The Global AI Bubble

A Cross-National Suicide Pact

Oct 14, 2025

Prologue

I’ve been sitting on this analysis for weeks. The implications are sobering enough that I kept second-guessing whether to publish it. But the October 2025 data shows the situation isn’t stabilizing—it’s accelerating. Silence isn’t neutral anymore.

For the past two years, every major economy has been engaged in an arms race to build the physical substrate of artificial intelligence: data centers, power grids, and semiconductor supply chains. The spending now exceeds the Apollo Program in annualized terms. Each region believes it’s building strategic advantage. In truth, they’re co-constructing a synchronized crash.

Framework: A Mispricing of Time

The global AI buildout is not a technology bet—it’s a liquidity event searching for justification. The market is not valuing future revenue from AI applications; it’s valuing the act of building infrastructure for those applications. This distinction is critical.

The fundamental mispricing is temporal. AI is likely a General-Purpose Technology—comparable to electricity or the internet—that will take decades to realize productivity gains across the economy. Historically, GPTs require fundamental reorganization of work, institutions, and capital allocation before their transformative potential materializes. Yet the current capital formation operates on quarterly earnings timelines, pricing AI as if it possesses electricity’s long-term impact but will be adopted with a viral app’s speed.

This contradiction is unsustainable. The builders of the dot-com era’s fiber optic networks were right about the internet’s transformative power. They went bankrupt before that future arrived. Being correct about the destination doesn’t prevent insolvency during the journey.

The current bubble operates at a scale that makes previous financial crises look modest. The 50:1 ratio of infrastructure spending to consumer revenue isn’t a lead-time gap waiting to close—it’s a systemic mispricing of when AI’s economic value will actually materialize versus when the financial system needs returns.

The wrong clock is running the world.

What follows is the evidence for how three economic blocs, using three different financing models, have locked themselves into the same temporal error—and why the coordination makes the eventual correction more catastrophic.

TL;DR:

50:1 infrastructure spend vs consumer revenue = a time mismatch, not a lead time
US: Reflexive capital manufactures demand; GDP props up on capex itself
EU: Pensions become exit liquidity; double exposure (US equities + local buildout)
China: Control group proves overbuild even without Wall Street—idle megawatts are the new dark fiber

Opening: The Economist’s Prescription

On October 2, 2025, The Economist published an analysis of Europe’s innovation problem. The diagnosis: European labor regulations impose costs that discourage investment in disruptive technologies. An American firm shedding workers incurs costs equivalent to seven months of wages per employee. In Germany the figure reaches 31 months. In France, 38 months.

The prescription is clear: Europe needs American-style labor “flexibility” to foster innovation. Strip away the protections. Make firing cheaper. Become more dynamic.

The innovation The Economist is praising isn’t technological. It’s narrative: the art of turning systemic risk into a moral virtue.

Part I: What “Like America” Actually Means

To understand what Europe is being advised to emulate, we need to examine what American “innovation” currently looks like at the highest levels of AI infrastructure investment.

1A: Nvidia’s Circular Capital Machine

On October 6, 2025, OpenAI announced a multi-year chip supply agreement with AMD—and granted OpenAI the option to acquire up to 10% of AMD.

For AMD to give away equity options just to secure OpenAI as a customer isn’t strategy. It’s triage—a bid to break into a market dominated by Nvidia’s capture of major AI players.

That capture is Nvidia. According to Nvidia’s Q2 FY26 10-Q filing, two unnamed customers accounted for 39% of the company’s total revenue in that quarter. Not 39% of data center revenue—39% of everything Nvidia does. That’s not market dominance. That’s existential dependency flowing in both directions.

The structure becomes clearer when examining Nvidia’s September 2025 announcement of a $100 billion “investment partnership” with OpenAI. The investment is released progressively as each gigawatt of AI data center capacity is deployed—$10 billion per gigawatt. Each gigawatt costs roughly $50–60 billion total, including $35–40 billion in Nvidia hardware.

So Nvidia “invests” $100 billion in OpenAI, which then flows directly back to Nvidia as hardware purchases. This gets recorded as “revenue” which justifies the investment, which funds more hardware purchases. It’s not fraud—each transaction is legitimate, the accounting follows GAAP. But the aggregate creates something novel: revenue-recognized reflexivity. Accounting recognizes revenue; it can’t label reflexivity.

Nvidia’s total debt as of Q2 FY26 stands at approximately $10–11 billion, with the company increasingly acting as financier via gigawatt-linked investments and customer advances rather than pure chip vendor.

This isn’t investment. It’s choreography.

1B: Oracle’s Financial Engineering

Oracle offers a complementary case study. On September 9, 2025, Oracle’s stock surged 15% after announcing $455 billion in “remaining performance obligations”—contracted but not yet delivered revenue. The company described a multi-cloud future where Oracle handles AI infrastructure buildouts for the hyperscalers.

