Europe is Losing

[00:00:00]January 2025 Deep Seek.

[00:00:08]>> Ladies and gentlemen, too often I hear that Europe is late to the AI race.

[00:00:14]>> The Chinese model Deepseek was apparently developed at a fraction of the cost of the leading American models, but it performed about as well. Its weights were open source. Anyone could run it on their own infrastructure, free from American tech dependence. It hit number one in the app store within a week. Ever since Chachi PT, Europeans have been asking whether they could still catch up to the Americans in AI.

[00:00:37]>> The skeptics say we will repeat the main mistakes of the past and another generation of European talents will be forced to leave.

[00:00:45]>> But Deepseek is evidence that cutting edge AI can be trained without the resources of the US tech giants. A Sputnik moment where American superiority could be questioned. latest R1 model is as sophisticated as that of Anthropic's latest model or of OpenAI's latest model, but they've done it at a fraction of the cost.

[00:01:04]>> Pundits declare it's Europe's moment to enter the race.

[00:01:07]>> We will spare no effort to make Europe an AI continent, >> but there are voices of caution.

[00:01:13]>> Said, I think a lot of the claims are exaggerated and a little bit misleading.

[00:01:17]>> Deepseek has smart researchers and move fast on the new reasoning paradigm.

[00:01:21]Efficiency gains only go so far.

[00:01:25]What do you think?

[00:01:27]>> American export restrictions on AI chips will soon constrain its rise. Silicon Valley is largely unfased. The hyperscalers double down on building data centers. But in Europe, >> I welcome the European AI champions initiative. We aim to mobilize a total of€200 billion for AI investment in Europe. Deepseek leveling the playing field gives the illusion that they can take a bet and keep up.

[00:01:56]>> The European AI strategy and it will be a very important occasion. But this strategy will be a unique opportunity for Europe to accelerate.

[00:02:10]They were wrong. On the 12th of June 2026, the US federal government told Anthropic to block every foreign national from using its most powerful models. With the flick of a switch, an entire continent was pulled away from the leading frontier. A scenario had predicted exactly this. It just didn't expect it until 2029. This is a story about Europe in 2031. A scenario about what it means for Europe to lose on AI.

[00:02:36]It was written by a group of researchers who have spent their careers where frontier AI meets European policy. It's modeled on an American forecast called AI 2027, a future of radical AI takeoff.

[00:02:48]But what that takeoff means for Europe and its institutions. And the future it imagines doesn't look too hot. Some of this won't play out. Some of the claims will make you angry. Some you'll find ridiculous.

[00:03:01]But that's important. Talking about what that future might look like, no matter how much you disagree with these projections in particular are, in my view, a step in the right direction.

[00:03:09]That's why I'm making this video. I want to tell a story about what Europe might look like in 2031.

[00:03:16]February 2025, plug baby plug >> on nuclear energy, which means that more than 75% of our electricity is produced by this nuclear energy.

[00:03:29]>> 3 weeks after Deep Seeks R1, McLolin hosts the AI action summit in Paris. In 2025, the geopolitical environment has hardened.

[00:03:38]>> It's very important in this world where I have a good friend in the other part of the ocean saying, "Drill, baby, drill. Here there is no need to drill.

[00:03:49]It's just plug, baby plug.

[00:03:51]>> Europe is now focused on competitiveness, not safety. Catchup is possible. Europe signals a clear attempt to lead in AI. The continent may have been off to a slow start, but it's now waking up. European Commission President Ursul on the line announces a 200 billion euro invest AI fund, including a 20 billion euro AI gigafactories initiative to build five giant comput clusters on European soil. AI gigafactories, very large data and computing infrastructure for very large models.

[00:04:22]>> The British infrastructure provider Fluid Stack announces an even bigger 1 gawatt data center near Paris. Mistal is hailed as Europe's AI champion, but the energy is short-lived. Vice President JD Vance gives a historic anti-Europe speech in one of the summit's final sessions. US innovators of all sizes already know what it's like to deal with ownorous international rules.

[00:04:44]>> Later, he doubles down at the Munich Security Conference.

[00:04:46]>> Russia, it's not China. It's not any other external actor. And what I worry about is the threat from within the retreat of Europe from some of its most fundamental values.

[00:04:59]>> EU policy makers working on AI worry Europe must derisk from the US and build its own AI. Sovereignty, whatever that means, is the goal. But things would have been much simpler if they'd all just switched to UDu. Udu is a European company that makes an all-in-one business software with more than 45 integrated apps. It brings every app your organization could possibly need into a single platform like CRM, sales, invoicing, accounting, project management, inventory, website building, HR. Because they're all integrated, your sales, for example, links straight to your accounting and your inventory. This means it adapts and builds as your organization grows. and it can adjust itself to your country automatically with taxes, currencies, local regulations all handled. Whenever you start a business, there's a lot of regrets to have. But for me, one of the biggest is not grounding all the administration in one single source of truth. UDU is great at that. The first app is free for life with unlimited hosting and support included. And if you want the full suite, it starts at $19.90 a month. Link in the description.

