Is AI Circular Financing Inflating a Bubble?
Category: Pricing & Economics · Duration: 25 min · ▶ Watch
Speakers: Patrick Boyle
Segments (13)
- 00:00:00 · Introduction: The AI Spaghetti Diagram
- Major AI companies are engaging in massive, circular investments with their suppliers and customers, creating a complex web of financial interdependence.
- 00:00:21 · The CoreWeave IPO: An ‘IPOurorobos’
- The IPO of CoreWeave, a major Nvidia customer, was anchored by Nvidia itself, illustrating a self-fueling financial loop likened to a snake eating its own tail.
- 00:01:26 · The Central Players: OpenAI and Nvidia
- OpenAI and Nvidia are at the heart of this circular financing, making huge deals to ensure the success of their entire ecosystem, from suppliers to customers.
- 00:01:54 · A Web of Mega-Deals
- Details of multi-billion dollar deals are outlined, including OpenAI with Oracle, Broadcom, and AMD, and Nvidia’s massive investment pledge to OpenAI.
- 00:03:12 · Amazon and Google Join the Loop
- Amazon and Google are also creating their own investment loops, primarily with AI company Anthropic, providing capital in exchange for cloud service commitments.
- 00:04:49 · Elon Musk’s Inter-Company Dating
- Elon Musk’s companies (xAI, X, Tesla) are structured to invest in and provide data to each other, creating another form of circular, self-reinforcing ecosystem.
- 00:05:38 · The Mobius Strip of Capital and Electricity
- The current AI boom is characterized as a loop of venture capital and electricity, with the latter being a very real and growing constraint.
- 00:07:07 · The Scale of Spending vs. Revenue
- Trillions in projected capex for AI infrastructure far exceed current revenues from AI companies, raising questions about who will ultimately pay for it all.
- 00:07:45 · Historical Parallels: Keiretsu and Chaebol
- The AI industry’s structure is compared to Japan’s Keiretsu and South Korea’s Chaebol systems, which involved complex cross-holdings to secure supply chains.
- 00:09:38 · Signs of Stress: The GPU Rental Market
- Falling rental prices for GPUs, including Nvidia’s latest chips, suggest the market may be showing early signs of stress and oversupply.
- 10:45:00 · Leveraged on Optimism and the Winner-Take-All Bet
- The entire system appears leveraged on optimism, with investors betting on a ‘winner-take-all’ outcome, but the ease of model replication could challenge this.
- 11:58:00 · The Electricity Constraint
- The immense and growing demand for electricity is a major, often overlooked, physical constraint on the AI buildout, with permitting and construction of new power sources lagging far behind.
- 12:54:00 · Conclusion: A Fragile but Different Bubble?
- While the situation resembles a bubble, the underlying companies are financially stronger than in past cycles, but the outcome remains highly uncertain, with profitability and demand as key questions.
Specific Prices (22)
| Timestamp | Item | Value | Context |
|---|---|---|---|
| 00:00:06 | OpenAI computing power purchase from Oracle | $300 billion | A recent agreement for OpenAI to buy computing power from Oracle over about five years. |
| 00:00:24 | CoreWeave net loss (last year) | $863 million | Despite soaring revenue, the company reported a significant net loss, highlighting its capital-intensive nature. |
| 00:00:44 | Nvidia’s anchor investment in CoreWeave IPO | $250 million | Nvidia placed a $250 million order to anchor the IPO of its customer, CoreWeave, at $40 per share. |
| 00:01:39 | Nvidia market value | $4.5T | Shown on a Bloomberg diagram illustrating the AI money machine. |
| 00:01:39 | OpenAI market value | $500B | Shown on a Bloomberg diagram illustrating the AI money machine. |
| 00:01:56 | OpenAI cloud infrastructure agreement with Oracle | $300 billion | A major deal for cloud infrastructure. |
| 00:02:02 | OpenAI custom chip partnership with Broadcom | $10 billion | A partnership for developing custom chips. |
| 00:02:18 | Nvidia’s pledged investment in OpenAI | Up to $100 billion | Nvidia pledged this amount to invest in OpenAI, which in turn will buy Nvidia’s chips. |
| 00:02:31 | OpenAI’s chip purchase from AMD | Tens of billions of dollars | OpenAI agreed to buy chips from AMD, and in return, AMD gave OpenAI the right to buy 10% of its stock. |
| 00:03:16 | Amazon’s investment in Anthropic | More than $8 billion | Amazon invested in Anthropic, which in turn committed to using Amazon’s cloud services. |
| 00:03:52 | Google and Anthropic cloud deal | Tens of billions | A deal for Anthropic to access Google’s TPUs. |
| 00:04:58 | Valuation of X (Twitter) in sale to xAI | $33 billion | Elon Musk’s AI startup xAI acquired his social media firm X at this valuation. |
| 00:07:08 | McKinsey forecast for AI-related capex by 2030 | $5.2 trillion | Projected spending on chips, data centers, and energy over the next five years. |