The notable part: Microsoft, Google, and Amazon—companies with vastly deeper pockets and better AI expertise—have signaled reluctance to internalize comparable off-balance-sheet DC risk at this pace. Oracle taking on what better-capitalized companies declined was treated as pure upside.

Then came the financing structure. Bloomberg reported that banks arranged roughly $38 billion in debt packages for Oracle-linked data centers. Blue Owl and other private credit firms started structuring multibillion-dollar joint ventures for facilities in places like Abilene, Texas. Oracle wasn’t just building infrastructure—they were securitizing it, packaging long-term commitments into financial products that could be sold to institutional investors.

The structure echoes 2008’s mortgage-backed securities:

Illiquid AI assets with uncertain demand
Recast as “stable” RPO-backed paper
Risk distributed, reward concentrated

Oracle isn’t committing fraud. They’re responding rationally to distorted capital costs. The critical questions center on contract quality—term length, termination rights, price escalation clauses, counterparty creditworthiness—details rarely disclosed when these contracts are packaged into securities and sold to institutional investors.

The Scale

Using a Wicksellian cost-of-capital framework, an analyst from MacroStrategy Partnership calculated that this AI bubble is 17 times larger than the dot-com bubble and four times larger than the 2008 subprime mortgage crisis. The methodology: capital is efficiently allocated when corporate borrowing costs are about two percentage points above nominal GDP growth. When borrowing costs fall below that threshold—as they did during a decade of Federal Reserve quantitative easing—you get massive capital misallocation.

Tech companies are projected to spend about $400 billion in 2025 on infrastructure to train and operate AI models. That’s more than the Apollo program’s inflation-adjusted cost—except not over a decade, but annually. Total US AI capital expenditures are projected to exceed $500 billion in 2026 and 2027.

American consumers spend approximately $10–15 billion per year on consumer-facing AI services by most estimates—subscriptions, add-ons, and direct AI product purchases.

The 50:1 Mismatch: 2025 global AI infrastructure capex (US ~$400B + China ~$98B + EU/RoW ~$100B = ~$600B) versus direct consumer AI revenue (~$10–15B in subscriptions, add-ons, and AI product purchases). This calculation excludes enterprise SaaS pass-throughs to avoid double-counting cloud infrastructure spending. The ratio isn’t a lead-time gap; it’s a pricing error on when returns will materialize.

This is what The Economist wants Europe to emulate. Not technology—term sheets. The ability to rapidly hire and fire workers to chase these capital flows. The “flexibility” to scale operations up and down as circular financing schemes expand and contract.

The American model isn’t serving a market—it’s manufacturing its own demand through financial structures that blur the line between investment and revenue. When you’re simultaneously the supplier, the investor, and the customer, you haven’t built an ecosystem. You’ve built a closed loop that functions until any component fails.

Part II: Europe’s Exposure - 2008 With a Sequel

A: How Europe Re-Financialized Through Pensions

Europe didn’t just watch American financial innovation from a distance. They bought in—heavily.

Dutch pension funds provide a representative example: detailed analysis of their holdings shows significant US corporate exposure, though total pension assets remain diversified across regions. The broader European trend is clear in the aggregate statistics.

Over half of European pension funds’ equity investments now flow to US companies, with notable exposures to US corporate and government bonds as well. Between 2013 and 2023, the share of US equities in European pension fund portfolios rose from 23% to 39%. US debt holdings rose from 6% to 11% over the same period.

This acceleration was aided by policy. Auto-enrollment pension reforms across Europe channeled worker savings automatically into equity markets—precisely as those markets concentrated into US tech. Workers who never chose to speculate on American AI valuations found their retirement security systematically redirected there anyway.

Workers became the collateral pool.

This isn’t diversification. It’s concentration dressed as sophistication.

European investment funds have increased their allocation to AI-related companies specifically, raising concerns about higher concentration and lower diversification that might shift their overall risk profile. As funds increasingly allocate larger portions of their portfolios to AI-related companies, correlated adverse conditions—shifts in expectations, operational setbacks, regulatory challenges—could significantly affect fund portfolio valuations.

The European Central Bank itself sounded the alarm in its November 2024 Financial Stability Review. Quoting directly: The concentration of market capitalization in a handful of US tech stocks “raises concerns about the possibility of an AI-related asset price bubble.” More critically: “In a context of deeply integrated global equity markets, this points to the risk of adverse global spillovers, should earnings expectations for these firms be disappointed.”

Translation: Europe’s pension system is structurally exposed to American AI valuations. If those expectations disappoint, European retirees pay the price.

The ECB’s May 2025 Financial Stability Review noted that despite the market volatility following Trump’s tariff announcements, US equity valuations remained high. More concerning: “credit spreads appear to be out of sync with the currently very high level of geopolitical and policy uncertainty.” The market is underpricing the risk, and European pension funds are holding the bag.

