The AI Dependency Dilemma: Why the U.S. Economy Might Be Built on Sand

At Amedios, we examine the forces shaping the future of economies, societies, and technology. Few developments in recent history have triggered as much optimism and anxiety as the explosive rise of artificial intelligence. But as investment surges and stock markets soar, a deeper question emerges: is the current AI boom a sustainable engine of growth, or are we building a new economy on foundations that may not last?

This essay explores a provocative warning from Deutsche Bank: that without the massive capital expenditures on AI infrastructure, the U.S. economy would already be near - or even in - recession. We unpack the evidence behind that claim, explore the systemic risks it reveals, and look ahead to what must happen before the tide of investment inevitably turns.

When a Research Note Shook Wall Street

In September 2025, Deutsche Bank released a report with a striking title: “The Summer AI Turned Ugly: Part 1.” Authored by chief strategist George Saravelos, the study examined the economic forces underpinning U.S. growth and reached a conclusion that sent ripples through global markets.

According to Saravelos, without the surge in technology-related capital expenditures, and especially those tied to artificial intelligence infrastructure, the U.S. economy would likely be “close to, or already in, recession.” The report went further, suggesting that “AI machines (in quite a literal sense) appear to be saving the U.S. economy right now.”

The analysis painted a revealing picture. While consumer spending remained resilient and inflation was moderating, much of the real growth momentum came from a narrow but powerful source: massive investments in data centers, GPUs, cloud infrastructure, and semiconductor fabrication. These investments, largely driven by a handful of Big Tech companies, were not just boosting corporate earnings. They were materially inflating GDP.

But this reliance on AI CapEx raises fundamental questions. Is this sustainable growth, or a temporary sugar high? What happens if these investments plateau or reverse? And most importantly, how should policymakers, businesses, and societies prepare for that moment?

The Claim Under the Microscope: Is AI Really Holding Up the Economy?

The data suggests that Deutsche Bank’s warning is not hyperbole. In the last 18 months, U.S. GDP growth has consistently outperformed forecasts, but when analysts strip out AI-related investment - particularly in data center construction and AI hardware - the picture looks far more fragile.

Private non-residential fixed investment, typically a broad-based indicator of business confidence, is now disproportionately concentrated in a handful of AI-related categories. Meanwhile, broader business investment has remained relatively flat.

Stock market performance tells a similar story. The so-called “Magnificent Seven” tech companies - Microsoft, Apple, Amazon, Meta, Alphabet, Nvidia, and Tesla - account for an outsized share of market gains, with much of their valuations now tied to AI infrastructure bets. Their combined capital expenditure is on pace to exceed $300 billion in 2025. This a historicallly unprecedented figure.

But the crucial nuance is that most of this spending is infrastructure-first. Like the railways of the 1800s or the fiber-optic networks of the 1990s, today’s investment boom is laying the groundwork for future productivity. Whether that productivity arrives quickly enough to justify the spending remains an open question — and one that will define the economic trajectory of the next decade.

The Hidden Risks of an AI-Fueled Economy

If AI investment is truly propping up the economy, then the economy’s future hinges on the outcome of that investment. Beneath the optimism lie several profound risks - each of them with the potential to reshape growth, markets, and policy if it materializes.

1. The CapEx Bubble Risk: Infrastructure Without Immediate Returns

Today’s AI boom is overwhelmingly a capital expenditure phenomenon. Vast sums are flowing into hardware, facilities, and infrastructure — but the commercial returns of many AI applications remain unproven.

If revenues fail to materialize at the pace investors expect, corporate spending could slow dramatically. Such a pullback would not only hit GDP growth but also ripple through construction, energy, and manufacturing sectors that have become dependent on AI-related demand. A sudden contraction could resemble the telecom crash of the early 2000s. Only this time, the stakes are larger and more interconnected.

Moreover, public markets have priced in decades of future growth. There could be sharp corrections if the promise of AI monetization lags too far behind its costs. The psychological and financial shock could erode consumer confidence, tighten credit, and spark broader economic volatility.

