Top Shot – Employees work on a circuit breaker production line at an electronics company’s factory in Fuyang, eastern China’s Anhui province, on January 16, 2024. (Photo credit: AFP)/China Out (Photo credit: STR/AFP, Getty Images)
AFP (via Getty Images)
Traditional boundaries between venture capital and private equity are beginning to melt, creating a hybrid investment ecosystem that may amplify systemic risks in the technology sector. This convergence, coupled with what economists call “cyclical lending” among large AI companies, could create a situation reminiscent of previous market bubbles, with far-reaching effects across the economy.
Investment boundary breakdown
Historically, venture capital and private equity have operated in different spheres. Venture capital focused on early-stage, high-risk investments with the promise of exponential returns, while private equity typically acquired mature companies with established cash flows. Today, these lines are becoming significantly blurred, especially in the field of AI.
Private equity, private credit, and asset management institutions such as Blackstone, Pimco, Magnetar Capital, Apollo, Blue Owl Capital, Carlyle, and BlackRock are currently issuing bonds to data center developers, a traditionally venture-backed space. As of early 2025, private bond funds had lent about $450 billion to the technology sector, an increase of $100 billion from early 2024.
This shift represents a fundamental shift in the flow of capital in the technology ecosystem. Blue Owl Capital’s involvement in Meta’s $29 billion data center deal exemplifies this trend, as it is a private credit company that provides infrastructure financing typically handled by traditional project finance or venture-backed development.
Circulating loan issue
Perhaps the most worrying development is what financial analysts call “roundabouts,” or circular lending, a phenomenon in which big technology companies simultaneously invest in and acquire each other, creating a tangled web of dependencies.
OpenAI’s deals with NVIDIA, AMD, and Oracle exceed $1 trillion. The complexity of these arrangements is staggering.
Oracle is spending $40 billion to buy NVIDIA GPUs to power the data centers it leases for 15 years to support OpenAI, and OpenAI is paying Oracle $300 billion over the next five years. NVIDIA has invested $100 billion in enAI, which will be repaid over time through OpenAI’s lease of NVIDIA’s GPUs. $11.9 billion over 5 years to rent GPUs NVIDIA has committed to buy CoreWeave’s unsold cloud computing capacity in a $6.3 billion order by 2032
Bloomberg reports that one analyst determined that half of NVIDIA’s revenue comes from these companies alone.
Financial Times columnist Bryce Elder captured this absurdity: “Imagine a caravan manufacturer. The caravan manufacturer sells caravans to caravan parks that only buy one type of caravan. The caravan park rents much of its land from another caravan park. The first caravan park has two big customers. One of the big customers is the caravan manufacturer. The other big customer is the caravan manufacturer’s biggest customer.”
The circular loan problem is also known as a “roundabout.” A phenomenon in which large technology companies simultaneously invest in and acquire each other, creating a tangled web of dependencies.
Josipa Majik Predin
Mob psychology: Herd behavior in AI evaluation
The venture capital industry’s traditional “spray-and-pray” investment strategy has evolved into something more concerning: synchronous, momentum-based capital deployment that resembles crowd behavior more than independent risk assessment.
At first, ventures were small, crude, and completely uncertain. Early venture partnerships were structured to produce asymmetric outcomes. In other words, it was a small check into an unproven market where differences matter. A $100 million fund that writes 12 checks of $5 million to $10 million, nine of which lose money, and one that returns $3 billion can become legendary.
That model has fundamentally changed. When someone tells a venture capitalist, “They raised $150 million to invest in an Irish space startup,” people’s faces fall. Because there aren’t enough Irish space startups to absorb this investment. The greater the fund’s growth, the flatter the distribution curve. Return the concentrate. And the whole model moves from voluntary to mandatory.
The concentration of the Magnificent Seven
Market concentration in AI stocks has reached unprecedented levels. The “Magnificent Seven” of tech stocks now account for more than 50% of the S&P 500 index’s market capitalization, according to an analysis by Employ America.
Mag7’s second-quarter profits rose 26% this year, while the rest of the S&P 500 rose just 1%. This concentration creates a dangerous feedback loop. When these stocks rise, the index fund has to buy more, which drives up the price and attracts more money in a self-reinforcing cycle.
cash flow problems
There is a fundamental problem underlying the soaring valuation. The idea is that AI companies are bleeding cash without a clear path to profitability.
Behind-the-scenes calculations using hyperscalers’ public statements and publicly available news sources show that hyperscalers will invest more than $560 billion in AI technology and data centers from 2024 to 2025, with only $35 billion in reported revenue, or no profit at all.
Individual examples are even more stark.
xAI reportedly spends $1 billion a month, but is expected to make only $500 million this year Recent forensic accounting suggests that OpenAI lost $11.5 billion between July and September 2025, or $3.8 billion per month
Sequoia Capital’s analysis found a $500 billion gap between the revenue projections implied by technology companies building AI infrastructure and the actual revenue growth in the AI ecosystem. Bain’s report states that the expanding trend in the AI sector requires $2 trillion in new revenue, more than five times the size of the existing software subscription market.