B: The 2008 Echo Chamber

We’ve seen this movie before. The 2008 financial crisis originated in American financial engineering—subprime mortgages packaged into securities, circular relationships between originators and rating agencies, systemic risk disguised as diversification. Europe got devastated.

The Eurozone crisis. Greece, Spain, Portugal, Ireland drowning in debt. Youth unemployment hitting 50%+ in some countries. A decade of economic stagnation. Austerity measures that gutted public services. Political upheaval that fractured the European project. The rise of extremist parties. Brexit, partially traceable to economic wounds that never healed.

Europe’s structural mistake in 2008 was being financially integrated with American markets while having less fiscal flexibility to respond—trapped by a shared currency without fiscal union.

Now we face a bubble that analysts calculate is four times larger than 2008’s subprime crisis. Europe is MORE exposed through pension funds than it was through bank holdings of mortgage-backed securities. The infrastructure is MORE concentrated (a handful of AI companies versus thousands of mortgage originators). And The Economist’s advice is to weaken the labor protections that at least cushioned workers somewhat during the last collapse.

Europe didn’t hedge against the next American crisis. It bought more of it.

The Double Exposure

But European exposure isn’t just passive investment in American companies. Europe is building its own AI infrastructure bubble in parallel.

European data center operators and infrastructure providers saw significant stock gains in 2025, driven by the AI boom. French telecom group Iliad committed €3 billion to AI-focused infrastructure. Through its cloud computing division Scaleway, it established what it claims is the largest commercially available AI computing resource in Europe.

European AI startups can’t even raise money domestically. They’re forced to turn to US investors—who accounted for over 71% of European AI and machine learning venture capital deals by value in 2025, up from 57.5% in 2024. European “innovation” is being financed by the same American capital fueling the US bubble.

So when it pops, Europe faces double exposure:

Pension fund losses from US equity holdings
Stranded infrastructure investments in European data centers and AI companies
Funding collapse as US capital withdraws from European startups
All with weakened labor protections if The Economist’s advice is followed

Part III: The Chinese Mirror - The Future That Already Happened

China’s AI sector is the fast-forwarded version of the global future: a glut of capacity, a drought of returns, and a refusal to stop building.

China’s AI capital spending is projected to reach $98 billion in 2025, with government investment accounting for $56 billion and major internet firms contributing $24 billion. The approach is fundamentally different from the US model—state-directed rather than private capital, focused on self-sufficiency due to US export controls, building toward a National Integrated Computing Network that will integrate private and public cloud computing resources.

In 2023 and 2024, over 500 new data center projects were announced across China, from Inner Mongolia to Guangdong. At least 150 of the newly built data centers were finished. The central government designated AI infrastructure as a national priority, urging local governments to accelerate development of “smart computing centers.”

By the end of 2024, the excitement had curdled into disappointment. MIT Technology Review documents widespread project failures and distressed assets. Energy was being wasted. Data centers had become assets whose investors were keen to unload them at below-market rates. Industry observers predicted the Chinese government would likely step in, take over, and hand them off to more capable operators.

The demand never materialized. In 2024 alone, over 144 companies registered with China’s Cyberspace Administration to develop large language models. Yet only about 10% of those companies were still actively investing in large-scale model training by the end of the year.

Venture capital funding for Chinese AI startups dropped nearly 50% year-over-year in early 2025, falling to just $4.7 billion in Q2—its lowest level in a decade. The bubble had already popped in terms of actual investment appetite, even as the physical infrastructure continued rising.

The DeepSeek Irony

The catalyst for reassessment came from an unexpected source. DeepSeek, a Hangzhou-based startup, captured global attention by releasing two advanced open-source AI models developed at a fraction of the typical cost and computing power required for large language model projects. The company’s reported training run cost for its V3 model was approximately $5.6 million—a figure that excludes broader R&D but demonstrates the potential for efficient model development.

DeepSeek proved you could do more with less. The entire infrastructure buildout thesis—that more compute capacity equals better AI—was undermined by a scrappy startup demonstrating superior efficiency.

Yet the building continues. Alibaba announced a 380 billion yuan ($52.4 billion) capital expenditure plan in February 2025, targeting computing resources and AI infrastructure over three years—the largest private-sector investment in computing infrastructure within China. Tencent’s fourth-quarter 2024 capital expenditure nearly quadrupled year-over-year to 36.6 billion yuan ($5.1 billion).

Many of the freshly built facilities sit idle. Industry reports indicate many intelligent computing centers run well below capacity, with some massive installations in western provinces drawing gigawatts of contracted power while operating at a fraction of design utilization. Idle megawatts are the new dark fiber.

