Is the AI Bubble About to Burst? Circular Investment, Debt Risks, and What Happens Next

The question isn’t whether ai valuations have gotten ahead of themselves. It’s whether you’re positioned to survive when the reckoning arrives.

Since ChatGPT launched in November 2022, we’ve witnessed one of the most aggressive capital deployments in market history. Nvidia’s market cap has exploded past $5 trillion. OpenAI’s valuation has jumped from roughly $29 billion to nearly $500 billion. And ai startups have absorbed more than half of every venture capital dollar flowing in 2025.

But beneath the surface of this ai boom, a troubling pattern has emerged: circular investment structures where the same dollars flow between a small group of interconnected players, inflating valuations without generating proportional real value. This isn’t speculative conjecture—it’s the explicit warning coming from the IMF, the Bank of England, and even insiders like openai ceo sam altman himself.

In this analysis, I’ll break down exactly what’s happening, why the circular nature of these ai deals creates systemic fragility, and how you can protect your portfolio whether the bubble pops tomorrow or holds together for another few years.

Answer first: Is the AI bubble about to burst right now?

Let me give you a straight answer: AI markets are showing classic late-stage bubble characteristics, but calling the exact timing of a burst in 2025-2027 is impossible. What I can tell you is that the fragile piece isn’t artificial intelligence as a technology—it’s the circular financing structures propping up current valuations.

The numbers are staggering. Tech giants have committed roughly $3 trillion in ai infrastructure spending by 2028. OpenAI has floated plans for up to $1.4 trillion in data centers over eight years. In September 2025, Nvidia poured $100 billion into OpenAI with the explicit expectation that OpenAI would purchase Nvidia chips for new data centers. Money flows from chipmaker to model developer and straight back through hardware purchases.

This is the circularity that should concern you. When Nvidia invests in OpenAI, which buys Nvidia chips, which boosts Nvidia’s revenue, which justifies Nvidia’s valuation, which enables more investment in OpenAI—you’re not looking at organic demand. You’re looking at a closed loop where the same capital creates the appearance of growth.

A sharp market correction becomes plausible if any of these conditions materialize in the next 12-24 months:

• Enterprise adoption of ai services fails to deliver measurable ROI, causing companies to cut pilot budgets
• Consumer willingness to pay for ai technologies plateaus below growth projections
• Interest rates stay elevated, making it harder to roll over the billions of dollars in debt financing these buildouts
• A major ai companies stumbles, triggering contagion across interconnected deals

Immediate red flags investors should monitor:

• Valuations: Nvidia trading at price-to-sales multiples well above historical chip-sector norms
• Debt surge: Hyperscalers taking on roughly $121 billion in new debt over a recent 12-month window—triple typical levels
• Circular revenues: Major ai companies booking revenue from partners who are simultaneously investors and customers
• Concentration: AI accounting for 75% of S&P 500 returns and 80% of earnings growth since ChatGPT’s debut

What is the “AI bubble” and why are investors talking about it?

An ai bubble refers to stock and private-market valuations of AI-related firms detaching from realistic cash flow expectations, amplified by reflexive hype and cheap capital. When stock prices climb not because of revenue growth but because investors expect other investors to keep buying, you have the textbook definition of speculative excess.

This isn’t abstract. The ai market has delivered concrete milestones that echo previous manias:

• Nvidia’s market cap jumped from approximately $1 trillion in mid-2023 to over $5 trillion by 2025—a 5x increase in roughly 18 months
• OpenAI’s valuation moved from around $29 billion in 2023 to roughly $500 billion by early 2025
• AI startups raised $192.7 billion in venture capital in 2025, capturing nearly two-thirds of U.S. deal value
• AI-related capital expenditures surpassed U.S. consumer spending as the top gdp growth driver at 1.1% in early 2025
• Single month funding rounds: June 2025 alone saw $40 billion flow into data center expansions

The historical parallels are unavoidable. During the dot com bubble, internet companies raised capital on promises of future monetization while building out fiber-optic networks that reached 93% excess capacity at peak. It took years—in some cases over a decade—for actual demand to absorb that infrastructure.

What’s particularly notable in 2024-2025 is that institutions that typically avoid alarmist language are now explicitly using “ai bubble” terminology. The IMF has equated AI investment risks to the dot com era. The Bank of England has warned of escalating bubble risks. These aren’t Reddit commenters—they’re central banks and global financial institutions.

