Lambda sells insurers $1 billion of GPU debt at 6.78% for 30,000 Nvidia chips

Lambda sold $1.008 billion of fixed-rate debt to insurance companies at 6.78%, secured by contracts with two investment-grade customers it does not name. Days earlier, similar secured data-center paper cleared at 7.875%.

Vincent JiangVincent Jiang · 3 min read
Share
Four Nvidia H100 data-center GPUs stacked in a server chassis, gold heatsinks and Nvidia logos visible.
Nvidia H100 GPUs, the kind of hardware securing Lambda's $1.008 billion insurer loan. The proceeds buy more than 30,000 Nvidia chips.

Insurers now hold the collateral

On October 1, Lambda closed a $1.008 billion delayed-draw term loan marketed to insurance companies and fixed-income investors, and the book was oversubscribed 1. The fixed rate is 6.78%, inside the company's target range, on paper rated Baa1 by Moody's and A (low) by Morningstar DBRS 12. The proceeds buy more than 30,000 Nvidia GPUs for three committed deployments with two investment-grade offtakers across multiple data centers 12.

Capital draws only as clusters enter service, and the loan amortizes fully to May 30, 2033 1. It is secured by the servers and the contracted cash flows they produce, and J.P. Morgan arranged it alone 1.

The rating rides on unnamed signatures

Lambda will not say who the two offtakers are. The Baa1 rides on their signatures: if GPU values break, the amortizing schedule pays lenders first and the insurers own the chips 1. Microsoft is the reported customer behind a separate $1 billion private debt deal in August, per Bloomberg; the rated facilities name no one 3.

The cadence is the point: a $1 billion bank facility in May, a $926 million term loan B on August 27 at SOFR plus 3.00% with a Baa2 rating, then this, about $2.9 billion in six months 134. Combes calls it the third new credit market Lambda has opened in 18 months 1.

Lambda has borrowed about $1 billion in each of three 2026 debt markets

$0M$500M$1,000M$1,500MMay 2026, bank facilityAug 2026, term loan BOct 2026, insurer loan6.78% fixed, inside target range
Data
Facility size
May 2026, bank facility$1,000M
Aug 2026, term loan B$926M
Oct 2026, insurer loan$1,008M
Announced principal of each facility, millions of US dollars: May secured credit facility, August senior secured term loan B, October delayed-draw term loan.1,3,4

A $3 billion pre-IPO round is already in the works

Michel Combes, newly installed as chief executive, says the capital "underwrites infrastructure in decades, not quarters" and has funded a private company "on the strength of our customer contracts" 15. Behind the debt sit a $5.43 billion post-money valuation from November 2025 and reported talks for a $3 billion pre-IPO round 3.

Everyone else pays the gap

Below the top tier, the market is splitting, not tightening. Banks and technology companies have raised more than $400 billion of AI-linked debt worldwide this year 3. At the bottom of the stack, about $18 billion of loans tied to Oracle's Project Jupiter campus were quoted at 89 to 91 cents, with syndicate sales stalled and Oracle one notch above junk 6.

CleanSpark's data-center subsidiary paid 7.875% for $2.276 billion of secured notes on September 25 7. IREN counts about $14 billion of funding capacity against a $25 billion to $30 billion fiscal-2027 build 8. The roughly 110 basis points between Lambda's 6.78% and CleanSpark's 7.875% is what two signed, rated customers are worth 17.

The rule for the sector: model AI capex at 6.78% only where the offtake contracts are rated and signed, and treat GPU paper clearing near investment grade without two rated offtakers as mispriced. Lambda targets 3 gigawatts by 2030 5. Watch the next deal for whether the offtakers get named, and what this paper is worth when its chips are three years old.

Deepdive

AI-generated from this story and its cited sources. Not investment advice.

Reader comments

0 comments

    Sign up

    Get your curated digest

    After email confirmation, you will receive a daily digest of the most relevant news that matter to your portfolio