Many of the freshly built data centers were quickly strung together and don’t offer the stability that companies like DeepSeek would actually want. China built hundreds of data centers for an AI boom that shifted to efficiency rather than scale—then kept building anyway because stopping means admitting the strategy failed.

The pattern is instructive: local officials prioritized short-term projects that demonstrated quick results to gain favor with higher-ups rather than long-term development. Large, high-profile infrastructure projects have long been tools for local officials to boost political careers. Many projects were led by executives and investors with limited expertise in AI infrastructure. Some relied on middlemen who exaggerated demand forecasts to pocket government subsidies.

Beijing has done this before: LGFV-driven “ghost growth”—assets first, use later (or never). The ghost cities of the 2000s and 2010s—Ordos, Tianjin Eco-City, dozens of half-occupied “new districts”—were built using local government financing vehicles that borrowed off-budget to hit growth targets. Visible GDP today, hidden leverage tomorrow. When those projects underperformed, higher levels backstopped via policy banks or restructuring. Losses were socialized, momentum preserved.

The AI buildout is the same governance technology upgraded for racks and megawatts. SPVs replace LGFVs in name only; the fiscal physics are unchanged. Investment and installed capacity were the KPIs; utilization and ROI were afterthoughts. A 200-megawatt “AI park” photographs better than efficiency retrofits or boring software wins.

China is showing us in real-time what happens when you build massive infrastructure ahead of actual demand. And the US and Europe are watching this cautionary tale while accelerating their own versions of the same strategy.

Part IV: Convergence—Three Pipes, Same Pressure

Three Financing Models, One Outcome:

US: Private credit and SPVs create circular demand loops
Europe: Pension funds flow to US megacaps; domestic mini-bubble in data centers
China: Policy banks and LGFVs enable soft budget constraint; stranded megawatts accumulate

United States: Private Capital and Circular Flows

The American model concentrates massive private capital into closed-loop investment structures. Nvidia invests in OpenAI, which buys Nvidia chips. Oracle securitizes infrastructure commitments that better-capitalized companies rejected. OpenAI gives equity to secure chip supply diversification—not from strength, but from capture.

The scale is staggering. US tech companies are spending approximately $400 billion in 2025 on AI infrastructure. Total AI capital expenditures are projected to exceed $500 billion in both 2026 and 2027—roughly the annual GDP of Singapore, every single year.

Deutsche Bank issued a stark assessment in September 2025: “AI machines—in quite a literal sense—appear to be saving the U.S. economy right now. In the absence of tech-related spending, the U.S. would be close to, or in, recession this year.”

Read that again. The US economy is being propped up by speculative infrastructure spending. Not by productivity gains from AI. Not by revolutionary applications generating massive revenue. By the spending itself. Growth by excavation: GDP via capex, not customers.

The big AI firms are shifting huge amounts of spending off their books into special purpose vehicles (SPVs) that disguise the true cost of the buildout. They’re using accounting methods to depress reported infrastructure spending, which has the effect of inflating their profits. The financial engineering is becoming as important as the technology itself.

US Department of Justice antitrust probe activity focused on AI hardware market dynamics has been reported, examining potential concentration and competitive concerns in the sector. When two customers represent 39% of your revenue and you’re simultaneously their investor, supplier, and essential infrastructure provider, you haven’t built a market—you’ve built a dependency trap that regulators recognize as problematic.

The collapse mechanism is straightforward: when circular capital flows break—when one major player can’t maintain the spending, when revenue expectations disappoint, when the cost of capital rises—the synthetic demand evaporates. And because the relationships are so concentrated, the failure cascades rapidly.

If this is flexibility, Europe might want to stay rigid.

Europe: Double Exposure Through Financialization

Europe’s model is more passive but potentially more catastrophic. European pension funds have systematically increased their exposure to US equities over the past decade, with over half of their equity investments now in American companies. They’re not building the bubble—they’re betting workers’ retirement savings on it.

Europe as Exit Liquidity (by Design): European pension funds, seeking yield in a low-rate environment, became the bag holders for America’s financial engineering. By concentrating worker retirement funds into the most speculative assets—US tech companies building AI infrastructure ahead of demand—Europe made its social safety net structurally dependent on the continuation of circular capital flows it doesn’t control.

Simultaneously, Europe is constructing its own smaller AI infrastructure bubble. European data center operators saw their stocks surge 23% in 2025. Billions are flowing into AI-focused infrastructure across the continent.

But European AI companies can’t secure domestic funding. Over 71% of their venture capital comes from American investors—the same capital pools fueling the US bubble. European innovation is structurally dependent on American speculative finance.