| 00:07:17 | Bain estimate for annual revenue needed to justify capex | $2 trillion | The annual revenue from AI companies needed to justify the massive infrastructure spending. |
| 00:07:26 | OpenAI’s full-year revenue and cash-burn target | $13 billion revenue, $8.5 billion cash-burn | Showing that the company is not profitable despite significant revenue. |
| 00:07:32 | Anthropic’s run-rate revenue by August 2025 | Over $5 billion | Demonstrating rapid growth but still a fraction of the required industry-wide revenue. |
| 00:09:44 | Nvidia B200 chip rental price | $2.80 per hour | The price has dropped from $3.20, indicating potential market stress. |
| 00:09:52 | Nvidia A100 chip rental price | $0.40 per hour | An older chip now available for a price below the break-even point for many operators. |
| 10:06:00 | Cost of an 8-chip A100 cluster in 2020 | $199,000 | Used in a break-even calculation by the Financial Times. |
| 10:06:00 | Break-even rental rate for an 8-chip A100 cluster | About $4 an hour | The FT calculates this is the rate needed to break even, which is higher than current market rates for the chip. |
| 11:42:00 | Projected data center spending over the next 5 years | Almost $7 trillion | McKinsey’s total estimate for both AI and traditional data center capex. |
| 12:20:00 | OpenAI’s revolving credit line | $4 billion | Secured from a consortium of banks to fund its infrastructure build-out, an unusual move for a cash-burning tech startup. |
Predictions (4)
- [00:07:08, by 2030] The AI industry will require $5.2 trillion in capex for chips, data centers, and energy by 2030.
- [00:09:24, Unspecified] If demand for AI infrastructure doesn’t materialize, the fallout will impact not just startups but also lenders, landlords, and utilities.
- [10:56:00, Unspecified] Pandemic-era GPUs may end up in liquidation, never having earned back their cost.
- [12:00:00, Long-term] AI could end up boosting productivity across the economy, while the AI labs themselves struggle to monetize and profit.
Companies Mentioned (18)
OpenAI · Microsoft · Nvidia · CoreWeave · AMD · Oracle · Intel · Broadcom · Amazon · Anthropic · Google · xAI · X (formerly Twitter) · Tesla · DeleteMe · McKinsey · Bain & Company · Toyota
Notable Quotes (5)
Bryce Elder described the deal in the FT at the time as an ‘IPOurorobos’, an ancient symbol of a snake or a dragon eating its own tail. — Patrick Boyle @ 00:00:51
A similar metaphor might be an extension cord plugged into itself. If you don’t know much about electricity, that might look like a perpetual energy machine, but trust me, I’ve tried it out… it won’t power your home appliances. — Patrick Boyle @ 00:01:01
Okay. Why don’t we pay you cash for the value of the chips, and you give us back stock, and when we announce the deal the stock will go up and we’ll get our $78 billion back. — Matt Levine (paraphrased by Patrick Boyle) @ 00:02:49
The whole thing is starting to look less like a tech boom and more like a Mobius strip made of venture capital and electricity. And the electricity part isn’t a metaphor either. — Patrick Boyle @ 00:05:39
Investors are now asking whether these interdependencies could pose risks if AI demand or monetization falls short of investor expectations. — Patrick Boyle @ 00:08:01
Key Topics
circular financing · AI infrastructure investment · GPU pricing · data center economics · venture capital · AI bubble · Nvidia economics · OpenAI business model · cloud computing costs · electricity demand · Keiretsu model · capital expenditure (capex) · AI monetization
Takeaways
- Major AI companies like Nvidia, OpenAI, Microsoft, and Amazon are engaged in massive, circular financing deals, where they invest in their own customers and suppliers to fuel growth.
- This self-reinforcing loop, likened to a snake eating its own tail (Ouroboros), inflates revenue projections and makes it difficult to assess true, organic demand.
- The structure resembles historical industrial conglomerates like Japan’s Keiretsu, designed to secure supply chains but also obscuring financial risk and propping up firms.
- While the companies involved are more financially sound than those in the dot-com bubble, the entire system is leveraged on the optimistic belief in a ‘winner-take-all’ market.
- The GPU rental market is already showing signs of stress, with prices falling below break-even costs for some operators, questioning the sustainability of the massive hardware investment.
- A critical, non-financial constraint is the immense demand for electricity, with projected data center needs requiring the equivalent of dozens of new nuclear power plants, a buildout that is not happening.
- The ultimate question is who will pay for the trillions in infrastructure spending, as current revenues are a fraction of the cost, and it’s unclear if the AI model builders or the businesses using the models will capture the profits.