Lambda, the AI cloud and GPU infrastructure provider, has raised around $1 billion in private short-dated debt arranged by JPMorgan to buy Nvidia GPUs that Microsoft will lease. The deal, which was marketed to private placement investors, underscores how the AI boom is reshaping corporate finance. Lenders are increasingly securitizing hardware and lease contracts, rather than relying on more traditional corporate credit metrics.
According to people familiar with the financing, Lambda is borrowing against Microsoft’s willingness to continue paying under the lease agreement, not against Lambda’s own revenue or cash flow. That distinction matters. The structure transfers the credit risk from Lambda’s operating business to the creditworthiness of Microsoft, one of the most valuable companies in the world. Microsoft gets access to additional computing capacity without having to add debt to its balance sheet, while Lambda obtains the capital needed to buy the expensive chips.
A Second Deal in Weeks
The transaction is the second similar raise this month. Two weeks earlier, Lambda borrowed $917 million against a contract with Nvidia. Nvidia’s relationship with Lambda is multifaceted: the chip giant is an investor in Lambda, its primary supplier of GPUs, and, according to the earlier report, also a customer of Lambda’s cloud services. This circularity reflects a broader trend in the AI industry, where the largest players often wear multiple hats in overlapping contracts, making it difficult for outsiders to assess the true risk exposure.
The exact terms of the new $1 billion transaction were not independently verified. The people who described the deal were not authorized to speak publicly, and JPMorgan declined to comment. Companies arranging private placements typically do not disclose full details, which is one reason regulators have begun to express concern about transparency in the private credit market.
How the Collateral Chain Works
In this structure, Lambda takes on debt to acquire physical assets: Nvidia GPUs. Those GPUs are then leased to Microsoft. Microsoft’s lease payments are the stream of income that supports the debt. If Microsoft stops paying, the GPUs remain as collateral. A lender can seize and sell them, though the resale value of specialized AI chips is uncertain and volatile. This is a relatively short collateral chain: a lender is ultimately underwriting one American technology company’s continued appetite for compute it does not own.
The arrangement is reminiscent of project finance or asset-backed lending, but with an AI twist. Instead of airplanes, ships, or real estate, the underlying asset is a rapidly depreciating semiconductor that can become obsolete within a few years. Nvidia’s GPUs, especially the latest generation used for AI training, are in high demand, but their useful life is short compared to traditional collateral. That makes the valuation of these assets inherently linked to the pace of AI adoption and the competitive dynamics of the chip market.
Microsoft Is the Anchor Throughout
Microsoft appears as the anchor tenant in several such deals. The software and cloud giant is one of the largest buyers of AI computing power in the world, alongside companies like OpenAI, Anthropic, and Meta. By leasing GPUs instead of buying them outright, Microsoft can increase its AI capacity while preserving capital and keeping debt off its balance sheet. This approach is common among cloud providers, but the new financing structure adds a layer of debt that is secured by the lease itself.
Another company, Nebius, has used a similar model. Nebius raised $775 million against its own GPUs earlier this year, the first secured debt taken by the Amsterdam-headquartered company. Nebius holds a five-year Microsoft contract worth $19.4 billion, and its management has said it has around $40 billion of contracts that could be securitized. The fact that Nebius and Lambda are both using Microsoft as the anchor tenant highlights how concentrated the AI infrastructure market has become. A handful of hyperscale cloud providers and their affiliates are responsible for a large share of global GPU demand, and they are increasingly using their purchasing power to finance their suppliers’ expansion.
Record AI-Related Debt
The scale of these financing arrangements is growing rapidly. Lambda is also reported to be in talks to raise as much as $3 billion ahead of a possible stock market listing next year. Meanwhile, more than $400 billion of AI-related debt has been raised globally in 2026 alone, according to estimates cited by financial media. That staggering number includes a wide range of instruments: corporate bonds, loans, asset-backed securities, and private placements arranged by major banks.
The private credit market is especially active in AI deals because banks and direct lenders can structure bespoke contracts with speed and confidentiality. Investors in these private placements are typically institutional buyers such as pension funds, insurance companies, and sovereign wealth funds, who accept higher risk and lower liquidity in exchange for higher yields. The short-dated nature of Lambda’s debt means that investors are exposed to a quick repayment schedule, which could be a source of stress if the AI market cools.
Regulators Sound the Alarm
Central banks and financial regulators have begun to warn about the risks embedded in these structures. In May, the European Central Bank cautioned about the private credit market’s “opaque valuation practices and limited liquidity.” It also pointed to portfolios that are heavily concentrated in a few US issuers whose valuations are driven by the AI narrative. The ECB’s concerns are not purely theoretical; several European financial institutions have been lending against US AI assets, creating transatlantic exposure to a relatively new and volatile sector.
In June, the Bank for International Settlements was more blunt. It warned that an AI investment collapse could disrupt credit markets on the scale of the 2008 financial crisis. The BIS also highlighted a specific risk: poor disclosure of deal terms makes it difficult to determine whether the same asset has been pledged to multiple lenders. Collateral reuse, sometimes called “rehypothecation” in other contexts, could amplify losses if an AI downturn leads to a rush of margin calls and forced sales.
The Uncomfortable Position in Europe
These warnings land awkwardly in Europe. When European Union officials speak about digital sovereignty and the need for homegrown AI compute, they often point to companies like Nebius and Lambda as evidence that the region can play a role in the AI infrastructure boom. Yet the actual financing of these companies is tied to American corporate balance sheets, especially Microsoft. The EU’s ambition to build sovereign compute capacity is therefore, in part, underwritten by an American company’s willingness to lease servers and pay for them over time.
Some analysts argue that the “sovereign compute” framing is thinner than it appears. Countries want to host data and operate AI models on their own territory, but if the underlying chips are financed through a US-based lease, the strategic value may be less than advertised. The chips themselves can be moved, and the data flows may still pass through US-controlled systems. The financing structure also means that European lenders are, in effect, betting on the continued growth of American AI companies, which may not align with European policy goals.
What Happens in a Downturn?
The key question for investors is what happens if demand for AI compute falters. In a downturn, Microsoft might reduce its leasing commitments, or Nvidia might face order cancellations, or the GPUs might be worth less than the outstanding debt. The collateral is highly specialized, and there may be only a handful of interested buyers if a lender tries to liquidate assets. Unlike real estate or corporate loans, AI chips lose value quickly and become technically obsolete, making them poor collateral in a prolonged crisis.
That is why the structure of these deals places the ultimate loss burden in unexpected places. If a debtor defaults, the lender seizes the GPUs and tries to sell them. The company that was leasing the GPUs, Microsoft in this case, may simply walk away without a loss. The equity holders in Lambda, the debt investors, and possibly the insurers of the debt would bear the losses. In other words, the party that is using the compute does not necessarily carry the financial risk of the hardware. This is a crucial point for anyone trying to understand the AI credit boom.
The chips will sit somewhere, and someone will own them. On this structure, the party carrying the loss is not the one using the compute. As the AI bubble debate continues, that asymmetry is one of the most important financial details to watch.