Lambda is selling a $917 million leveraged loan to finance a chip purchase, Bloomberg reported on Monday. Jeannine Amodeo, Paula Seligson and Gowri Gurumurthy broke the story. The money pays for GPUs and other infrastructure as part of a contract with Nvidia, according to a person with knowledge of the matter. Morgan Stanley leads the transaction.

Lambda Compute II LLC and Lambda Cloud Canada Inc will issue the debt. A lender call opened at 10:30am New York time on Monday. Commitments fall due on Thursday. Order books had reached nearly $2 billion after a pre-marketing effort.

Bloomberg frames the risky debt market as a new front in the borrowing binge funding the AI buildout. Its own compiled data puts global AI-linked borrowing at nearly $600 billion since last year. Lambda belongs to the group of firms known as neoclouds, which rent out access to microchips and other AI infrastructure.

The circle closes on itself

Start with who sits on the other side of the contract. Nvidia already held an investment in Lambda. It also supplies every chip Lambda runs, because the company uses Nvidia silicon exclusively.

Then it went further. In September 2025 Nvidia agreed to lease GPUs back from Lambda, in a $1.3bn four-year deal covering 10,000 servers plus a $200m contract for 8,000 more. RCR Wireless reported that the arrangement made Nvidia its largest customer, worth roughly $1.5bn across 18,000 servers.

So Nvidia plays four roles at once. Investor, supplier, biggest tenant, and now counterparty to the contract a $917m loan exists to fund.

The shape is familiar. Google built the same loop around its own TPUs, guaranteeing rent for data centres that buy its chips to serve Anthropic. The Bank for International Settlements flagged the pattern in June, warning that an AI investment collapse could disrupt credit markets on the scale of 2008.

The BIS singled out chipmakers taking equity stakes in firms that then commit to buying chips from those same investors. The terms of such deals, it said, “are typically poorly disclosed, with risks of the same asset being pledged multiple times.”

The venue matters more than the number

$917m looks small next to the sums moving through AI infrastructure. The venue does not. This deal lands in the institutional leveraged loan market, where investors buy the debt of companies rated below investment grade.

CoreWeave opened that door in April. Bloomberg called its $3.1bn deal the first of its kind to finance chips in the institutional loan market. CoreWeave then sold debt backed by customer contracts, including one with OpenAI, in May. A later deal backed by different contracts forced it to pay a hefty yield, which pushed its borrowing costs up sharply.

Compare the alternative route. Nebius raised a $775m GPU-backed facility in July at SOFR plus 2.50 percentage points, roughly 6.8%. Banks syndicated that one. An investment-grade customer contract sat behind it, throwing off cash flows that covered more than 100% of the capital expenditure.

Lambda pays as much as 3.75 percentage points over the benchmark rate, at a discounted price of 99 cents on the dollar. Same collateral class, different market, higher cost.

The terms name the nervous party

Read the structure and you can see what lenders demanded. The loan matures in 4.4 years. Institutional loans usually run seven.

It also amortises fully, so the debt repays over roughly four years rather than landing as a lump at the end. Bloomberg notes that amortisation ranks among the protections investors have been seeking. Here it means lenders carry no refinancing risk at all.

A call-protection clause adds a penalty if Lambda redeems the debt early. Bloomberg describes that combination as closer to a bond deal than a loan.

Those features do one specific job. They shorten the window in which a GPU has to keep earning. TNW noted the open question when Nebius borrowed in July. The whole asset class rests on residual value assumptions that nobody in this industry has had to test yet. A four-year amortising schedule sidesteps that question rather than answering it.

What Lambda actually is

Machine learning engineers founded the company in 2012, working out of the Noisebridge hackerspace in San Francisco’s Mission District. It now calls itself an AI-only company and builds modular AI factories with power, liquid cooling and high-bandwidth interconnects, at what it describes as gigawatt scale. Its mission statement promises to “make compute as ubiquitous as electricity and give everyone in America the power of superintelligence.”

A management overhaul followed a $1.5bn Series E in November 2025. Lambda named Michel Combes chief executive in May, after his runs at Sprint, SoftBank International and Alcatel-Lucent. Former AT&T Communications boss John Donovan became chairman. Co-founders Stephen and Michael Balaban moved to chief technology officer and chief product officer. The company targets 3GW of AI compute under management by 2030.

Microsoft signed a multibillion-dollar agreement with Lambda last November to deploy tens of thousands of Nvidia GPUs, CNBC reported. Lambda has held preliminary discussions with bankers about a public listing. It sells only in North America, and it has started building its own data centres rather than renting space in other people’s.

What the deal does not settle

Investors clearly want the paper. Order books of nearly $2bn against a $917m loan point to real demand, and the amortising structure shows they went in with their eyes open.

Several things stay unresolved. Lambda and Morgan Stanley did not respond to Bloomberg’s requests for comment. The detail tying the debt to an Nvidia contract comes from an unnamed person. Lambda’s revenue and loss figures sit behind paywalled reporting, so no audited number exists in public.

The wider market has also turned fussier. Loan investors started pushing back this month as fear rose, Bloomberg reported. Big Tech alone has lined up $350bn of AI debt over five years, and private credit keeps stacking more on top. Blue Owl’s Stack Infrastructure is chasing a $5.9bn data centre loan of its own.

Lambda leaves lenders one question, and it has nothing to do with chips. Nvidia sits on both sides of this trade. If AI demand holds, that reads as alignment. If it slips, the same fact reads as concentration, and the loan matures long before anyone finds out which.

Get the TNW newsletter

Get the most important tech news in your inbox each week.