Lambda is taking out a loan of $917 million to purchase chips from the company that is investing in it.

      Lambda is in the process of selling a $917 million leveraged loan to finance the acquisition of chips, as reported by Bloomberg on Monday. The article was authored by Jeannine Amodeo, Paula Seligson, and Gowri Gurumurthy. The funds will be used to purchase GPUs and other essential infrastructure as part of a deal with Nvidia, according to a knowledgeable source. The transaction is being led by Morgan Stanley.

      Lambda Compute II LLC and Lambda Cloud Canada Inc. will be responsible for issuing the debt. A lender call commenced at 10:30 AM New York time on Monday, with commitments due on Thursday. Following a pre-marketing campaign, order books had swelled to nearly $2 billion.

      Bloomberg characterizes the volatile debt market as a new dimension in the borrowing spree funding the AI expansion. Their aggregated data indicates that global borrowing linked to AI has approached $600 billion since last year. Lambda is classified among the neoclouds—companies that provide access to microchips and other AI infrastructure.

      The situation loops back on itself. Nvidia, which already has an investment in Lambda, also supplies all the chips Lambda utilizes, as the company exclusively employs Nvidia silicon. Further complicating the relationship, Nvidia agreed in September 2025 to lease GPUs from Lambda in a $1.3 billion four-year agreement covering 10,000 servers, along with an additional $200 million contract for 8,000 more. RCR Wireless reported that this made Nvidia Lambda’s largest customer, representing approximately $1.5 billion across 18,000 servers.

      Thus, Nvidia occupies four distinct roles: investor, supplier, major tenant, and counterpart to the contract that the $917 million loan is intended to support. This scenario mirrors a similar setup by Google, which created a closed loop around its TPUs, offering rent guarantees for data centers purchasing its chips to support Anthropic. The Bank for International Settlements flagged this pattern in June, cautioning that a breakdown in AI investments could disturb credit markets on a scale comparable to that of 2008.

      The BIS specifically pointed to chipmakers taking equity stakes in companies that promise to purchase chips from these same investors. It stated that the terms of such agreements are often poorly disclosed and carry risks of the same asset being pledged multiple times.

      The scale of the deal may seem modest at $917 million compared to larger financial movements in AI infrastructure, but the context is significant. This deal is positioned within the institutional leveraged loan market, where investors buy debt from companies rated below investment grade. CoreWeave set a precedent in April with a $3.1 billion deal, termed the first of its kind to finance chips in this market. Subsequently, CoreWeave issued debt backed by customer contracts, including a deal with OpenAI in May. A later transaction associated with different contracts incurred high yields, sharply increasing its borrowing costs.

      In comparison, Nebius raised a $775 million GPU-backed facility in July at SOFR plus 2.50 percentage points, roughly 6.8%, with banks participating in the syndication. An investment-grade customer contract supported this debt, generating cash flows that exceeded 100% of the capital expenditure.

      Lambda, in contrast, incurs up to 3.75 percentage points over the benchmark rate, offered at a discounted price of 99 cents on the dollar. While the collateral class is the same, the market conditions lead to more expensive borrowing.

      A closer analysis reveals what lenders have requested. The loan has a maturity of 4.4 years, while institutional loans typically extend to seven. Additionally, it amortizes fully, meaning the debt is repaid gradually over approximately four years instead of as a lump sum at maturity. Bloomberg highlights that this amortization is one of the investor protections being sought, mitigating refinancing risk.

      A call-protection clause imposes a penalty if Lambda repays the debt prematurely, with Bloomberg labeling this combination more akin to a bond agreement than a standard loan. These features serve a specific purpose: they limit the time frame during which a GPU must continue generating income. When Nebius borrowed in July, an open question arose regarding the viability of the entire asset class based on residual value assumptions that remain untested in this sector. The four-year amortizing schedule circumvents this inquiry without addressing it directly.

      Lambda's origins trace back to its founding by machine learning engineers in 2012, who began operating from the Noisebridge hackerspace in San Francisco's Mission District. The company now identifies itself as an AI-exclusive entity, creating modular AI factories with capabilities for power, liquid cooling, and high-bandwidth connections, conceptualized at what it refers to as gigawatt scale. Its mission is to "make compute as ubiquitous as electricity and give everyone in America the power of superintelligence."

      A management overhaul followed Lambda's $1.5 billion Series E funding round in November 2025. In May, Michel Combes was appointed as CEO after leading roles at Sprint,