[00:06:02]August 2025 is AI hitting a wall. In the US, the AI fixation only intensifies.

[00:06:09]Reasoning models like OpenAI's 03 are the new rage. The talent races in full speed. Metapoaches researchers for football star salaries. Engineers sleep intense at the office. The Trump administration publishes its AI action plan calling for American AI dominance.

[00:06:25]But in Europe, people roll their eyes.

[00:06:28]Americans are Americans. Large claims by boisterous tech CEOs doesn't help the illusion that this is being made bigger than it actually is. Some are calling AI a bubble or saying the tech is fake. The narrative hardens when Chad GPT5 arrives in August. Despite the hype when it lands, the reaction is disappointment.

[00:06:46]GPT5 still hallucinates. It makes silly mistakes. Its agentic capabilities are limited. AI skeptics take victory laps.

[00:06:55]But other experts are doubtful. 03 is only months old. GPT5 raised the capability floor rather than the ceiling. More powerful models are accessible to free users. But in Europe, they are talking to deaf ears. Even in America, some investors are getting cold feet. But the money keeps going.

[00:07:13]>> So interesting that an IPO we're all worrying about on the private market.

[00:07:16]>> Open AAI raises $40 billion.

[00:07:18]>> Committing to OpenAI at this time.

[00:07:20]>> Yes, there is still a lot of craziness happening on the private markets when it comes to AI. Still a lot of optimism.

[00:07:25]Nvidia signs a letter of intent to invest hundred billion in OpenAI. Oracle and OpenAI strike a $300 billion comput deal. In Europe, these circular looking arrangements draw comparisons to past financial crisis. In September, MSL raises 1.7 billion, 1.3 billion from the Dutch semiconductor giant ASML. The round is 20 times smaller than OpenAIs.

[00:07:48]But amid the bubble talk, few Europeans worry that Mistl is raising too little.

[00:07:53]Meanwhile, the Gigafactories initiative is slowly grinding. Amendments show that industry will have to fund more than 2/3 of the Lion's 20 billion. Media outlets claim the 200 billion euro invest AI initiative was largely recycled public funds repackaged with hopes of massive private industry AI investment.

[00:08:10]>> But the urgency of early 2025 has faded.

[00:08:13]>> If AI is overhyped anyway, there is no need to worry.

[00:08:21]November 2025, two worlds.

[00:08:24]By November 2025, Anthropic releases Claude 4.5 Opus. AI agents have actually started to work. Instead of computer use agents and assistants, the Frontier Labs have doubled down on agents that excel at coding. The market is lucrative and proficiency in software engineering also helps the lab speed up their own research and development. Cloud code goes viral. People learn that it can do much more than code through bash commands and HTML. Coding agents can also send emails, organize documents, and navigate the web. Developers burn through thousands of dollars in tokens a month, refusing to go to bed without first queuing up overnight tasks for multiple agents to coordinate on as they sleep. Anthropic's revenue skyrockets from $1 billion annual recurring revenue in December 2024 to $9 billion at year's end. It becomes one of the fastest growing companies in history. Meanwhile, Anthropic is already training Opus 4.6.

[00:09:16]Release cycles have compressed from 6 months to three. Top researchers share that most of their code is now written by agents like cloud code and companies start monitoring their AI deployments with other AI systems. European leaders are only partially aware. They lack trusted tech advisers and institutional policies often forbid the use of standalone AI tools like Claude or Chad GPT.

[00:09:37]>> March 2026, the Frontier AI initiative in Europe. Sovereignty talk is abound with little to show. Back in November, Germany and France announced the Frontier AI Initiative with backing from the European Commission. It's a plan to establish by Q1 2026, the world's best funded nonprofit AI research organization. But when Q1 ends, the only concrete thing the Frontier AI initiative has produced for the public is an expert form. The Promised Research Organization is delayed. The details need further fleshing out. The experts that Germany, France, and the Commission consult all point in widely different directions.

[00:10:12]Some confidently claim that LM are a dead end and that the initiative should try to leaprog the Americans through a different paradigm. Others argue that Europe should find its own niche, reliable AI, AI for science, industrial foundation models. The list is long and consensus is nowhere to be found.

[00:10:28]Resources are scarce, too. France has little physical maneuverability, and the commission's new budget only starts in 2028. Germany has money, but a long wish list in other domains, and it's just committed to building the largest conventional army in Europe.

[00:10:42]It's a tough spot. Partners already committed. Something needs to be built.

