Is artificial intelligence the next big bubble? The truth behind the “AI Ponzi scheme” claims.

Is the AI industry turning into a ‘artificial intelligence Ponzi scheme?’

In recent months the term “AI Ponzi scheme” has surfaced with increasing regularity in commentary about the tech sector. On its face, the rapid rise of AI infrastructure investing, strategic partnerships among chip-makers, cloud providers and AI labs, and spectacular headline deals may seem to signal a genuine revolution. But a closer look suggests something less benign: a vast, highly inter-connected web of cross-investments and purchase commitments that raise troubling questions about whether the entire industry is more self-referential than value-creating.

Below we examine how the AI value chain is evolving, why the circular financing and mutual stock-buying among firms resembles classic Ponzi/Pyramid-type structures, what the technical and financial risks are, and why investors, despite strong underlying fundamentals in some firms, should remain cautious.

Circular financing: the ecosystem becomes its own customer

At the centre of the current AI boom you’ll find companies such as Nvidia, OpenAI, AMD and major cloud/data-centre firms. The business model is increasingly inter-woven: chip-maker invests in AI lab; AI lab agrees to buy large volumes of chips from that chip-maker; cloud/data-centre firm in turn invests in both chip-makers and AI labs; and the firms publicly announce multi-billion-dollar commitments that feed headline valuations and exuberant press coverage.

For example: OpenAI has committed to buying tens of billions of dollars worth of AMD chips and AMD gave OpenAI the right to buy a roughly 10% stake in AMD, contingent on share-price and deployment thresholds. At the same time, Nvidia pledged up to US$100 billion in investment into OpenAI and OpenAI will buy large volumes of Nvidia chips.

The same dollars seem to be circulating through the system: chip-maker → AI lab → chip-maker, with each transaction supporting the other’s narrative. In short: company A invests in company B, company B buys from company A, and each uses the announcement to boost its market cap.

Financially and technically this gives the appearance of growth. But when your customers are also your investors, and your suppliers are your partners, you run the risk of circular revenue that is, money flowing around the loop rather than outwards in genuinely new business. Analysts have warned that this “investment carousel” may obscure the true quality of revenues and demand. One recent piece flagged that “the billions pouring into AI startups reveal a tangled web of cross-investments where the same companies buying, selling and funding each other” are raising concern.

Why this smells like more than hype: Ponzi mechanics

A bona fide Ponzi scheme typically works like this: early investors are paid returns not from legitimate new business, but from the capital of later investors. The structure must keep growing and recruiting fresh capital to keep the illusion alive, until the inflow cannot keep up and the scheme collapses. While the AI industry isn’t illegal or fraudulent in this sense, many of the same mechanics are present:

Mutual investing among firms: Firms buy stakes in one another and announce multi-billion dollar commitments to purchase each other’s hardware or services.
Headline-driven valuations: The announcements generate exuberance and drive share-price gains especially in firms whose fundamentals (revenues/profits) are still weak.
Dependence on future demand: Many of these deals assume massive scale-up of AI infrastructure usage and monetisation that is yet to materialise.
Circular flows of funds: A chip-maker invests in an AI lab; that lab commits to buy chips from the chip-maker; the chip-maker benefits from its own investment. This is analogous to a loop where funds rotate rather than create independent value.
Need for fresh capital: If demand disappoints, there could be no one left to pay for the infrastructure or services, imposing stress on the entire chain.

Reddit users have commented on this phenomenon in plain terms:

“Nvidia is giving a company $100 billion so that company can buy Nvidia chips with it.”
“The same money being shuffled around across different companies… each cycle making balance sheets look far healthier (than reality).”

Although the players involved are legitimate and some have real earnings, the scale, complexity, and tight inter-linking of these deals mirror the structural risks of a pyramid: the growth narrative is being supported, in part, by internal circularity.

Technical & financial fragility: cracks in the foundation

Beyond the structural finance concerns, there are serious technical and financial issues that compound the risks.