The crypto boom of 2020-2021 offers another comparison point. Like AI, crypto saw story-driven valuations, massive infrastructure buildouts, and the belief that adoption would inevitably validate prices. The subsequent crash erased over $2 trillion in value.

The difference with AI? Larger incumbents with real current revenue are leading the charge. But as we’ll see, even profitable companies can see their share prices collapse when valuations disconnect from fundamentals.

Signs we might already be in an AI bubble

Survey data paints a concerning picture. Over 75% of surveyed economists in 2024-2025 described current AI investment levels as “bubbly” or “bubble-like.” At Yale’s CEO Summit in June 2025, 40% of 150+ executives—including venture capitalists and tech founders—foresaw an imminent correction amid AI exuberance.

The valuation extremes are hard to ignore:

Venture capital behavior has reached irrational exuberance territory. AI captured 1 in every 2 VC dollars in 2025, up from 23% in 2023. Funding is flowing to “AI-in-name-only” firms and copycat GenAI tools with no clear path to revenue. When the hottest trend attracts this much undiscriminating capital, the industry inevitably produces losers that will lose money in spectacular fashion.

The concentration is extraordinary. ai stocks have driven 75% of S&P 500 returns since late 2022. They’ve accounted for 80% of earnings growth and 90% of capital spending growth. This echoes how internet stocks dominated indices before the 2000 crash—when those stocks fell, they dragged down entire portfolios that appeared diversified but weren’t.

Corporate behavior tells its own story. Hyperscalers like Microsoft, Google, Meta, Amazon, and Oracle are dedicating up to 50% of operating cash flow to AI and data centers in 2024-2025. KPMG surveys show average enterprise AI investments rising 14% quarter-over-quarter to $130 million per firm. This spending is tied to expected productivity gains that haven’t yet materialized at scale.

The clearest bubble signals:

• Majority of index returns coming from a small group of AI-related names
• Tens of billions flowing into startups with unproven business models
• Corporate capex running ahead of demonstrated enterprise demand
• Valuations assuming perfect execution and limitless growth

Circular investment: how AI companies are funding their own demand

Circular investment is the practice of ai companies financing each other’s projects in ways that make demand, revenue, and valuation all feed on each other. It’s the financial equivalent of taking money from your left pocket, putting it in your right pocket, and counting it as income.

Let me walk you through how this actually works in the ai industry:

The Nvidia-OpenAI Loop:

1. Nvidia invests approximately $100 billion in financing arrangements for OpenAI-linked data centers
2. OpenAI uses this capital to build infrastructure that requires Nvidia GPUs
3. OpenAI purchases those chips from Nvidia, generating Nvidia revenue
4. Nvidia’s revenue growth justifies its valuation, enabling more investment
5. The cycle repeats, with each participant booking “growth” from the same underlying capital

The OpenAI-AMD-Nvidia Triangle:

• OpenAI holds a 10% stake in AMD alongside Nvidia’s massive investment
• This intertwines the fates of chip rivals with model developers
• Investment returns depend on the success of entities that are simultaneously customers, suppliers, and equity holders

The CoreWeave Model:

• CoreWeave rents Nvidia-powered computing power to AI firms
• It receives equity stakes in exchange for capacity
• It uses that valuation to borrow more capital
• It deploys that capital to buy more Nvidia hardware
• The company that provides supply is equity-linked to demand

These loops create several dangerous dynamics:

• Inflated AI revenue: Cloud spend, chip orders, and data center leases all get booked as “growth” even when the underlying cash demand comes from circular partners rather than genuine end users
• Masked demand: It becomes nearly impossible to distinguish organic enterprise adoption from circular financing dressed up as market traction
• Opacity: When special purpose vehicles (SPVs) and off-balance-sheet arrangements enter the picture, you get Enron-style complexity that hides true exposure

The critical vulnerability: when one part of the loop slows—AI startups fail to monetize, enterprises cut pilot budgets, or hedge funds stop funding growth at any cost—the entire chain can reprice rapidly.

This is similar to structured-finance contagion in 2008, where interconnected mortgage securities meant that problems in one corner of the market cascaded everywhere. A handful of players—OpenAI, Nvidia, CoreWeave, Microsoft, Google—dominate multibillion-dollar deals where failures could trigger similar domino effects.

