The Hidden Bubble Inside the AI Boom: Circular Money, Power Limits, and Investor Risk
These days, whether you look at the stock market or the headlines, everything seems to revolve around AI. Massive amounts of capital are flowing into the sector, fueled by the belief that artificial intelligence will reshape the world.
But behind this dazzling narrative lies a far less discussed reality: a circular financing structure that resembles a snake eating its own tail. Companies invest in one another, then use that same capital to purchase each other’s products.
Today, let’s take a closer look at the warning signals embedded in the current capital flows of the AI industry, and what they mean for investors.
1. The Reality of AI’s Self-Perpetuating Circular Financing
A peculiar phenomenon has emerged in the AI ecosystem: circular financing.
In simple terms, NVIDIA invests in companies that buy its chips, such as CoreWeave, and those companies then use the investment capital to purchase even more NVIDIA hardware.
In the short term, this inflates sales figures and boosts market expectations. Some experts compare it to plugging an extension cord into itself.
It may look like a perpetual motion machine, but without real profits flowing in from outside the system, the loop eventually breaks.
At the center of this structure are companies like OpenAI and NVIDIA.
- OpenAI
Signed a $30 billion cloud partnership with Oracle and a $10 billion custom chip deal with Broadcom - Amazon
Invested $8 billion in Anthropic, then required the use of AWS and its proprietary chips - Google
Invested $3 billion in Anthropic while securing TPU usage contracts
2. Echoes of the Past: Japan’s Keiretsu and Korea’s Chaebols
This kind of interdependent investment structure is not new. It closely mirrors Japan’s postwar keiretsu system and Korea’s chaebols.
These systems were originally designed to stabilize supply chains and ensure survival in capital-scarce environments.
Their fatal flaw, however, was risk opacity.
When Japan’s asset bubble burst in the 1990s, complex cross-shareholdings made it nearly impossible to isolate and liquidate bad assets.
Today’s AI giants appear strong on the surface, but they are quietly constructing a fragile ecosystem where the entire structure could wobble the moment fresh capital stops flowing in.
3. The Gap Between Infrastructure Investment and Profitability
The scale of capital expenditure required by the AI industry today is almost unimaginable. According to estimates from firms like McKinsey and Bain:
| Category | Projection | Timeline |
| AI infrastructure investment | ~$5.2 trillion | By 2030 |
| Total data center spending | ~$7 trillion | Within 5 years |
| Annual revenue required to justify investment | ~$2 trillion | Ongoing |
| OpenAI’s current annual revenue | ~$13 billion | As of now |
This is a critical point investors must understand.
To justify current investment levels, the industry needs around $2 trillion in annual revenue. Even the most advanced players are nowhere close. OpenAI, despite its prominence, remains heavily cash-burning.
4. The Hidden Bottleneck: Power Shortages
Chips may arrive on time. Power likely will not.
OpenAI’s Stargate project alone is estimated to require around 10 gigawatts of electricity, roughly equivalent to ten nuclear power plants.
- Typical US nuclear plant construction time: over 10 years
- Current reality: data centers are expanding faster than grid connections
- Interim solution: some operators are installing on-site gas power facilities
For investors, this sends a clear signal.
The long-term value of power infrastructure, copper, and energy-related assets may rival that of chip manufacturers themselves.
5. Market Stress Signals: Falling GPU Rental Prices
Early signs of strain are already visible.
Rental prices for NVIDIA’s latest B200 GPU have fallen from $3.20 to $2.80 per hour within months. Older A100 units are renting for as little as $0.40 per hour, below break-even for many operators.
*Note: TrendForce [News] Why GPU Rental Prices Keep Falling
If demand fails to keep pace with supply, massive data centers risk becoming underutilized assets, much like the unused fiber-optic networks left behind after the early-2000s telecom boom.
⚡ Investment Insights
- Understand the difference in fundamentals
This is not a replay of the 1999 dot-com bubble. Today’s Big Tech firms generate real cash flow. The outcome is more likely to be extreme polarization between companies with genuine profitability and those selling only future promises. - Beware winner-take-all assumptions
If AI models become commoditized through open-source or low-cost alternatives, as recent developments suggest, the valuation of pure model builders could collapse. The real winners may be ordinary businesses that use AI to meaningfully improve productivity. - Focus on physical constraints
As digital competition intensifies, the value of power generation, cooling systems, and energy infrastructure becomes increasingly durable. Often, the most defensible investments lie beneath the software layer.
Actionable Steps for Investors
- Review your portfolio and limit exposure to AI companies dependent solely on external funding
- Prioritize real profit generation over price-to-sales multiples
- Study the energy sector carefully, as power, not chips, is emerging as the true bottleneck
The future belongs to those who can distinguish substance from speculation. Those who learn to see value even inside a bubble may ultimately emerge stronger from the AI revolution.
