The artificial intelligence industry is undergoing a unique phase of transformation: capital is being raised not through venture rounds or IPOs, but through the traditional debt market. Based on my analysis of market flows, I estimate that by 2026, the volume of AI company bonds issued could reach $570 billion. And the main beneficiary of this process is Morgan Stanley.
How the Mechanism Works
By the end of May this year, the volume of so-called "AI debt" had already reached $236 billion — four times more than in the same period last year. In the second half of 2025 alone, Morgan Stanley arranged bond issuance deals for tech giants totaling $65 billion. The result: $2.3 billion in fees over six months, nearly double the $1.4 billion figure from a year earlier. The bank has overtaken Goldman Sachs and now trails only JPMorgan Chase.
Key drivers are Nvidia and the new Chinese neural network Kimi K3 from Moonshot AI. Funding is provided by pension funds and insurance companies, which operate without much fanfare. They structure the credit obligations of IT giants and contracts for computing power, while conservative investors eagerly buy the finished bonds.
Case Studies: TeraWulf and Meta
A telling example is former Bitcoin miner TeraWulf, which repurposed its capacity for neural network needs. A $3.2 billion debt issuance attracted $10 billion in bids at a yield of 7.75% per annum. The high demand is explained by Google's involvement: the tech giant guaranteed $3.2 billion in lease payments if operator Fluidstack defaults. In return, Google received the right to buy 14% of TeraWulf's shares.
Meta went even further. With Morgan Stanley's assistance, the company raised $27 billion for the Hyperion complex in Louisiana. The deal set a record for the private credit market: partner Blue Owl holds an 80% stake, so the obligations do not appear on Meta's balance sheet.
Risks and Prospects
However, the market is starting to show restraint. In February, over-lending to IT giants exceeded supply by five times, but by July, the ratio had fallen below two times. The cost of insuring against Oracle's default rose to its highest level since 2009. Participants' concerns are growing alongside talk of an industry overheating.
Nevertheless, infrastructure needs are enormous: by 2028, investments in data centers could reach $2.9 trillion, while companies' own funds cover only 50% of the requirements. The remainder will have to be raised through debt capital.
My conclusion: The bond market is becoming the main benchmark for assessing real demand for AI. If the cost of borrowing continues to rise, it could slow down the adoption of technology. But for now, the mechanism works like clockwork — every microchip is financed through credit funds, and Morgan Stanley is at the helm of this process.