[00:10:46]But spinning up a public frontier AI lab turns out to be difficult. The initiative wants to attract worldclass talent. But it cannot offer anywhere near the compute and compensation packages of the American labs. Without proof that they can hire top-notch talent, none of the partners want to throw serious money at it. Meanwhile, in Silicon Valley, OpenAI raises $122 billion, putting to bed any remaining worries that the company can't pay its bills this year.

[00:11:13]April 2026, Claude Mythos. In April 2026, Anthropic announces Claude Mythos, the most capable model ever built. For the first time, the company does not release its new model publicly. Mythos has become so proficient at coding and security research that it has autonomously identified thousands of zeroday vulnerabilities across every major operating system and browser. Many are 1 to two decades old, sitting undetected in code bases reviewed by millions of developers. Anthropic determines it's not safe to release.

[00:11:44]Without many people noticing, Anthropic has supposedly become one of the most competent offensive cyber organizations on the planet. Spooked by their own pace, they launched Project Glasswing, a defensive coalition with AWS, Apple, Google, Microsoft, Nvidia, and Crowdstrike, giving partners exclusive access to Mythos to find and patch vulnerabilities, before adversaries can exploit them. The goal is to harden the world's critical software before open- source models reach the same offensive capabilities.

[00:12:12]The announcement lands in the middle of a bitter standoff between Anthropic and the US government. When Enthropic announces it wants to expand Mythos access to 70 additional organizations, the White House objects. It's not secure, they say. Besides, Anthropic's comput is finite. More partners means less capacity for US government workloads. A month earlier, the administration had wanted nothing to do with Anthropics AI.

[00:12:35]>> All government agencies to cut off ties with the AI company Anthropic after it refused to agree to the Pentagon's demands.

[00:12:42]>> Now, it wants the company's most powerful model kept largely for itself.

[00:12:46]For Europe, the implication is deeply uncomfortable. The most consequential defensive cyber security tool in the world is being rationed by a US administration that has just spent a month trying to break the company that built it. Most European governments and businesses are not on the access list.

[00:13:03]May 2026. Disillusionment. After 18 months, Deepseek has finally released its next model, but it barely registers.

[00:13:10]It's impressive given the company's size, but still 6 months behind the Frontier. Deepseek admits it lacks the compute to serve it widely, likely because of US export controls. The playing field has not leveled. Mal has raised another 815 million, around 150 times smaller than OpenAI's last.

[00:13:27]Looking for capital, it has turned to the US. Infrastructure tells a similarly grim story. Fluid Stack will no longer build its promised 1 gawatt data center in France, pivoting to American buildouts after relocating its headquarters to the US. EU Gigafactory applications have only just opened, a year after the announcement, and budget constraints mean the EU can likely fund only two of the five plan facilities, none realistically up and running before 2029.

[00:13:55]By May 2026, the largest AI supercomput in the US runs at 1.5 GW. Europe's largest sits at 75 megawatt, 20 times smaller. Even a gigafactory appearing overnight would be more than 10 times smaller than America's leading facility.

[00:14:12]By now, national security staff have briefed European leaders on mythos. In return, they've asked uncomfortable questions. Why don't European companies and governments have access? Do we have domestic fallback options? Wasn't AI supposed to hit a wall? But while leaders and CEOs are looped in, the mythos breakthrough is not reaching the masses. Cyber security is not sexy, and there's no product to try. Some people deny anthropics claims entirely. The gap between public perception and actual capabilities have never been higher.

[00:14:42]But things will change.

[00:14:44]>> Up to this point, all the events can be verified by historical records. We'll no longer single out individual AI companies. Prometheus stands for the leading American AI company, Lumos for the leading European company, and Phoenix for the leading Chinese one.

[00:14:58]In February, German Chancellor Felix Meritz had flown to Hango and tore Chinese EV and robotics factories. He returned chasened, convinced that German industry would not survive uninherited prestige. In August 2026, he makes a similar trip to San Francisco. He wants to witness the AI revolution from up close. He meets researchers towards labs and is shown under NDA capabilities that do not yet have names. When he returns to Berlin, an insider reports that Mertz has been in an unusually brooding philosophical mood. Meritt has come to believe that European elites have catastrophically misread the past 18 months. While Brussels debated whether AI would hit a wall, it had punched through several. While ministers weighed AI strategic's importance in white papers, it quietly swallowed software engineering and cyber security. While European capital celebrated their sovereign AI initiatives, dependence on American providers only deepened. Across the border in France, President McClon has just been briefed on worrying news.

[00:15:55]Lumus is preparing to stop pre-training its own frontier AI models. The feedback loops between data, capital, and talent have hardened. The compute gap has become a chasm. A pivot looks rational.