When the US bubble pops, Europe experiences:

Massive pension fund losses from US equity holdings
Collapse of European AI infrastructure investments
Funding withdrawal from European startups as US capital retreats
Potential sovereign debt pressures as pension shortfalls require government intervention

And if The Economist’s advice is followed—if Europe weakens labor protections to be more “flexible”—European workers will face this crisis with less security than they had in 2008. Unemployment with fewer benefits. Job losses with less severance. Economic shock with diminished social safety nets.

The ECB has explicitly warned about this risk. They see it coming. And the policy prescription from influential voices is to make Europe MORE vulnerable, not less.

Europe didn’t hedge against the next American crisis. It bought more of it.

China: State-Directed Building Despite the Bust

China’s model is government-led infrastructure development with state-directed capital allocation. The government is investing $56 billion in 2025, with private companies adding another $42 billion. China is building a National Integrated Computing Network, integrating public and private cloud resources into a unified platform.

China as the Control Group: By replacing Wall Street’s financial engineering with state-directed capital, China isolates the core variable. Remove the circular capital flows, the securitization, the pension fund exposure—and you’re left with pure state capacity building ahead of demand. The result? The same outcome. Underutilized infrastructure. Funding collapse. Continued building despite obvious misallocation.

This proves the problem isn’t the financing model but the underlying premise: building massive infrastructure far ahead of proven, organic demand produces stranded assets regardless of who writes the checks.

The critical difference: China is already experiencing the bust phase. Five hundred data center projects announced, many sitting empty or becoming distressed assets. VC funding down 50% to decade lows. The majority of companies that registered to develop large language models have abandoned the effort.

Yet the building continues. Alibaba committed $52.4 billion over three years. State-owned telecom operators are increasing computing power investments by 20%+ annually. Local governments continue breaking ground on new facilities.

Why? Because China’s system doesn’t require market validation in the same way. State capacity can absorb losses that would bankrupt private companies. Local officials advance their careers through visible infrastructure projects regardless of utilization rates. The political logic of building supersedes the economic logic of demand.

The West financializes overcapacity via pensions and private credit; China financializes it via policy banks and LGFVs. Different pipes, same pressure to overbuild.

This makes China’s model the most resilient to capital market shocks—the state can keep building even as private investors flee. But it also means China is accumulating the most stranded assets. Resources, energy, and materials flowing into infrastructure that sophisticated AI developers don’t actually want or need.

China treats cycles as structural features, not moral failings. The Party doesn’t cling to exceptionalism—it clings to continuity. Ghost cities became mechanisms of recapture: absorb surplus labor and capital during overcapacity cycles, warehouse losses inside SOEs, fold stranded assets into state grids or telecoms, reset the political storyline around “common prosperity” or “modernization.” Each collapse becomes a phase transition rather than system failure.

Western systems run on belief in exceptional, irreversible growth. When the loop turns, it feels like failure rather than weather. China exhales and keeps building. The US and Europe panic.

China is the proof of concept for what happens when you build ahead of demand. The cautionary tale is already written. And yet the US and Europe are accelerating into their own versions of the same mistake, lacking the governance structures that allow China to absorb and redistribute the losses.

The Arithmetic of Absurdity

Let’s establish the baseline reality:

Total global AI infrastructure spending in 2025: Approximately $600 billion

United States: ~$400 billion
China: ~$98 billion
Europe and rest of world: ~$100 billion+

Annual consumer spending on AI services: $12 billion

For every dollar of revenue generated by AI applications, roughly fifty dollars of infrastructure are being built.

This isn’t a temporary mismatch between infrastructure development and application maturity. This is a fundamental disconnect between capital deployment and economic reality. The infrastructure being built today assumes demand that isn’t appearing tomorrow—it assumes demand orders of magnitude larger than current trends suggest will materialize over the next decade.

Infrastructure-first buildouts can precede killer applications—this has happened before. The critical question isn’t whether AI will eventually matter, but whether the current pace and scale of building can be sustained through the maturation period. The 70% utilization threshold in the falsification criteria represents the minimum bar for proving organic demand is catching up to supply.

The dot-com bubble left us with “dark fiber”—between 85% and 95% of fiber optic cable laid in the 1990s remained unused for years after the bubble burst. Companies like Global Crossing and Qwest built massive networks for demand that never materialized. The internet did eventually transform the economy, but the timing mismatch between infrastructure and utilization caused complete financial collapse for the builders.

The AI parallel isn’t idle fiber—it’s stranded power capacity and overbuilt GPU data centers. The difference is scale. The dot-com bubble was measured in hundreds of billions over several years. The AI bubble is measuring $500+ billion annually, sustained year after year, financed through increasingly complex mechanisms, distributed across global pension funds and sovereign balance sheets.

Three economic blocs. Three different models. Same fundamental error: building for a future that exists in presentations but not in revenue statements. And all three are accelerating because stopping means admitting what they’re building—not the infrastructure for an AI revolution, but the scaffolding for a synchronized financial collapse.