2. Concentration Risk: A Few Giants Carry the Economy

The AI boom is highly concentrated. Fewer than ten companies are driving the bulk of investment, infrastructure, and innovation. While their scale enables transformative projects, it also introduces fragility.

If even one or two of these companies face regulatory hurdles, antitrust action, or strategic missteps, the consequences could cascade across the economy. A slowdown in capital spending by a single hyperscaler could shave significant points off GDP growth, disrupt supply chains, and trigger layoffs across dependent industries.

This concentration also distorts market dynamics. Smaller firms often cannot compete with the capital intensity required to train state-of-the-art models or build hyperscale infrastructure. As a result, innovation risks could becoming centralized, reducing competitive pressure and slowing the diffusion of AI benefits across the broader economy.

3. Productivity Paradox: The Lag Between Investment and Impact

History offers a consistent pattern: transformative technologies deliver productivity gains only after significant time lags. Electricity, the steam engine, and the internet all triggered massive investment waves before their economic benefits fully materialized.

AI may follow the same arc. Today, it is already transforming sectors like coding, logistics, and drug discovery, but at a macro level, productivity growth remains modest. If that lag extends too long, political patience could erode. Policymakers may face pressure to scale back incentives, investors could grow disillusioned, and corporate boards might reallocate capital to more immediate-return projects.

The economic implications of such a scenario are profound: weaker productivity would mean slower wage growth, reduced fiscal space, and heightened inequality — a difficult backdrop for sustained prosperity.

4. Energy and Environmental Risks: Powering the Future Has a Price

AI’s computational hunger is enormous; and it is growing fast. Data centers already consume more electricity than some nations, and demand is projected to triple within a decade. This surge is reshaping energy markets and forcing utilities to reconsider long-term infrastructure plans.

If supply fails to keep pace, energy shortages or price spikes could ripple through the broader economy. Inflationary pressures might intensify, undermining monetary policy. Industries competing for the same energy resources (from manufacturing to transportation) could face higher input costs, reducing competitiveness.

Additionally, the environmental footprint of AI infrastructure may become a flashpoint in climate policy debates, potentially leading to stricter regulations or public backlash that slows deployment.

5. Policy and Geopolitical Risks: Strategic Dependencies in a Fractured World

AI’s supply chains are deeply global and deeply vulnerable. From advanced semiconductors to rare earth minerals, many critical inputs are concentrated in geopolitically sensitive regions.

A trade dispute, sanctions regime, or conflict involving a key supplier, such as Taiwan’s semiconductor industry, could disrupt AI production capacity, sending shockwaves through global markets. Such a disruption could trigger inflation spikes, stifle innovation, and even compromise national security strategies that increasingly rely on AI capabilities.

This interconnectedness means AI is not merely a technological dependency. It is a strategic one. If governments fail to coordinate industrial policies, diversify supply chains, and invest in domestic capabilities, the world’s economic resilience could erode.

6. The “Shovel Seller” Dilemma: Who Really Profits from the AI Gold Rush?

In every gold rush, the people selling shovels often make more money than those digging for gold. The current AI boom is no exception. Companies like Nvidia, which designs the GPUs powering generative AI, have seen their market valuations skyrocket. Nvidia has even invested directly in OpenAI - one of its largest customers - tightening its position in the ecosystem.

Meanwhile, many companies building AI platforms are still struggling to monetize their products at scale. This imbalance highlights a deeper structural issue: the value capture in the AI economy is heavily skewed toward infrastructure providers, not application developers.

If this dynamic persists, we could see a bifurcated economy in which hardware and energy suppliers reap outsized rewards while downstream innovators face thinner margins. Over time, that could deter risk-taking, slow innovation, and leave the broader economy overly dependent on a narrow segment of the value chain.

The Next Critical Phase: From CapEx Boom to Sustainable Productivity

The question is no longer whether AI investment can support short-term growth. It clearly can. The real challenge is what happens next: how economies, businesses, and governments navigate the transition from today’s infrastructure surge to tomorrow’s productivity revolution.

This transition will not happen automatically. It requires deliberate action, strategic foresight, and coordinated policy. The following four transformations are essential - and each of them demands bold decisions.

