Ahead of IPOs, AI Giants Keep Making Circular Deals. Here’s Why That’s a Risk
As OpenAI continues to gear up for an IPO expected this year, the unusual deals between artificial intelligence companies and their suppliers are growing in importance as questions about revenue streams come to the forefront.
Rather than AI suppliers simply paying their vendors for products and services, these so-called circular arrangements involve vendors making payments back to—or taking stakes in—AI companies. OpenAI has engaged in such deals with Nvidia NVDA, CoreWeave CRWV, and others. Competitors Anthropic and xAI have made similar agreements, in which corporate investors provide capital in exchange for commitments of future computing resource contracts.
The pattern has been on full display over the past six months. In March, OpenAI closed a historic $122 billion funding round, with Amazon AMZN committing up to $50 billion and Nvidia $30 billion. OpenAI also expanded its existing $38 billion multi-year agreement with Amazon’s cloud computing unit AWS by $100 billion, and it committed to deploy next-generation Nvidia chips.
Three months earlier, Microsoft MSFT and Nvidia jointly committed $15 billion to Anthropic, which in turn pledged to spend $30 billion on Microsoft’s cloud unit, Azure.
Ahead of expected IPOs from OpenAI and Anthropic, revenue quality is likely to be one of the key metrics investors will scrutinize, according to PitchBook senior analyst Harrison Rolfes.
With The Wall Street Journal reporting that OpenAI has missed internal growth targets, investors are questioning the assumption that AI labs will grow revenues large enough to meet their obligations. OpenAI CFO Sarah Friar has warned that if revenue doesn’t accelerate, the firm could fail to pay for future computing contracts. (OpenAI has disputed the Journal’s reporting.) If that happens, it could significantly impact companies that have been banking on the AI giant’s growth.
“The revenue miss matters more than the market is pricing in—OpenAI lost ground to Anthropic and Google in coding and enterprise earlier this year, and its infrastructure obligations now exceed $1.15 trillion across Oracle, Microsoft, [and] Amazon,” explains Rolfes. “The question now is whether the growth can stay ahead of the cost curve. The deeper structural problem is that the obligation stack isn’t just debt; it’s circular. The same hyperscalers writing the equity checks are also the vendors collecting the compute bills.”
Because of the structure of these deals, Rolfes says, the risks are not even. He points to the deal wherein Microsoft has a stake in OpenAI, takes 20% of its revenue, and is owed years of Azure commitments. “That’s not a partnership with aligned incentives, but a fixed claim on your upside, regardless of how your margins perform. When revenue misses, the hyperscaler doesn’t absorb the pain alongside you; they get paid first.” It’s a similar story, Rolfes says, with Alphabet’s $40 billion investment in Anthropic, alongside the multi-gigawatt TPU deal involving specialized chips for Google.
“These deals made sense when the priority was scale at any cost,” Rolfes explains. “At the point where margin trajectory becomes the primary valuation driver, the circularity stops being a feature.”