In June 2025, graphics visualizing these circular investments went viral on social media, amplifying public scrutiny. The ai revolution may be real, but the financial structures funding it deserve skepticism.

Debt, SPVs, and the hidden leverage behind AI data centers

The magnitude of ai infrastructure spending is unprecedented. Morgan Stanley projects around $3 trillion in AI-related capex by 2028. OpenAI alone has discussed up to $1.4 trillion in data center buildout over eight years, plus a separate $300 billion compute deal with Oracle over five years. These aren’t budget line items—they’re commitments that rival the GDP of medium-sized countries.

How are tech companies financing this? Increasingly through debt and special purpose vehicles that keep the true leverage off headline balance sheets.

Hyperscalers have taken on roughly $121 billion of new debt in a recent 12-month window—approximately triple their typical annual levels. This represents a pretty significant shift from the cash-flush, low-debt model that made Big Tech resilient during previous downturns.

The SPV playbook works like this: A tech company creates a separate legal entity to buy GPUs and build data centers. The SPV takes on debt, not the parent company. The parent leases capacity from the SPV, guaranteeing cash flow to service the debt. The parent may own a minority stake while limiting consolidated balance sheet exposure.

A concrete example: Blue Owl Capital arranged roughly $27 billion in financing for a Meta-linked data center. Meta leases capacity and owns a minority stake while keeping the debt off its main balance sheet. If AI economics disappoint and the data centers become stranded assets, the debt obligations remain.

Key mechanics of these financing structures:

• Off-balance-sheet debt: Billions of dollars in obligations that don’t appear in headline debt figures, making companies look more financially healthy than they are
• Asset-liability mismatch: Highly specialized data centers designed for current-generation AI workloads may become obsolete faster than the debt matures
• Lease guarantees: Parent companies often provide implicit or explicit guarantees that could create sudden liability recognition if deals unwind
• Interconnected exposure: Many SPVs involve the same counterparties appearing in circular investment loops
• Limited transparency: Private company SPV structures face minimal disclosure requirements compared to public markets

The historical precedent that should concern investors? Enron. That company used off-balance-sheet vehicles to hide losses and inflate apparent profitability until the entire structure collapsed. The specific mechanics differ, but the principle—keeping liabilities invisible until they become unmanageable—is identical.

If AI demand growth disappoints, these oversized, specialized data centers cannot easily be repurposed. You can’t convert a facility optimized for Nvidia H100 clusters into a warehouse. The debt remains, the assets depreciate, and the cash flow projections evaporate.

What could actually pop the AI bubble?

A bubble burst doesn’t require a single catastrophic event. The dot com bubble deflated over two years, not two days. Here are the concrete scenarios that could trigger an AI bubble bursting in the 2025-2028 timeframe:

Revenue Disappointment (Most Likely)

Enterprises fail to see measurable ROI from GenAI pilots. This is already the concern voiced by Goldman Sachs CEO David Solomon, Amazon’s Jeff Bezos, and even sam altman of OpenAI. If the “AI productivity gains” don’t materialize in corporate earnings, cloud and compute growth slow, and the entire capex thesis unravels.

Hardware Obsolescence

Rapid efficiency gains or architectural breakthroughs (including quantum computing advances) strand today’s GPU-heavy data centers. The past few years have shown how quickly AI chip generations evolve. Hundreds of billions invested in current infrastructure could become stranded assets, much like the unused fiber optics from the dot com era that took over a decade to fully utilize.

Regulatory or Governance Shock

A major AI incident—systemic model failure, large-scale security breach, catastrophic autonomous system error—triggers regulatory clampdown or moratorium. Governments have already shown willingness to intervene in AI development. A single high-profile disaster could freeze investment across the ai industry.

Credit Tightening

Higher interest rates or widening credit spreads make AI capex impossible to roll over. The debt-financed buildout depends on continued access to cheap capital. If rates stay elevated or credit markets tighten, the SPV structures become unsustainable, exposing over-leveraged positions.

Startup Cascade

A wave of ai startups failing to monetize triggers venture capital pullback. With AI capturing the majority of VC investment, a loss of confidence in the sector reverberates across the entire startup ecosystem. Funds that lose money on AI may have to mark down other positions, creating selling pressure beyond the sector.

Demand Saturation

Consumer and enterprise willingness to pay for AI services plateaus below growth expectations. If users won’t pay premium prices for AI-enhanced products, the economics that justify current valuations collapse regardless of technical capabilities.