Part V: Why It Persists

If the arithmetic is this absurd, the warning signs this clear, and the Chinese preview this recent—why doesn’t anyone stop?

Career Incentives: No CEO Calls the Top

No executive wants to be the one who missed the AI revolution. The reputational and career risk of underinvesting in AI is immediate and severe. The risk of overinvesting won’t materialize until the bubble pops—and by then, everyone will have overinvested together. There’s safety in coordinated failure.

Tech CEOs face a simple calculation: spend billions on AI infrastructure and fail alongside your peers, or underinvest and fail alone while watching competitors capture the narrative. Even if executives privately harbor doubts about return timelines, the incentive structure demands they build.

The same dynamic applies to pension fund managers. Underweight US tech when it’s the dominant theme? Explain to beneficiaries why their returns lagged the benchmark. Overweight US tech when it crashes? Everyone else crashed too—you were following best practices and appropriate diversification models.

When individual rationality produces collective irrationality, you get bubbles. Everyone knows the music will stop. No one can afford to leave the dance floor early.

Policy Capture: Governments Can’t Withdraw

Governments are deeply invested—literally and structurally. The US economy, according to Deutsche Bank, would be “close to, or in, recession” without AI infrastructure spending. That spending is generating construction jobs, manufacturing orders, energy demand. It’s showing up in GDP growth, employment statistics, and tax revenue.

For a politician to question the AI buildout is to question economic growth itself. It’s to risk being blamed for any subsequent slowdown. The political logic is clear: ride the wave, claim credit for the growth, let your successor deal with the consequences.

China’s State Council has designated AI leadership as a national strategic priority. Xi Jinping himself has emphasized that China’s AI industry should be “strongly oriented toward applications.” When the paramount leader makes AI a cornerstone of national development plans, local officials don’t slow down—they accelerate to demonstrate loyalty and competence.

European governments face a different bind. Their pension funds are structurally exposed to US AI valuations. Any policy that acknowledges the bubble risk or attempts to reduce exposure would trigger exactly the capital flight and market volatility they’re trying to avoid. They’re trapped by their own portfolios.

Narrative Inertia: “AI is the Future”

The statement “AI is the future” has become definitionally true in a way that makes it unfalsifiable and therefore useless as investment thesis. Of course AI will be important—that tells us nothing about whether current valuations, infrastructure spending, or timeline assumptions are reasonable.

But the narrative has become load-bearing for too many institutions. Media companies built coverage models around AI hype. Consulting firms sold transformation roadmaps. Investors pitched funds. Academics secured grants. Politicians crafted industrial policy. The ecosystem that formed around “AI is the future” is now self-reinforcing.

Questioning the timeline or scale isn’t positioned as prudence—it’s positioned as being “anti-innovation” or “against progress.” The narrative has enough truth (AI will matter) to immunize against criticism of the details (but maybe not like this, not at this scale, not yet). The slogan is true enough to launder everything that isn’t.

Western systems, particularly, struggle with this because they’re built on narratives of exceptionalism and irreversible progress. Downturns are framed as accidents in an otherwise linear trajectory rather than predictable phases of cyclical systems. China’s soft budget constraint converts under-utilized AI campuses into public ballast—just as ghost-city land banks once did. The West has no equivalent mechanism, which means the correction will be sharper and more politically destabilizing.

This is how bubbles persist past the point of obvious absurdity. It’s not that participants don’t see the arithmetic. It’s that the system has made it individually rational to ignore it, politically impossible to act on it, and professionally dangerous to articulate it.

The building continues not because the case is compelling, but because stopping requires coordinated acceptance of reality—and the system has aligned incentives against that coordination.

What Would Change My Mind

This analysis is falsifiable. The bubble thesis would be invalidated by:

Consumer AI revenue sustainably exceeding $100B annually with demonstrated unit economics and margin proof across multiple quarters
Utilization of new AI data centers consistently above 70% for four consecutive quarters, indicating organic demand matching supply
Capital expenditure shifts from training infrastructure to inference and applications with clear ROI on deployed systems
European pension fund exposure to US megacap tech reduced by more than 10 percentage points, showing institutional risk management
China consolidating idle data centers into greater than 50% sustained utilization via policy bank restructuring

Any of these would suggest the timing mismatch is closing faster than the current evidence indicates.

Part VI: The Asymmetric Risk

The mechanics of bubble formation are symmetrical—everyone builds at once, everyone believes the same narrative, everyone accelerates together. But the distribution of consequences is profoundly asymmetric. When this coordination collapses, the pain won’t be evenly shared.

The system is structured to privatize gains and socialize losses. The impending correction is not a risk to the system’s architects but to its inhabitants.

The gains are captured immediately: executives via stock options based on capex-driven “growth,” financiers via fees on debt and equity deals, politicians via GDP figures inflated by the buildout itself. The losses are deferred and distributed downward to workers, retirees, and taxpayers who never participated in the upside.