[00:16:08]The thought deeply disturbs Mcomem. The French startup has often been called Europe's last hope in AI. Over the following days, he and Merit hold a series of long encrypted calls. Both feel they have reached an inflection point. AI will keep getting more capable. It will rearrange labor markets, security architectures, and the balance of power between continents.

[00:16:26]Europe is dependent and unprepared. Both men in private are drawn to drastic measures. Interventions actually proportionate to the moment. Both know what that would cost them politically.

[00:16:35]Their political advisers plead for restraint. Their technical adviserss plead for the opposite. They are scheduled one week out to deliver a joint address at the FrancoGerman Digital Sovereignty Summit in Strawburg.

[00:16:47]By the night before, they still have not decided. In the morning, Meritz Mcome will walk on stage together.

[00:16:55]August 2026, a positive vision.

[00:16:59]The speech Meritz and McL deliver is the one their political advisers wrote. It draws a natural conclusion. If the United States can no longer be trusted, Europe must build its own frontier AI.

[00:17:10]Merit speaks of resolve. Lum speaks of destiny. Neither mentions what they talked about in private. Over the following weeks, France, Germany, and the European Commission announced a cascade of interventions. The media strategy is volume. Overwhelm the euroskeepics with motion. The messaging is hope. Imagine a world where we run our own European more trustworthy models. Billions of additional euros are committed to the Frontier AI initiative.

[00:17:35]Subsidized compute and AI and science.

[00:17:38]Several American AI companies are charged under the AI act and the digital services act. The measure that actually makes the headlines is a new EUI commitment. By 2035, member states must be 100% digitally sovereign with civil services no longer relying on American AI companies or cloud providers. It also convinces Lumos to keep investing in pre-training with a guaranteed pool of future customers. Investors line up again. Most measures are wellreceived.

[00:18:04]They show resolve. They align with the popular instinct to derisk from the US.

[00:18:09]But some experts are doubtful. The investments are too little too late. The adoption measures fail to address deeper bottlenecks. The fragmented single market strike labor protections that need to be paired with astronomically high salaries for talent. Sector specific rules that effectively ban AI across much of healthcare and legal services. But most European elites are done with that kind of negativity.

[00:18:30]Europe needs a positive vision and private merits and makon tell each other that they did what was possible.

[00:18:38]March 2028 taking stock. More than 18 months have passed in Strawborg. AI progress has been relentless. Agents now reliably operate spreadsheets, enterprise software design tools, and procurement systems. The leading labs have doubled down in automating AI R&D itself, racing to be the first to build a fully automated AI researcher.

[00:18:57]Prometheus has gone public, its market cap surging to 5 trillion. Nobody is calling AI overhyped anymore. Through a heroic effort, Lumis has held its position roughly 12 months behind the frontier, raising billions from new investors, keeping its talent from moving to the US. But its relative position hides that the absolute distance in capabilities has grown. This is no longer an abstract problem. Until recently, a few hold out Chinese AI companies still open source their models. As a result, cyber capabilities meaningfully beyond Mythos are now widely available, including to anyone who wants to misuse them. Without active AI assisted defense, such systems breach most software. People are furious at the labs. They unleash the disease and now sell the cure. Prometheus's AI is noticeably better at cyber defense than Lumos's.

[00:19:48]Some governments are only allowed to use Lumos and are now regularly hit by ransomware attacks built on better models.

[00:19:56]April 2028 cautious optimism.

[00:19:59]And yet politicians are hopeful. Europe got lucky on cyber. The CCP has banned Phoenix and its peers from openly releasing systems that meaningfully advance offensive capability. The worst case scenario may have been averted by Beijing rather than Brussels. Europe's adoption pilots, after a rocky start, are returning positive results. Doctors are becoming more productive with AI.

[00:20:19]Teachers are giving more personalized guidance to students.

[00:20:23]The 2028 EU budget has unlocked substantial funding for AI and science.

[00:20:27]With heavy investment in medicine, material science, and clean energy, areas where Europe could still possibly win. The job doomsayers, meanwhile, have not been proven right yet. Even in sectors of high adoption, AI is not behaving like a labor destroying technology. Quite the opposite. Workers who know how to use it have become marketkedly more valuable. Consultants, software engineers, lawyers, analysts, and designers who can orchestrate AI systems see their output and wages rise sharply. The first macroeconomic effect of AI is not mass unemployment, but a productivity boom. The public, however, is less convinced. Polls across the EU show majorities expecting AI to disrupt their work, their children's prospects, and the information environment more than it improves any of them. Protests against AI have begun to sprung up in European city centers.

[00:21:15]May 2028, the compute crunch. By May 2028, the whole world is screaming for more AI compute. Every previous bottleneck in the chip supply chain found a creative workaround. Now, for the first time, a bottleneck is emerging with no obvious solution. Assumel, the Dutch semiconductor giant, is the only company that can manufacture the EUV machines TSMC uses to fab AI chips.

