Lambda Borrows $1 Billion to Buy Nvidia Chips for Microsoft: Inside AI's Debt Machine

The most telling number in AI infrastructure this week was not a model benchmark or a valuation. It was a loan. On August 28, TechCrunch reported that Lambda — the Nvidia-backed “neocloud” that buys computing chips and rents them out — has raised $1 billion in private, short-dated debt to buy Nvidia AI chips that it will then lease to Microsoft. The deal was arranged by JPMorgan Chase, according to Bloomberg’s original reporting.

The structure tells you everything about how AI capacity actually gets built in 2026: a specialized cloud borrows against contracted future revenue from an investment-grade customer, buys GPUs from its own investor, and repays the loan from the lease checks. Equity rounds get the headlines. Debt builds the data centers.

What we know about the deal

Details of the private placement are sparse by design — JPMorgan declined to comment, and Lambda, Nvidia, and Microsoft had not issued statements. But the shape of the transaction is clear from the reporting:

• ~$1 billion in private, short-dated debt, marketed to institutional investors through private channels.
• Proceeds earmarked for Nvidia GPUs that will serve Lambda’s November 2025 agreement with Microsoft, under which Lambda deploys tens of thousands of Nvidia GPUs dedicated to Microsoft’s AI workloads.
• JPMorgan Chase arranged the financing.
• The short tenor is the point: TechCrunch notes the terms signal Lambda is betting it can deploy the chips and start generating revenue fast enough to repay the debt quickly from incoming cash — a working-capital instrument, not a moonshot.

This is Lambda’s third significant debt transaction in under four months. In May, the company closed a $1 billion secured credit facility. And this week, it announced the closing of a $926 million senior secured term loan B to fund Nvidia GB300 GPUs — one of Nvidia’s newest chips — for a deployment it is under contract to provide to Nvidia itself.

The $926 million deal is the more remarkable one

If the $1 billion Microsoft deal is aggressive, the term loan B that closed alongside it is arguably the more structurally significant piece of news. Per Lambda’s own announcement, the $926 million facility is the first investment-grade-rated term loan B financing ever executed by a private neocloud, drawing a Baa2 rating from Moody’s.

The mechanics deserve attention, because they mark the arrival of AI infrastructure as a mainstream credit asset class:

• Priced at SOFR + 3.00%, issued at 99.5% of principal — with pricing tightening by a full 75 basis points during syndication as investor demand flooded in.
• Heavily oversubscribed, attracting what Lambda calls a broad group of institutional investors.
• Fully amortizing, with a maturity of December 31, 2030, aligned to the contracted cash flows and useful life of the underlying GPUs.
• Secured by the GPU servers themselves and the contracted cash flows they generate.
• Structured as Lambda’s first large-scale private cloud GPU asset-backed SPV financing — a template the company explicitly says it will reuse for committed customer deployments.
• Morgan Stanley acted as lead left arranger and administrative agent, with MUFG as joint bookrunner.

CEO Michel Combes framed the milestone plainly: the transaction “will establish Lambda as the first private neocloud to access the term loan B market with an investment-grade rating,” reflecting “the strength of our customer commitments, the quality of our infrastructure and growing institutional conviction in AI infrastructure as an asset class.”

That last phrase is doing a lot of work. Two years ago, lending against rapidly depreciating GPUs would have been a niche private-credit story. A Baa2 rating from Moody’s on a private GPU cloud’s debt means the rating agencies now treat contracted AI compute as an annuity-like cash flow — because, contractually, it is.

$400 billion and counting

Lambda is not an outlier; it is a leading indicator. According to data Bloomberg compiled, banks and tech companies have raised over $400 billion in AI-related debt globally in 2026 so far. The AI capital cycle has decisively bifurcated: equity funds the platforms and the model makers, while debt funds the physical plant — chips, racks, power, and the buildings that house them.

The debt route has obvious attractions for a company like Lambda. It is reportedly in talks for a $3 billion pre-IPO round at a valuation above $12 billion, after raising $1.5 billion in venture capital last November at a $5.43 billion post-money valuation. Every dollar of asset-backed debt that funds a contracted deployment is a dollar of equity the founders do not have to sell before an anticipated IPO. Debt investors get paid from Microsoft’s lease; equity keeps the upside.

The circularity question

Skeptics — and there are many in credit markets — point to the loopiness of the arrangement. Nvidia is simultaneously Lambda’s investor, supplier, and (via the GB300 contract) customer. Microsoft leases GPUs from Lambda rather than buying them outright, converting capital expenditure into operating expense while keeping capacity off its own balance sheet. JPMorgan and Morgan Stanley underwrite the middle. The whole structure works beautifully — as long as AI demand keeps absorbing every chip that gets plugged in.

The risks are equally structural. GPUs depreciate on a brutal schedule; a rack of GB300s is not a toll road. If model architectures shift, if inference moves to custom silicon, or if a hyperscaler walks away at renewal, the “contracted cash flows” securing these loans compress fast. And short-dated debt is a roach motel in a liquidity squeeze: easy to roll in good markets, punishing when sentiment turns. The 2008-era lesson that “investment grade” is a trailing indicator, not a shield, applies.

But the near-term read is hard to argue with. Lambda has now demonstrated that a private neocloud can borrow at investment-grade rates, in size, against contracts with blue-chip counterparties — and that institutional credit investors will fight for allocation. That is a new financing primitive for the entire sector, and every Lambda competitor from CoreWeave to Crusoe just saw its cost of capital potentially drop.

What to watch

Three things determine whether this is sound finance or a leveraged treadmill. First, deployment speed: the short-dated Microsoft loan assumes revenue starts flowing within months, not years. Second, the pre-IPO round: a $3 billion equity cushion at a $12 billion-plus valuation would reset Lambda’s balance sheet entirely. Third, renewal economics when the initial Microsoft and Nvidia contracts expire — because at that point, the debt is still there and the GPUs are four years older.

For now, the machine is running. Nvidia sells the chips, Lambda borrows against them, Microsoft leases them, and the banks take their fee at every hop. The AI boom’s second act is being financed not by venture conviction but by credit committees — and this week, the credit committees said yes to another billion.