“Circular lending” has become a loaded phrase in 2025, typically hurled at large technology companies (especially those in the artificial intelligence sector) by investors arguing that the stocks of those companies are in a bubble and that bubble will soon burst. The argument from the bubble-callers is that circular lending (sometimes called “circular funding”) is juicing the sales of each other’s companies, as if that by itself would be an indictment. However, the question an investor should ask when a bubble-caller condemns circular lending—often citing the term with a snide tone—is “so what?”
Strip away the unearned stigma and you will find an old playbook of vendor financing being rerun at a modern scale. The benefit for investors is a capital-expenditure cycle that can sustain innovation and earnings power.
Recent circular lending examples driving today’s AI build-out
- Microsoft and OpenAI. Microsoft’s multi-year, multibillion-dollar partnership with OpenAI (the parent company of the large language model ChatGPT) was formalized in 2023, with OpenAI committing heavily to Azure (Microsoft’s cloud computing platform) for training and product workloads.
- Amazon and Anthropic. Amazon invested up to $4 billion in Anthropic (the parent company of the LLM Claude), and Amazon Web Services (AWS) became Anthropic’s primary cloud provider.
- Google and Anthropic. Google’s financing includes a multi-billion-dollar commitment by Anthropic to utilize Google’s cloud computing services.
- NVIDIA and CoreWeave. NVIDIA, a supplier to and shareholder of CoreWeave, agreed to a $6.3 billion order from CoreWeave through 2032, effectively backstopping CoreWeave while securing future demand for its hardware. (CoreWeave builds hardware for AI models, using NVIDIA’s graphics processing units, or GPUs.)
- Oracle and Cohere (and others). Oracle invested in Cohere and promoted training/deployment on Oracle’s cloud computing platform. Oracle and NVIDIA deepened their infrastructure ties to put thousands of next-generation GPUs onto Oracle’s cloud. Oracle also struck a deal under which OpenAI would purchase hundreds of billions of Oracle’s compute resources over several years.
- OpenAI and AWS. OpenAI will run workloads on AWS infrastructure, tapping hundreds of thousands of Nvidia’s GPUs. The deal opens up the potential for OpenAI to have access to Amazon’s custom-built Trainium chips, which Anthropic is currently using.
Critics cite the circularity as if it were a bug, but it is actually a feature. It is how big platforms seed ecosystems, and it is nothing new. Businesses have funded customers and suppliers for decades; what is new is only the size of it all.
Vendor financing did not begin in Silicon Valley. General Motors established GMAC (General Motors Acceptance Corporation, GM’s lending arm; it was later renamed Ally Financial) in 1919 to finance dealer inventory and consumer purchases, thereby accelerating the distribution of cars and increasing GM’s market share. Sears popularized installment plans, enabling households to purchase big-ticket appliances and catalyzing the mass-market adoption of these items. IBM formalized IBM Credit (now part of IBM Global Financing) in the early 1980s, allowing clients to finance mainframes and software, thereby keeping upgrade cycles alive even when budgets were tight.
The naysayers point to the dot-com/telecom buildout of the late 1990s, when vendor financing by equipment makers led to overbuilding networks for what was needed at the time. It did not take long for the demand to match the overbuilt supply; it was not so much the existence of the infrastructure that was the problem, but rather that stock prices had gotten way ahead of themselves. Are stock prices ahead of themselves today? Yes, but it is not in the bubble territory reached in the late 1990s.
The technology-heavy Nasdaq stock index rose 400 percent between 1995 and 2000, with a price-to-earnings (P/E) ratio of 73. Today, the Nasdaq’s P/E ratio is approximately 31, and over the last four years, the index has risen by about 100 percent, which is roughly the same level as it has been since 2013.
Today’s companies are better run than those from a quarter century ago, which means they deserve higher P/Es now. Total corporate profits in 1999 were approximately $750 billion; by 2024, they had increased to $3.8 trillion. The average operating margins of the five biggest companies in 2025 are 40 percent better (39.9 percent versus 28.4 percent) than those of their 1999 cohorts. Net margins are 65 percent better (34.2 versus 20.9). Return on equity is 2.3 times better (66.5 percent versus 28.1 percent). And the average revenue and net income of today’s big five are 4.6 and 9.1 times higher, respectively, than the inflation-adjusted results of the big five of 1999.
More broadly, profit margins for the aggregate companies of the S&P 500 index have been increasing over the decades. The margins for those companies were 5.6 percent in the 1990s, 6.3 percent in the 2000s, and 8.8 percent in the 2010s. Profit margins have averaged 10.3 percent in the 2020s and currently stand above 11 percent.
Year to date, the S&P 500’s roughly 14 percent rise has been driven chiefly by earnings growth (eight percentage points) and dividends (one percentage point), with just five percentage points from multiple expansion. Over the last five and half years, the S&P 500 has increased by approximately 125 percent, with roughly 77 points attributed to earnings growth, 19 points to dividends, and 30 points to multiple expansion. That is a fundamentals-led stock market rise, not a speculative melt-up.
Still, I watch everything; too much of a good thing can become a red flag.
What could go wrong?
- Demand mirage. Critics say circular funding can blur “real” demand with financed demand. (That is fair, but in the near term, an overbuild risk feels more like a concern for 2027 than 2026. I frequently use and discuss AI, and while it has been widely adopted, there is still more potential for improvement and adoption to come.)
- Smaller balance sheets are at risk. Second-tier AI infrastructure players that incur capital expenditures without matching cash flows are vulnerable if a hyperscaler (e.g., Meta, Alphabet, Microsoft) reduces its capital expenditure.
- Financing complexity. Rapid growth, combined with bespoke loans, can create technical defaults even without missed payments. For example, CoreWeave triggered technical defaults on a $7.6 billion loan facility after breaching specific terms. Complexity is a feature of fast cycles, but it makes it more challenging for a company’s administration to keep up.
Why the net effect is bullish for investors
First, this financing aligns incentives. When a platform invests in or backstops a company, it underwrites future consumption of its own compute resources, software, or chips. That increases visibility into capacity utilization and smooths the capital expenditure curve.
Second, it accelerates diffusion. IBM showed decades ago that embedded financing shortens sales cycles and broadens adoption. The same logic applies when an LLM startup can scale training new users because its cloud partner is an owner or financier.
Third, it strengthens the ecosystem. Consider NVIDIA’s pact with CoreWeave: It de-risks CoreWeave’s buildout and secures long-dated demand for NVIDIA’s GPUs. For equity holders, this can lengthen the runway for earnings as margins normalize.
There are real risks to the stock market (inflation, labor, trade policy, U.S. debt, valuations, etc.); however, circular funding is a structural catalyst, and not a red flag by default.
Allen Harris is an owner of Berkshire Money Management in Great Barrington and Dalton, managing more than $1 billion of investments. Unless specifically identified as original research or data gathering, some or all of the data cited is attributable to third-party sources. Unless stated otherwise, any mention of specific securities or investments is for illustrative purposes only. Advisor’s clients may or may not hold the securities discussed in their portfolios. Advisor makes no representation that any of the securities discussed have been or will be profitable. Full disclosures here. Direct inquiries to Allen at AHarris@BerkshireMM.com.