[00:21:37]Output has risen from 60 to 85 machines a year, a real improvement, but wholly insufficient against demand growing three times annually.

[00:21:46]The US government does not like that the company constraining its AI ambitions sit outside its control. It also does not like that ASML still exports older DUV machines to China where they used to fab the country's AI chips. USChina tensions have risen sharply as the AI race is heated up. Washington worries China can outbuild it once it fully indigenizes its semiconductor ecosystem.

[00:22:07]It is already better at manufacturing and robotics and its energy supply is skyrocketing. Beijing in turns worries the US will use its AI advantage to force geopolitical concessions. Now Washington pressures the Netherlands to halt all DUV exports. This is highly escalatory. The machines do not only make Chinese AI chips, but also chips for smartphones, cars, and laptops sold worldwide. The Dutch resist. They can see the reasoning, but do not like being bossed around, let alone dragged into a global power conflict. The commission backs them, but many member states fear American retaliation. Seeing fragmented European support, the Dutch turned to Japan and South Korea, only to find the US has gotten to them first. When Washington threatens to cut off Simer, the US-based laser supplier, Osmo cannot operate without the Netherlands caves.

[00:22:53]August 2028. Bottlenecks.

[00:22:57]A new line of AI systems has arrived that no longer thinks in English.

[00:23:00]Instead of writing thoughts on a digital scratch pad, it cycles through highdimensional vectors, long lists of numbers humans cannot interpret.

[00:23:08]People in the know are alarmed. In Brussels, experts urge the AI office to act, but the commission is understaffed.

[00:23:15]To most observers, the big news is capability. Freed from compressing complex thoughts into English, the systems think more quickly and more deeply. In a matter of months, this produces a significant jump in intelligence. Models that cannot reliably complete multi-day projects now can. In the US, employers on the fence about cutting jobs pull the trigger.

[00:23:35]Entry-level vacancies dry up.

[00:23:37]Unemployment remains low in absolute terms, but is rising sharply. In Europe, the picture is grim in its own way. The same job pressures are building but without a company in growth to absorb them. US GDP grew at an annualized 4.7% last quarter. The EU posted 2.9. The productivity gap is commonly attributed to differences in AI diffusion. On the surface, adoption rates look similar.

[00:23:58]Dig deeper and the picture is more worrying. Markets restructure around AI significantly slower in Europe. America produces more disruptive startups and its companies adopt AI faster because the environment rewards it. Less regulatory friction. a larger internal market, more capital, weaker labor protections, and a management culture more comfortable with radical automation. Europe has access to the same models, but it cannot extract the same economic gains. The dynamic plays out vividly in Europe's labor market, where pretend work has become common.

[00:24:29]Strong labor protections means that employers cannot let go of staff whose jobs the AI systems can automate.

[00:24:35]Employees log in, attend the meetings, send the emails, and let their AI tools do most of the actual output in a fraction of the time. The arrangement has its charms. People enjoy the extra hours with their families, the long lunches, the afternoon walks, but the costs compound. Salaries are paid for output that no longer requires them.

[00:24:52]Firms carry headcounts they cannot justify.

[00:24:56]Government spending will need to rise to cushion the labor disruption when unemployment inevitably catches up to reality. But the same transition is eroding the tax base. European firms are losing share to American AI first ones and a larger share of GDP is flown to capital tax at lower rates and often routed through tax havens. Meanwhile, global interest rates are rising amid the capital hungry AI boom. In countries like France with state debt at 120% of GDP, an ever larger share of the government revenue is being spent just to service it. The only real remedy is growth. But Europe is not growing nearly as fast as the US.

[00:25:32]>> November 2028. The people don't want AI.

[00:25:36]The world holds its breath as the 2028 US election reaches its conclusion. AI has been a defining issue of the campaign. During the primaries, populist factions on both sides ran on strong anti-AI platforms, data center moratoriums, worker protection against AI automation, bans on AI and education.

[00:25:54]The general election offered more centrist candidates with stronger ties to the AI industry. But the underlying sentiment has not gone away.

[00:26:01]Increasingly, people take to the streets. Fractures show along two lines.

[00:26:05]The classic Democrat versus Republicans, but now also elites who typically welcome AI versus a middle class that finds it scary, immoral, and dehumanizing. Increasingly, anti-AI movements are turning violent. There have already been multiple attempts on the lives of AI leaders. Just last week, a data center was set on fire. Much of the American anti-A sentiment spreads to Europe. People are angry and they want the EU to come down harder on American tech companies. They demand stronger social safety nets, but most of the value of the AI boom that could fund them is acrewing to the US. Meanwhile, Europe's AI companies are falling further behind. The gap between Lumos and Prometheus is widened further.