Infrastructure scale and power demand

Consider this: one announced project, OpenAI’s “Stargate”, reportedly aims to build 10 gigawatts of AI data-centre capacity across the US a quantum leap in power, equivalent to roughly ten nuclear power plants in output. Another commitment is 23 GW of new capacity costing well over a trillion dollars. Yet new nuclear reactors in the U.S. take more than a decade to build; grid, permitting and power-supply bottlenecks remain major hurdles. In other words, the electricity supply may not match the hype. If compute cannot scale as promised, then the down-side is large.

Monetisation and usage unknowns

Many of the big AI labs are still unprofitable or burning cash at pace. One key metric: although OpenAI claims ~700 million weekly users, only about 5% are paying customers. For enterprise AI projects, success rates are low (McKinsey estimates fewer than 15% of AI pilots scale into meaningful production). If usage and monetisation do not scale, then the huge infrastructure investments could under-perform.

Chip rental yields weakening

An example: rental prices for Nvidia’s A100 chips have fallen from ~US$2.40/hour to ~US$1.65/hour in just a few years; older chips now being rented for as little as US$0.40/hour below breakeven for many operators. Stranded hardware is a clear risk. If utilisation doesn’t meet expectations, the investments become stranded assets just like the fibre-optic and rail booms of previous eras.

Valuation disconnect

Valuations remain elevated. AI-leaders are trading at ~35× forward earnings, better than the dot-com era (Internet stocks had ~60×+), but still leave little room for disappointment. These valuations presuppose wide thick growth, dominant market share and monetisation, none of which are guaranteed.

A whole investment firm of one.

Investing doesn’t have to be that hard.

Access stocks, ETFs, and more. Oh, and no commission fees. That’s right. Zero. Nada. Zilch. Your first stock is even on us.

*Conditions apply

Get started

How cross-share-holding and circular purchases inflate the story

Here’s how the “Ponzi-ish” mechanics manifest in concrete transactions:

Chip-maker → AI lab: Nvidia commits up to US$100 billion investment into OpenAI; OpenAI then agrees to purchase millions of Nvidia chips.
AI lab → Chip-maker: OpenAI agrees to buy tens of billions of AMD chips, while AMD grants OpenAI warrants equating to ~10 % of its shares.
Cloud/data-centre → AI lab / Chip-maker: Amazon invests in Anthropic; Anthropic commits to using Amazon’s chips and cloud services—Amazon thereby funds a company which uses Amazon’s services, boosting Amazon’s ecosystem.
Inter-firm stakes & supply-commitments: A chip-maker owns a stake in a data-centre provider which is a major customer of its GPUs; the data-centre provider uses those GPUs to service the AI lab; the AI lab has other deep ties to both.

In each case the press release emphasises “billions of dollars” in investment or product commitments. The market hears: “This must be huge growth.” Share-prices rise. But much of the growth is internal to the loop. It’s not always new end-customers buying new services, it’s companies within the ecosystem buying from each other.

Because of the internal loop structure, it’s hard to disentangle real organic demand (outside the loop) from self-referential demand (inside the loop). When revenue is secured via commitments from a sister entity, the longevity and authenticity of that revenue is more questionable.

Why the “revolution” narrative helps conceal risk

The AI revolution narrative, transforming business, medicine, science, productivity, helps shore up sentiment and valuations. Investors want to believe in a transformative wave. But that very optimism helps mask the built-in fragility of circular investment structures.

Press coverage tends to focus on bold numbers (“AI infrastructure spending to reach US$5.2 trillion by 2030”).
Management teams frame multi-year commitments as proof of future demand, even when they depend on sister firms’ purchase decisions.
Investors, flooded with hype, may pay more attention to headline deal-size than margin sustainability.

This makes for a potent combination: bold vision + inter-linked deals + internalised revenue flows = a strong upward spiral of sentiment. But just like a Ponzi-type structure, if new organic external demand falters, the feedback loop may unwind.