United States: Flexibility as Fragility

American workers already operate in the most precarious labor market in the developed world. At-will employment means termination without cause or warning. Healthcare is tied to employment—lose your job, lose your insurance, potentially lose your home to medical bills. Unemployment benefits are minimal compared to other developed economies. Paid leave, sick days, parental leave—all dependent on employer discretion rather than legal requirement.

This is the “flexibility” The Economist praises. It’s what allows American companies to scale workforces up and down rapidly, to chase capital flows as they appear and vanish, to hire thousands for projects that might be cancelled next quarter.

When the AI bubble pops, American tech workers will experience mass layoffs. But unlike previous tech busts, this one is structurally different. The 2001 dot-com crash affected primarily software engineers and web developers—a relatively small, well-compensated segment of the workforce. The AI infrastructure boom has pulled in construction workers, manufacturing employees, logistics operators, energy sector workers. The ripple effects will be broader.

And American workers will face this with minimal safety nets. Severance packages are not legally required—they’re discretionary. Unemployment benefits replace only a fraction of wages and expire quickly. Healthcare costs continue regardless of employment status. The “flexibility” that enabled rapid scaling up provides no cushion for the scaling down.

The system optimized for building, not using—and workers will be told they weren’t “flexible” enough when it collapses.

Europe: The Pension Catastrophe

European workers face a different but potentially more devastating exposure. Their retirement security is systematically invested in American AI valuations.

When the bubble pops, European pension funds will experience massive losses. Over half of their equity holdings are in US companies, with significant exposure to the exact firms at the center of the AI infrastructure boom. The losses won’t be abstract—they’ll be concrete reductions in retirement income for millions of workers who never chose this exposure, who never invested in speculative AI ventures, who simply paid into their mandatory pension systems.

European workers aren’t just exposed to American valuations—they’re more exposed than they were to subprime mortgages in 2008. The concentration is higher. The sums are larger. And unlike 2008, where the contagion was initially hidden in complex securities, this exposure is direct and measurable.

The policy response will likely follow the 2008 playbook: pension shortfalls declared a crisis, government intervention required, austerity measures imposed to fill the gaps. Public services cut, retirement ages raised, benefits reduced. European workers will pay for American financial engineering through diminished retirement security and reduced social services.

The double exposure makes it worse. European AI infrastructure investments will fail simultaneously. European startups, dependent on American venture capital, will collapse as funding evaporates. The European data center boom will leave stranded assets. Workers face both direct pension losses and indirect economic contraction.

And if The Economist’s advice has been followed—if European countries weakened labor protections to be more “competitive”—workers will face this crisis with reduced job security, smaller severance packages, fewer unemployment benefits. They’ll have neither the financial cushion from stable pensions nor the employment protection from strong labor laws.

The political consequences could be severe. The 2008 crisis already fragmented European solidarity, fueled nationalist movements, and contributed to Brexit. A pension crisis affecting millions of workers across multiple countries—perceived as resulting from German, Dutch, and Scandinavian pension funds chasing American tech returns—could deepen those fractures.

European workers didn’t create this bubble. They won’t profit from it. But they’ll absolutely pay for it.

China: State Absorption, But at What Cost

China’s exposure is different in structure but similar in waste. The Chinese government can absorb losses that would bankrupt private entities. State-owned enterprises don’t face the same survival pressures as private companies. When hundreds of data centers sit empty, when billions of yuan of infrastructure goes unused, the Chinese state can carry those costs indefinitely.

This is genuine resilience—the ability to sustain investment through downturns, to maintain employment during transitions, to avoid the cascading failures that characterize market-based busts. Chinese workers won’t face mass layoffs in the American style. Pension funds aren’t systematically exposed to foreign market valuations. The social safety net, while imperfect, won’t collapse under the weight of investment losses.

But resilience doesn’t mean costlessness. The resources, materials, and energy flowing into unused AI infrastructure represent opportunity costs. Every yuan spent on empty data centers is a yuan not spent on healthcare expansion, education improvement, infrastructure that people actually need. Every engineer designing facilities for inflated projections is an engineer not working on problems with real demand.

China is also facing broader economic challenges that make this misallocation particularly painful. The real estate sector has plummeted. Youth unemployment exceeds 17%. Consumer confidence is declining. In this context, massive AI infrastructure investment represents a doubling down on the same growth model that created the real estate bubble—state-directed building ahead of organic demand, prioritizing visible construction over actual utilization.

Chinese workers won’t experience the acute shock of American-style layoffs or European pension collapses. But they’ll experience the chronic drain of resources misallocated at scale, the opportunity cost of an entire development strategy oriented toward infrastructure that sophisticated users don’t need.