[00:26:43]American labs are now powered by swarms of internal AI agents, writing most of their own code. They're making algorithmic progress at more than twice their human only speed. Lumos has a fraction of the processing power. Every month, the gap grows wider. Europe's public efforts are delivering public goods. The Frontier AI initiative has made impressive progress on interpretability. That work benefits AI reliability everywhere, but its world model program has been hijacked. As soon as the results started picking up, Prometheus took notice. It spun up its own team, and it poached four of the initiative's best researchers in exchange for astronomical compute budgets. The researchers wanted to stay.

[00:27:20]They wanted to believe in the European project. But at some point, belief is not enough.

[00:27:25]The public can see that the EU's grand strategy is failing. The investments haven't let champions like Lumos catch up. The regulatory actions under the DSA haven't leveled the playing field. To admit failure now would mean admitting it spent 2 years and tens of billions of euros on a dead end. So, it doubles down. The European Commission announces a new fund, 20 billion for AI sovereignty. It's clearly a long shot.

[00:27:49]The same institutions that already failed to turn money into capability are being asked to try again with a harder target and less time.

[00:27:58]January 2029 dreams of a robot economy.

[00:28:01]China's AI strategy isn't so different from the EU's, except in one crucial way. It's working. Like Brussels, Beijing backs its most promising companies. It protects the industry with subsidies and procurement mandates, and it pushes hard for adoption.

[00:28:16]When China's compute was split between rival labs, the Communist Party simply ordered them to pull it. The European Commission has no such power. China's talent pool is deeper. Its energy is cheap and abundant, and its frontier labs stay within a year of the US. And China doesn't believe its strategy depends on leading the cognitive frontier. The government always wanted AI to boost the physical economy first.

[00:28:37]And in robotics manufacturing, it holds a huge advantage. Massive state subsidies have pushed humanoid production above a million units a year.

[00:28:45]In cities like Shenzen, humanoids clean the streets. Quadripeds deliver the packages. It's ordinary now. Beijing's bed is simple. Stay close to Silicon Valley and AI and let its industrial advantage pay off as more robotics capabilities arrive. The bet still looks viable. And in the US, concern is mounting. American politicians have started calling the eye race of China the second cold war. Tensions rose further when the US forced ASML to halt exports to China. Since then, both Beijing and Washington have tightened security around their AI development. In the US, researchers are screened by the NSA. They need clearance just to touch frontier model weights, and the government regularly accuses China of stealing algorithmic secrets. China, for its part, has semi-ationalized Phoenix and thrown its full weight behind it.

[00:29:30]April 2029, access all areas.

[00:29:34]AI demand is soaring. Compute supply can't keep up. The labs start rationing access to their frontier models and raising prices. Businesses are screaming for tokens. American firms run 70% of the world's AI compute and they sell it everywhere. So, American infrastructure is making foreign business more productive, foreign militaries more capable, and foreign labs more competitive. The American National Security Review has been formalized. It no longer pretends to be voluntary.

[00:30:01]Access to the most capable models is throttled by default. Domestic government agencies get them first.

[00:30:06]Allied governments next. Adversary nations not at all. But access is one thing. The comput shortage means inference itself is now at a premium.

[00:30:15]And by April, Washington has had enough.

[00:30:17]It starts rationing not just access but usage. Even for the countries inside the 10. It's called the frontier inference services rule. A country-based licensing regime. Tier one is for close allies, the five eyes nations plus Japan, South Korea, Taiwan, and the Netherlands. They get unrestricted access and light touch reporting. Tier 3 is for hostile countries, Iran, Russia, China. By default, they get nothing. Most of Europe lands in tier 2 along with about 100 other countries in between. And tier 2 has a cap. No more than 25% of any provider's frontier inference can go to tier 2 customers. Europe depends on American AI. The US has no such dependence on Europe. In Brussels, an emergency European Council meeting is convened. French and German ministers fly to Washington to demand tier 1 status. They're told tier 2 better reflects quote the current state of the bilateral relationship.

[00:31:10]May 2029 tilt. A week after the cap, the phones and European capitals won't stop ringing. Panicking businesses trying to put off the inevitable. The European companies smart enough to delay or drop their sovereign AI policies spent years building everything around frontier agents. Now those agents are being pulled out from under them. The European alternatives are almost 2 years behind.

[00:31:31]The Chinese ones aren't really an option for anyone with a compliance department.

[00:31:35]Demand for Lumis' models has shot far beyond what the company can supply. The Gigafactories are finally under construction, but they won't come online until next year, and even then, they'll be a fraction of what's needed to close the gap. European leaders decide something drastic has to be done. For the first time ever, the commission opens a formal case under the anti-corion instrument, but the assessment reaches an uncomfortable conclusion. The obvious retaliation is self-defeating. Tariffs on US cloud and digital services would just raise the price of the inference European businesses are already scrambling for.