1. From Infrastructure to Application: Turning Capacity Into Capability

Building data centers and training massive language models is just the beginning. The next phase must focus on integrating AI into real-world processes - from supply chains and logistics to healthcare, education, and public administration.

This requires significant investment in adoption: workforce training, organizational change, data governance, and regulatory clarity. If this shift doesn’t occur, trillions in CapEx risk becoming stranded assets — impressive infrastructure with limited economic impact.

What's needed: Governments should incentivize AI deployment across sectors beyond Big Tech, while enterprises must focus on operational transformation, not just experimentation.

2. From Potential to Productivity: Measuring What Matters

Many AI projects today are measured by technical milestones like model size, parameter counts, or benchmark scores rather - and not so much by their economic contribution. The next decade will demand a different focus: return on intelligence (RoI). In the future, there has to be a clear link between AI deployment and measurable productivity gains.

If organizations fail to bridge that gap, they risk repeating the mistakes of past tech cycles, where hype outpaced value. That could trigger disillusionment, investment pullbacks, and stalled innovation.

What's needed: Businesses must align AI investments with core KPIs like efficiency, revenue growth, and customer satisfaction. Policymakers have to refine economic metrics to capture intangible gains like algorithmic decision-making and knowledge automation.

3. From Concentration to Diffusion: Spreading the Benefits

AI’s current economic impact is concentrated among a small group of companies and industries. For the technology to transform the broader economy, adoption must diffuse to small and medium-sized enterprises (SMEs), the public sector, and developing regions.

Failure to do so will widen inequality between digital leaders and laggards, creating structural imbalances that drag down long-term growth. Broad diffusion, on the other hand, could unlock new waves of entrepreneurship, job creation, and regional revitalization.

What's needed: Policymakers should support open-source initiatives, digital infrastructure for SMEs, and AI-ready education. Corporates must build accessible platforms and ecosystems that empower partners, suppliers, and smaller innovators.

4. From Hype to Governance: Building a Framework for Trust and Resilience

AI’s societal impact will not be determined by technology alone but by the governance frameworks around it. Ethical safeguards, privacy standards, liability regimes, and workforce transition policies will shape how and how widely AI is deployed.

If governance lags behind, public trust could erode, regulatory crackdowns could intensify, and adoption could stall. A proactive governance approach, however, can foster confidence, attract investment, and ensure that growth is inclusive and sustainable.

What's needed: Governments should collaborate internationally on AI regulation, while companies embed ethics, transparency, and accountability into their AI strategies from the outset.

Future Outlook 2030: Three Possible Scenarios

The next five years will determine whether the AI boom becomes the foundation of a new era — or a footnote in economic history. 

Here are three plausible scenarios:

The Optimistic Scenario – The Productivity Renaissance: AI drives measurable productivity growth across industries, boosting GDP by 1–2 percentage points annually. Infrastructure investments pay off, new business models emerge, and societies adapt smoothly to workforce changes.

The Realistic Scenario – The Uneven Transition: AI delivers strong gains in some sectors but remains underutilized in others. Productivity improves, but not enough to fully offset slowing demographics and debt pressures. Policymakers scramble to manage inequality and market concentration.

The Risk Scenario – The AI Bust: Monetization lags, investment slows, and markets correct. The economy faces a “post-CapEx hangover,” revealing structural weaknesses. Governments intervene with industrial policies, but growth remains sluggish for years.

Which path becomes reality will depend on the decisions made today.

Conclusion: Leading Beyond the Boom

Artificial intelligence has become the invisible scaffolding of the U.S. economy. It is a force that is not just transforming industries but sustaining growth itself. Yet this very dependency exposes vulnerabilities that must not be ignored. CapEx alone cannot build a prosperous future. Productivity, diffusion, trust, and governance must follow - or otherwise the economic tide could turn far faster than many expect.

The Deutsche Bank warning is more than a headline. It is a signal. It serves us as a strong reminder that technological revolutions are not defined by how much we invest, but by how wisely we guide them.

The next decade will test whether this AI boom becomes a new industrial revolution or a cautionary tale. Leaders in business, government, and society alike have a narrow window to ensure it’s the former.