The political costs may be subtle but significant. Xi Jinping personally emphasized AI as a national priority. When that strategy produces hundreds of empty data centers and a decade of lowest-ever venture capital funding, the credibility costs accumulate. The Chinese system can absorb economic failure more easily than political failure.

The Distributive Pattern

The pattern across all three blocs is consistent: those who built the bubble won’t bear its costs.

American tech executives will have already exercised stock options, collected bonuses for “growth,” and moved to the next venture. European pension fund managers followed benchmark allocations and industry best practices—no individual accountability for collective failure. Chinese officials who built unnecessary data centers will have already been promoted based on visible construction achievements.

Meanwhile:

American workers lose jobs with minimal severance in a precarious labor market
European workers see retirement security evaporate and social services cut
Chinese workers experience opportunity costs as resources flow to unused infrastructure

The asymmetry isn’t accidental. It’s structural. The system is designed to distribute gains upward during expansion and losses downward during contraction. Bubbles don’t change this pattern—they amplify it. The larger the bubble, the wider the gap between who profits and who pays.

Closing: The Suicide Pact

Return to where we began: The Economist’s October 2025 advice to Europe. Strip away labor protections. Make firing cheaper. Become more like America. Embrace “flexibility” as the path to innovation.

Now we can see what this advice actually means:

Europe should weaken the protections that cushion workers during economic shocks
To better participate in a speculative bubble built on circular capital flows and financial engineering
That analysts calculate is 17 times larger than dot-com and 4 times larger than 2008
While European pension funds are already more exposed than they were to subprime mortgages
And China is providing real-time evidence that building AI infrastructure ahead of demand produces empty facilities and funding collapse
Using the 2008 financial crisis—which devastated Europe through American financial contagion—as the template for what happens when this goes wrong
All while European workers would face this next crisis with weakened protections, diminished pensions, and reduced social services

This isn’t policy analysis. It’s not even bad advice. It’s advocacy for accelerated participation in a globally coordinated march toward financial catastrophe.

The synchronization is remarkable. Three economic blocs using three different models—private capital, pension fund exposure, state direction—all building for the same inflated projections, all ignoring the same warning signs, all accelerating despite the obvious arithmetic showing 50 dollars of infrastructure for every dollar of revenue.

The infrastructure will eventually be useful. AI will matter. The technology will mature. Just as the internet eventually grew into all that dark fiber, AI will eventually grow into this computing capacity.

But the dot-com companies weren’t wrong about the internet being transformative. They were wrong about the timeline. That timing mismatch—between when you build and when demand actually materializes—is what bankrupts the builders. Being right eventually doesn’t save you from being insolvent immediately.

When the music stops—when one major player can’t sustain the spending, when revenue expectations finally disappoint, when the cost of capital rises and circular flows break—the cascade will be global. American workers will face mass layoffs with minimal protection. European workers will watch their retirement security evaporate while funding cuts to social services. Chinese workers will inherit hundreds of empty data centers as monuments to misallocated resources.

And the executives, fund managers, and officials who built this? They’ll have moved on. The system that created the bubble will declare the bubble’s collapse an unfortunate but unforeseeable event. Calls will be made for flexibility, adaptation, and resilience—the same flexibility that left workers exposed, the same adaptation that means workers absorb losses, the same resilience that somehow never applies to the institutions that created the crisis.

The miracle of this moment isn’t innovation. It’s synchronization—three empires mistaking acceleration for progress, and marching together into the same collapse.

The real innovation would be stopping. Questioning the timeline. Demanding that infrastructure match actual demand rather than projected demand. Building social protections for workers before economic shocks rather than after. Learning from China’s empty data centers instead of pretending they’re irrelevant to Western models. Ghost cities became ghost datacenters; the budget constraint stayed soft.

But that would require admitting what’s being built. And the system has made it individually rational, politically impossible, and professionally dangerous to tell the truth.

So the building continues. The pension funds increase their exposure. The circular capital accelerates. The warnings are dismissed as pessimism or anti-innovation sentiment. And millions of workers whose retirement security, job stability, and economic futures are being wagered on this coordination continue their daily lives, unaware that their financial security is the collateral for a speculation they never consented to join.

This is the cross-national suicide pact. Not an agreement between leaders or institutions, but an emergent coordination failure—where every actor making locally rational decisions produces globally catastrophic outcomes. It’s gravity, not design. The prisoner’s dilemma playing out across continents: everyone knows cooperation (slower, measured buildout) would be better, but the individual incentive is to defect (build faster, capture narrative). Where innovation rhetoric obscures financial engineering. Where flexibility means fragility. Where building ahead of demand isn’t preparation but speculation.

The Economist thinks Europe’s problem is that workers are too secure. The actual problem is that the workers aren’t the ones making these decisions—and they’re the ones who’ll pay for them.

The system will call it a correction; history will call it coordination.