Hybrid model of venture and private equity
The fusion of VC and PE approaches has created a new financial structure that spreads risk in an opaque way. Hyperscalers can use sale/leaseback transactions to commission and build single-tenant data centers from their balance sheets.
Under Meta’s $29 billion deal with Blue Owl Capital and PIMCO, Meta will spend on data center site preparation through an SPV co-owned with Blue Owl Capital (20% owned by Meta and 80% owned by Blue Owl), and the SPV will secure debt from PIMCO. Once the data center is completed, Meta plans to lease it back from the SPV as the sole tenant for a period of 15 to 20 years.
The structure, Bloomberg reported, allows companies to raise billions of dollars in debt without showing it on their balance sheets, making leverage levels appear lower than they actually are.
Side issue: GPU depreciation
Adding to these concerns is the rapid depreciation of the sector’s key asset, the graphics processing unit (GPU). H100 GPUs once cost $30,000 and had rental fees as high as $8 per hour, but they can now be rented for as little as $1 per hour.
GPU market leader NVIDIA has traditionally released new GPU models every two years, but now plans to release new GPUs on a one-year cycle. This accelerating obsolescence creates a dangerous situation for companies using GPUs as collateral for loans.
Analysts at Barclays believe a drop in GPUs poses enough risk to be worth cutting profit estimates for Google (Alphabet), Microsoft and Meta by as much as 10%, arguing that consensus modeling significantly underestimates the amount of profit amortization required.
Minsky Framework: Understanding Bubble Dynamics
Financial economist Hyman Minsky’s work on financial instability provides a useful framework for understanding current dynamics. Mr. Minsky identified a “Ponzi Finance” sector as one in which “operating cash flow is insufficient to meet either principal repayments or interest on outstanding debt.”
In the short term, the AI sector’s cash flow is simply insufficient to service its debt, and as the AI sector grows, this situation is unlikely to change in the short term.
This sector exhibits the classic bubble characteristics identified by researchers at State Street Associates: a combination of asset centrality (where one sector generates returns across the broader market) and high relative valuations.
Policy implications and systemic risks
Systemic risk extends beyond Wall Street. Data center capital spending is currently the main driver of US economic growth, with AI investment accounting for more than 40% of US GDP growth this year, according to an analysis by economist Ruchill Sharma.
Stock ownership accounts for about 30 percent of American households’ net worth, and we’re all incredibly exposed to the data center boom. The concentration of market profits in AI stocks means that pension funds, 401(k)s, and index funds are heavily weighted toward these companies.
JPMorgan notes that AI-related bonds are now the largest portion of the investment-grade bond market ($1.2 trillion), with AI-focused companies accounting for a higher share of the investment-grade bond market than U.S. banks, as reported by Bloomberg.
Government backstop problem
The intertwining of the AI sector and national industrial policy creates further complexity. In early November, OpenAI’s chief financial officer (CFO) indicated that the company was willing to receive federal “guarantees” on GPU purchases. Both the federal government and OpenAI CEO Sam Altman have emphasized that there are no plans for a federal fiscal backstop, although OpenAI has previously applied for loan guarantees from the federal government.
The AI Action Plan released by the White House directs federal agencies to support AI innovation, rapid energy access to data centers, and streamlined construction, creating what some economists call a “venture-state symbiosis.”
Historical similarities and warning signs
The current situation has unpleasant similarities to previous bubbles. The data center boom is now bigger than previous technology booms, rivaling the housing bubble as a share of GDP, although the railroad and dot-com bubbles are most often cited.
However, unlike previous tech bubbles, this one will have an immediate impact on the entire economy. Employ America’s analysis emphasizes that there is no offsetting economic dynamism to cushion the effects of a near-term market correction, saying that “the losses from a tech sector collapse will be too large and too rapid to be offset cleanly by an impending symmetrical boom elsewhere.”
Conclusion: Systemic vulnerabilities
The convergence of venture capital and private equity in the AI space, combined with circular funding arrangements and unprecedented market concentration, has created a financial structure that amplifies rather than spreads risk. The mob drive for AI valuations has little regard for traditional metrics such as cash flow, profitability, and sustainable business models.
The stock valuation of these companies is your ticket to financing capital investment. If the stock market is not active, short-term corporate bond spreads will rise, and long-term investment and SPV debt spreads will certainly start to rise as well.
The question is not whether these financial arrangements are sustainable; Cash flow data shows it’s not sustainable. The question is what happens when the music stops, and whether the public sector will be asked to backstop what has been a major driver of economic growth in the United States. As the economist John Maynard Keynes warned, “When a nation’s capital development becomes a byproduct of casino activity, the job is likely to go awry.”