[00:32:06]So the proposed counter measures reached for subtler tools. Suspending the intellectual property protections US labs hold inside the single market and screening or blocking American takeovers of European firms. It's an unattractive option, one that invites a response Europe may not be able to afford. When it comes to a vote, it falls short of the majority needed. On the record, a senior commission official says the result reflects, quote, differing national assessments of the strategic environment. Off the record, a French official puts it differently. They were too scared of Washington to fire the weapon they spent a decade building.

[00:32:40]February 2030, labor shock. Europe's pretend work honeymoon doesn't last. The one where people took up gardening while the agents did their jobs, and their firms weren't allowed to let them go.

[00:32:51]European firms are stuck paying full human staffs. They can't compete with leaner American rivals. The shock hits the most AI exposed industries first.

[00:32:59]And the AI systems just keep getting better. Continual learning was often called the last hurdle to fully automating knowledge work. Let Prometheus latest model sit inside a consultancy for 6 weeks and it starts to write the way that firm writes. It learns who defirms to whom, which clients take bad news badly, which senior names are worth leading on. The implementation is imperfect, but the failures are rare and worth the price.

[00:33:21]The jobs once thought safe, protected by judgment or by knowledge too tacit to write down are now exposed. The graduate jobs market is the worst in living memory. Street protests become routine.

[00:33:32]Some demand labor protections. Some demand bans on American AI. Some demand bans on all AI. All of them demand that someone somewhere do something.

[00:33:42]June 2030, an offer you can't refuse.

[00:33:45]China is leading on robotics, but now Prometheus is going all in on it too. It announces hundreds of billions in spending on industrial data and on manufacturing capacity to build robots at something like China's scale. Its CEO lays out the logic. America still leads AI software. With his new world model team, he's cracked the last software problem standing in the way of generalpurpose robots. Soon, robots will build the factories that build the robots, just as the AIs already write the code that improves AIS. But there's catching up to do. A new robot factory takes 2 years to build in the US. In China, it takes 7 months. Even America's rapid growth is now constrained by infrastructure, buy in quality, and by public opinion. So the CEO decides the fast path isn't to build, it's to buy.

[00:34:29]To retrofit factories that already exist, backed by friendly investment funds, he goes shopping for industrial firms with floor space. Space he can turn over to building wheeled robots, quadripeds, and humanoids. Europe's car manufacturers are high on the list.

[00:34:42]After years of Chinese competition, Germany's biggest car maker is close to bankruptcy. For a buyer who wants high-tech manufacturing to build tens of millions of robots a year, it's perfect.

[00:34:52]But Berlin disagrees. It blocks the sale on national security grounds. But insiders know it's really about pride.

[00:34:58]Prometheus doesn't back down. The CEO calls the US president. Within 72 hours, the White House hits European car imports with sky-high tariffs.

[00:35:06]Officially, it has nothing to do with the car maker. 3 weeks later, a sale is announced, dressed up as a partnership.

[00:35:12]The German state takes a 20% stake. The old board keeps nominal control behind the scenes. Prometheus is in charge. It holds operational control, the licensing rights to the manufacturing platform and first refusal on any future fund raise.

[00:35:26]The pattern repeats again and again over the next few months. One by one, Europe's high-tech manufacturers are bought up and restructured by American ventures. Robotics, aerospace, specialist tool makers. The official line is that it protects jobs and keeps vital plants on European soil. The reality is that Europe has no counter offer.

[00:35:46]August 2030 model collapse.

[00:35:49]The United States isn't trying to destroy Europe. It's trying to beat China in a race it believes is existential. In Paris, the finance minister presents a budget to the National Assembly. Nobody in the room believes it, including him. Three numbers simply won't reconcile. Welfare spending near pandemic levels, corporate tax down 9%, and a tenth of the whole budget going just to service the debt.

[00:36:11]The only way to make it add up is to assume growth nobody believes in and everyone sees through it instantly. The cost of French borrowing pulls sharply away from Germany's to the widest gap since the euro was created. The political consequences soon arrive. The protests build through the spring until they turn violent. The screens in the baramont show riots in Paris and Rome.

[00:36:30]Young people mostly united by little except the sense that the system has failed them.

[00:36:36]Across the Atlantic, the picture is different. Facing its own anti-a backlash, the US has rolled out job guarantees and direct cash transfers at a scale Europe can't even contemplate.

[00:36:46]The protests there haven't stopped, but the government has bought itself time.

[00:36:50]The contrast isn't lost on the European public or on European finance ministers.