A “pop” need not be a single day crash. It can be a multi-year derating where AI multiples fall and capex collapses, even if AI usage continues to grow underneath.

This distinction matters. The ai revolution can be real while the ai bubble still bursts. Technology adoption and stock prices are not the same thing.

Lessons from dot-com and other bubbles: why AI might be different this time

Every bubble survivor claims their era is unique. But the 1995-2000 internet bubble offers direct parallels worth examining:

What rhymes with dot-com:

• A genuine technology shift generating unrealistic near-term profit expectations
• Massive infrastructure buildout that was initially unprofitable but later became essential
• Valuations based on “story” rather than cash flow
• Circular financing where internet companies invested in each other and booked paper gains
• Concentration of index returns in a handful of high-flyers

The fiber-optic networks built during the dot com era eventually became the backbone of today’s internet economy. But investors who bought at peak valuations in 1999-2000 waited over a decade to recover—if they ever did. Many companies simply disappeared.

AI may follow a similar pattern. The hundreds of billions flowing into data centers today could be “overbuilt” relative to 2025-2027 demand but form essential infrastructure for 2030s applications. That’s cold comfort if you bought Nvidia at $5 trillion and watched it fall 60%.

What’s genuinely different:

Jensen Huang, Nvidia’s chief executive, has drew comparisons to the dot-com era to argue his case, calling the AI surge “very different” from a bubble and emphasizing tangible infrastructure like chips and cloud storage.

The uncomfortable middle ground:

Even strong companies can see their stocks collapse in a repricing. After 2000:

• Cisco fell roughly 80% and never recovered its peak
• Intel declined over 75% and remained below its high for two decades
• Microsoft dropped approximately 65% and took 15 years to set new highs

These weren’t failed businesses. They were market leaders that continued generating billions in revenue and profits. But their valuations had gotten ahead of fundamentals, and the market correction was brutal.

The lesson: AI as a technology can thrive while AI investors still lose money. Surviving a bubble isn’t about avoiding the technology—it’s about managing exposure to overextended valuations.

Investor playbook: how to protect yourself if the AI bubble bursts

Let me be direct: timing a bubble top is extremely difficult. Investors who shorted the dot-com bubble in 1998 were right about the fundamentals and still got destroyed before the crash. The ai hype can remain irrational longer than you can remain solvent.

Focus on risk management, not heroic market-timing calls.

Portfolio diversification strategies:

The tech giants driving the current ai boom—Nvidia, Microsoft, Alphabet, Amazon, and Meta—dominate S&P 500 and Nasdaq 100 weightings. If you own index funds, you likely have far more AI exposure than you realize.

Active hedging strategies:

• Avoid outright shorts on AI leaders. Short squeezes can be devastating, and stocks can remain overvalued for years.
• Consider options strategies for defined-risk hedging rather than unlimited-loss short positions
• Sector rotation—shifting from tech stocks to value, defensive, or international sectors—provides natural hedge without betting against momentum
• Reduce overweight positions gradually rather than trying to call exact tops

Your pre-crisis checklist:

Review your portfolio’s AI exposure by sector and single-stock concentration. Most investors underestimate how much their wealth depends on a small group of names.

Stress-test your holdings: how would a 50% decline in top AI names affect your overall net worth? Can you tolerate that drawdown without forced selling or lifestyle changes?

Decide in advance under what conditions you would reduce or add to AI positions. Having predetermined rules prevents emotional decision-making during market turmoil.

Consider this: even if AI tech stocks fall 50-70%, the ai industry will likely continue developing transformative applications. The investors who thrive will be those who maintained enough diversification to survive the correction and deploy capital at lower valuations.

The companies building AI aren’t going away. But their stock prices and the tech stocks that ride the wave don’t have the same guarantees.

This isn’t personalized advice—I don’t know your situation, timeline, or risk tolerance. But the principles of diversification, stress-testing, and predetermined decision rules apply regardless of whether you’re managing hundreds of millions through a family office or building wealth through a 401(k).

The ai revolution may indeed transform the world over the coming decades. That transformation doesn’t require today’s valuations to be correct. In fact, history suggests they rarely are.

Review your exposure. Make a plan. And remember that surviving to invest another day is more important than capturing every last dollar of upside.

Your Friend,

Wade