.

One could now try to salvage medical careers by pointing to the value of biomedical research. But again, this area of research is i) also far from neglected and ii) more impactful when pursued in the US. For instance, Boudrea et al. analyzed publications in four major medical journals between 2000 and 2009, noting that

“Harvard researchers account for around 5 percent of scientific articles published in the top four medical journals, a larger share than Germany or Canada as a whole.”

Any bright-eyed student fascinated by the life sciences should thus first try to get into a British or American university (and, if possible, apply for a scholarship by the DAAD, and the German or Swiss Scholarship Foundation!)

Conclusion

Smart high school graduates need better options than the current choice between homogeneous and middling universities, making too many of them opt for medical school. Importantly, the argument does not call for fewer students at medical schools but a lower share of particularly talented students. Generally, good physician performance likely requires above-average, but not outlier cognitive performance while also resting on factors such as social intelligence, conscientiousness, and stamina. Based on this, there are likely enough individuals fitting these criteria who would be happy to replace students currently attending medical school.

To get towards such a world, we could either lower the allure of medical school or offer a different high-status option that provides greater societal value.

To achieve the former, medical school admission officers could decrease their focus on GPAs, instead only requiring an entry test. Though testing would still be competitive, fewer students would think they would “waste” their A* grades if not attending medical school. 

To achieve the latter, one would want to change German higher education fundamentally by creating one or two clear champions among universities. Such change would likely need to come from outside current universities, as the status quo suits many senior German academics. I’ll dive into the merits of launching such a university in a future blog post.