[00:36:54]They know the American response is paid for by AIdriven growth their own economies can't produce. The ECB intervenes and intervenes again. Then it runs out of credible instruments. By late February, depositors in Italian and Greek banks are moving money north faster than the ACB can offset it.

[00:37:12]March 2031, between giants.

[00:37:16]By early 2031, power sits in two places.

[00:37:19]American labs hold the cognitive frontier. China still holds the physical one. Prometheus alone is worth more than every listed European company combined.

[00:37:28]The three biggest American AI firms each spend more on compute than the entire EU spends on defense. The leading Chinese humanoid maker ships more units in a month than Europe ships in a year. The rising tension across the Pacific has brought Taiwan to the center of everything. TSMC's FABS, where the world's most advanced AI chips are made, are mostly on that island. The gap between US and Chinese AI has grown over the past year. Taiwan's importance has grown with it. Both sides have tested AI run weapons platforms in public and more powerful ones in private. In Washington, the labs and the defense department are so entangled, the distinction barely means anything. In Beijing, integration is more formal. Analysts now use the word war. We have no qualifier in front of it. Europe isn't a disastrous state.

[00:38:12]Polarization is everywhere. In the country hit hardest by the AI shock, growth has stopped. In the rest, it's all but detached from how people actually live. Mortgage defaults have surged in the member states that rely on variable rate loans. But Europe has one last card to play. After 5 years of failing to build a frontier AI sector, it still owns the one bottleneck the whole race runs through. ASML is still the only company on Earth that can build the EUV machines used to print cutting edge chips. Without those machines, the US couldn't keep extending its lead.

[00:38:40]With them, China would probably have caught up long ago. Beijing has made real progress on its own DUVI machines.

[00:38:47]Mass production is expected within a year, but it's too little too late. A year now feels like a lifetime. The US is pushing ahead every day. The Chinese government is deeply worried that super intelligent AI is close, and nobody is sure what it could do. Some advisers fear the US could use advanced enough AI to neutralize China's nuclear second strike. Others fear such systems could be made persuasive enough to destabilize the party itself. China's lead in robotics is real, but it can't ease those fears. So, Beijing leans harder on the strategy it's been running for 2 years. The loans to Southern Europe grows bigger and the terms more generous. The information campaigns intensify and European leaders start getting signals about what a closer relationship might look like. Privileged market access, joint robotics production, a seat at a table Washington has mostly kept them away from. For the first time, ASML and its EUV technology are named explicitly. 5 years of being treated as a vassel by Washington has left its mark. And in several capitals, the alternative is for the first time being taken seriously. The White House decides it needs direct control of the company. Now, while it still can, the call goes to three countries that can stop it. The Netherlands, where ASML is based, and Germany and France, because no Dutch government could sign without them. The US vice president makes the offer in a 40-minute call on a secure line. ASML would be folded into a joint Dutch American holding company. A shared board would run it, and Washington would hold the controlling vote over what it makes, who it sells to, and what technology moves. In return, the US offers a capital ejection Europe can't match, enough to build a string of new production sites on American soil. And more striking still, it promises direct cash payments to European citizens. The three leaders decide not to tell their fellow EU heads of government. They don't find the offer acceptable, and they worry, the others would pressure them into taking it. When they politely decline, White House makes the offer public and adds a stick to the carrot.

[00:40:36]If the EU doesn't sign, the whole region drops its tier three and loses all access to American AI, present and future. And the US knows it holds all the cards. Several European capitals reach out to China hoping for a counter offer. Something that gives them room to say no to Washington. It's not what they were hoping for. Beijing has decided the charm offensive isn't enough. It switches registers, too. If the Dutch sign with Washington, rare earth export terms will be reviewed. Robot export licenses will be reassessed. And the deadline is shorter than the American one. Europe now has three options. Sign with Washington and Europe hands over its only leverage. It becomes an American protectorate in all but name.

[00:41:14]And if China follows through and cuts off exports, what's left of European manufacturing dies. Align with Beijing and maybe the South stops drowning, but it would also be handing China the keys to its future. Tier 3 would just be the opening move. Sign with neither and Europe gets nothing. Instead, it takes the full displeasure of both great powers at once, and it would likely tear the already strained union apart.

[00:41:37]Europe's slides into irrelevance was not inevitable, even as late as 2026. It could have still have changed course if it had found the courage and the will to act drastically.

[00:41:47]What you just watched is my version of this story. It's compressed. I cut a lot. The original goes much further.

[00:41:53]It's told through characters, a young policy worker in Brussels, and a founder who left Europe for Silicon Valley. And it walks through all of this from the inside yearbyear. There's dozens of details, arguments, and turns I I skip past. So, if any of this landed, if it worried you or annoyed you or made you think, and if you want to think about what are the right decisions, please go ahead and read the full thing. The link is in the description. Yeah, let me know what you think.