AI titans' ‘circular deals' are starting to look like ‘daisy chains'

The 'circular deals' among AI titans increasingly resemble the 'daisy chains' of the 1980s savings-and-loan crisis. Forty years ago, interconnected transactions obscured the dangers, multiplied systemic risks and helped inflate asset values before roughly a third of US thrift banks failed. Today's financial engineering – an incestuous ecosystem of interlocking multi-year commitments to provide financing, buy semiconductors, secure gigawatts of power and lease data centres – could meet a similar fate, especially if revenue fails to outpace costs or a black swan event eviscerates artificial intelligence (AI) trajectories. OpenAI has struck deals with Nvidia , CoreWeave and others ahead of an initial public offering planned for next year while competitors Anthropic and xAI have pursued similar agreements. Anthropic's US$35 billion cloud-computing deal with provider Lambda is the latest example. Last month, the AI developer signed a US$45 billion arrangement with another Nvidia-backed neocloud, Nscale, to rent data-centre capacity. Nvidia is involved in more than US$750 billion worth of AI investments, financing deals and partnerships, according to PitchBook. Demand from AI labs receiving Nvidia financial support will account for roughly a quarter of the company's business next year, CFO Colette Kress said. The rash of AI investments is unprecedented in scale – 4.5 times the level three years ago, according to Bank for International Settlements data. This raises a question: how much of the industry's extraordinary growth is being financed by the same companies benefiting from it? Meanwhile, hidden risks are colliding with a borrowing binge well under way. The five largest hyperscalers carry roughly US$1.65 trillion in debt through special purpose vehicles and other off-balance-sheet structures, exceeding the US$1.35 trillion they report directly, according to Nikkei.