So: Is it really a Ponzi scheme? Not legally but structurally analogous

It’s important to stress that the AI industry as currently structured is not necessarily engaged in illegal activity. These are legitimate enterprises, operating under regulation, announcing valid partnerships, making real hardware and software. What is comparable to a Ponzi scheme is the circular financing, mutual share-holding, and self-referential revenue flows that create an illusion of growth and demand more than guarantee it.

In other words: this industry may be experiencing a Ponzi-type dynamic, one in which the system grows on the assumption that the next wave of demand will materialise, and where internal momentum drives valuations ahead of actual long-term, sustainable earnings.

That’s right—earn 3% cash back on all categories.

It’s the only credit card you’ll need, and it’s exclusively for Robinhood Gold members.

What could go wrong, and what investors should watch

Here are key risks:

Demand fall-off or monetisation short-fall: If AI adoption or revenue growth falls short, the infrastructure investments may become under-utilised, turning into stranded assets.
Power & infrastructure bottlenecks: Many projects require grid, nuclear or turbine builds. Delays or cost-blowouts would erode business cases.
Margin squeeze: If chips/rentals decline in price (as they already have), profitability could materially lag build-out cost expectations.
Valuation shock: If the narrative stalls, companies valued at high multiples may face sharp corrections.
Systemic unwind: Because the ecosystem is highly connected, a setback at one node (e.g., AI lab fails to scale) could ripple across chip-makers, cloud providers and data-centres.

Investors should pay extra attention to who is really buying from whom (external customers vs sister-company deals), how much of the demand is public and verifiable, and whether revenue commitments are backed by cash flows or simply paper commitments.

For example: If a chip-maker books US$20 billion of orders from an AI lab in which it holds equity, how much is actual external end-market demand? If a cloud-provider commits to support an AI lab in which it has invested, how much is optional vs binding? These kinds of disclosures matter.

Balanced conclusion: fundamentals exist, but caution is warranted

While there are very real fundamentals driving parts of this wave, the growing performance of AI chips, headroom in infrastructure, genuine enterprise interest, these are intermingled with increasingly complex transaction structures, internal deals, and ambitious-but-uncertain roll-outs. The AI industry may thus be at a turning point: whether it becomes a meaningful revolution or a bubble dependent on self-referential growth will depend largely on execution, external end-user demand, and real monetisation.

In other words, yes, there are legitimate bets being made. But from the vantage point of an investor or observer, the structure carries features reminiscent of a pyramid: circular flows, mutual investments, reliance on the next tier of growth to justify the previous tier’s valuations.

It might still turn out spectacularly well. One of the firms may dominate and capture enormous value, thereby justifying the valuations. But owning the entire ecosystem? That may not play out so neatly. As one observer put it: if you invested in all of the search-engines in the 1990s you would have under-performed because only one (Google) dominated. The AI industry might follow a similar winner-take-most outcome, not a broad-based unlimited growth wave.

product hero invest 91b9077cf4788b508a013b9dda8c3ffe4d4fff969655c212a0201c9533237d46

Commission-free Stock Trading & Investing App | Robinhood

Trade on your time and your terms We have 24/7 Support to help with that too. Oh, and no commission fees. Your first stock is even on us.

Final thoughts

If you search for the phrase “AI Ponzi scheme” it’s easy to label this wave as hype-driven and structurally shaky. And indeed, many of the mechanics are concerning: inter-investments, circular purchase commitments, internal loops of capital, headline deal-size trumping underlying revenue. Yet the companies at the centre of this wave also have strong capabilities, impressive products, and sizeable market opportunities.

The prudent view? Treat the AI boom with respect, but also with scepticism. Don’t assume that all announced deals will convert into profitable business; don’t assume that every dollar of infrastructure build-out will generate sufficient utilisation; don’t assume that mutual investing among ecosystem players is a sign of independent value creation rather than circular momentum.

In short: the AI industry may be resembling a Ponzi structure in its mechanics, but whether it ultimately proves to be one depends on whether the underlying demand, monetisation and external growth materialise. Until then, investors and observers should keep one eye on the hype, and the other on the actual customer deals, external revenues and deployable power-and-utilisation metrics.

______________________