The largest companies in artificial intelligence are financing each other in an increasingly closed loop. Nvidia invests in OpenAI. OpenAI buys Nvidia chips. Nvidia invests in CoreWeave. CoreWeave buys Nvidia chips and provides cloud services to OpenAI. The money circulates, valuations rise, and everyone involved reports impressive growth.
This pattern has a name: circular financing. It’s accelerating an AI infrastructure buildout that may already be unsustainable. And it has historical precedent that should alarm investors.
How circular financing works in AI today
Circular financing occurs when companies invest in each other and then purchase each other’s products or services. It creates a self-reinforcing cycle that can make demand appear stronger than it actually is.
In the current AI ecosystem, the pattern is remarkably concentrated. Nvidia has invested roughly $53 billion across 170 deals between 2020 and 2025, spanning large language model developers like OpenAI and cloud providers like CoreWeave. These companies then use that capital to purchase Nvidia’s GPUs, often committing to billions in future chip purchases as part of the investment terms.
The deals have accelerated dramatically in recent months:
- November 2024: Amazon completes $8 billion total investment in Anthropic; Anthropic names AWS as primary cloud and training partner and will use AWS Trainium chips
- December 2024: Nvidia and AMD participate in xAI’s $6 billion Series C round; funds used to expand Colossus supercomputer with Nvidia hardware
- March 2025: OpenAI signs $11.9 billion contract with Nvidia-backed CoreWeave for GPU cloud capacity; OpenAI receives $350 million in CoreWeave stock
- May 2025: OpenAI expands CoreWeave contract by $4 billion
- July 2025: OpenAI and Oracle enter agreement to develop 4.5 gigawatts of Stargate capacity, a partnership exceeding $300 billion over five years
- September 2025: Nvidia announces up to $100 billion investment in OpenAI, deployed progressively as OpenAI builds at least 10 gigawatts of Nvidia systems
- September 2025: Nvidia agrees to purchase $6.3 billion in CoreWeave cloud services, guaranteeing to buy unsold capacity through 2032
- September 2025: OpenAI expands CoreWeave contract by $6.5 billion, bringing total to approximately $22.4 billion
- October 2025: AMD issues OpenAI warrants for up to 160 million shares at $0.01 per share, vesting as OpenAI deploys up to 6 gigawatts of AMD GPUs
- October 2025: Nvidia invests up to $2 billion in xAI as part of a $20 billion financing round structured through an SPV that purchases Nvidia GPUs
- October 2025: Amazon opens $11 billion Project Rainier data center in Indiana, built exclusively to train and run Anthropic models using 500,000 Trainium chips
- November 2025: Nvidia and Microsoft invest up to $15 billion combined in Anthropic ($10 billion from Nvidia, $5 billion from Microsoft); Anthropic commits to $30 billion in Azure compute purchases and up to 1 gigawatt of Nvidia capacity
- January 2026: Nvidia invests $2 billion in CoreWeave, adding to existing agreements worth over $6 billion in services through 2032
- January 2026: Nvidia-OpenAI $100 billion deal reported stalled after internal Nvidia concerns; CEO Jensen Huang states the investment was never “a commitment”
Why this mirrors the 1920s
The current AI financing structure bears uncomfortable similarities to the investment trusts that proliferated before the 1929 crash.
In the late 1920s, investment trusts became wildly popular vehicles for speculation. According to economist John Kenneth Galbraith in The Great Crash, 1929, the number of trusts exploded from around 160 in 1926 to over 750 by the end of 1929, with a new trust launching roughly every business day that year. These trusts issued $3 billion in securities in 1929 alone, representing a third of all capital issued that year.
The problem was their structure. Trusts invested in each other, creating layers of leverage and interdependence. Goldman Sachs Trading Corporation sponsored Shenandoah Corporation, which sponsored Blue Ridge Corporation. Each trust held shares in its sponsored trust and benefited exponentially when those shares traded at a premium to their underlying assets. One academic analysis described the structure as “highly levered and basically a pyramid scheme.”
The parallels to today are striking:
Then: Investment trusts traded at premiums to net asset value because investors believed in the managers’ ability to generate outsized returns. Now: AI companies trade at valuations that assume decades of continued hypergrowth.
Then: Trusts invested in trusts invested in trusts, creating chains of interdependence. Now: Chip companies invest in cloud companies that invest in AI labs that buy from chip companies.
Then: Everyone involved reported impressive returns, right up until they didn’t. Now: Everyone involved reports impressive growth.
When the 1929 market turned, the leverage worked in reverse. Trusts that had amplified gains on the way up amplified losses on the way down. The crash destroyed roughly 90% of market value by July 1932. The resulting regulations, including the Investment Company Act of 1940, explicitly restricted the practices that had proven so dangerous.
We are not predicting a 90% decline. But we are noting that the structures enabling today’s AI boom look remarkably similar to structures that have failed catastrophically before.
Two problems for shareholders
Circular financing creates two concrete problems for equity investors.
The first problem is that it obscures whether genuine demand exists. The fundamental question for AI is whether end users will pay enough for AI products to justify the infrastructure being built. We don’t know the answer yet. OpenAI is burning billions. Most AI applications aren’t profitable. Bain & Company estimates the industry needs $2 trillion in annual revenue by 2030 to sustain current infrastructure spending, with an $800 billion projected shortfall.
Normally, weak demand would show up in the numbers. Customers stop buying. Revenue slows. Markets self-correct. But circular financing delays that signal. OpenAI can keep buying chips because Nvidia keeps investing in OpenAI. Revenue looks strong. Growth looks healthy. But we haven’t actually answered whether any of this will generate sustainable returns.
The second problem is that Nvidia faces amplified downside. Nvidia isn’t just selling picks and shovels. It owns equity stakes in the miners. The company holds positions in OpenAI, CoreWeave, xAI, Anthropic, and dozens of smaller AI firms.
If AI demand disappoints, Nvidia gets hit twice. Revenue falls because customers buy fewer chips. And its investment portfolio loses value because those equity stakes decline. Same trigger, double the exposure. This isn’t diversification. It’s concentration masquerading as a portfolio.
The companies themselves will probably survive. Microsoft has software. Amazon has e-commerce. Google has advertising. And all three have cloud computing platforms. Their core businesses generate enough cash to service debt and absorb AI losses. Bondholders can sleep well at night.
But equity holders are in a different position. Stock prices don’t just reflect whether a company will survive. They reflect expectations about future earnings. And current prices assume AI growth that may not materialize. Survival and compounding returns are not the same thing.
What this means for index investors
For investors holding S&P 500 index funds, this risk is hard to avoid. The Magnificent Seven now account for more than 34% of the index, up from 12% in 2015. Nvidia alone is one of the largest weights. Owning “the market” increasingly means owning a concentrated bet on AI infrastructure spending continuing indefinitely.
The 1929 crash taught regulators to restrict investment trusts from owning each other in pyramids. We haven’t applied similar scrutiny to the current AI ecosystem. Perhaps we should.
A different approach
At Magnifina, we take a different approach to equity investing. By focusing on individual stock selection with deliberate attention to business fundamentals and valuation, we can avoid the circular dependencies and concentration risks embedded in today’s indexes. We look for companies with clear revenue sources, sustainable competitive advantages, and valuations that don’t require a decade of perfection to justify.
Sometimes the best investment decision is knowing to avoid going all in on certain trends. We believe this is one of those times.
If you’re concerned about AI concentration risk in your portfolio, we should talk.
