Lambda’s $4 Billion Round Rests on One Customer’s Balance Sheet

The Lambda pre-IPO round would value the company at $14.5 billion before the new money, according to TechCrunch, reporting on a Wall Street Journal story. Coatue Management and Blackstone are leading it. NVIDIA backs the company and joined an earlier round in February 2025.

The mechanism matters more than the headline. A neocloud buys GPUs, builds or leases space to run them, and sells the resulting capacity on multi year deals. Contracted revenue becomes the asset it borrows against. Lambda closed a roughly $1 billion term loan on 1 October at a fixed 6.78 percent rate, secured by the servers and the customer cash flows, with a delayed draw that releases money as clusters enter service. That structure ties the borrowing to specific deployments, which is prudent, and it also means the debt and the backlog rise together.

Backlog is not a shortcut to profit. It is an order book, meaning what a paying customer has promised, and nothing more. Lambda has published strong growth for two years and the pace of the sector is real. The question a buyer has to answer is what share of that order book would survive a change of mind at the top.

One Contract Carries Most of the Growth

The investor letter puts the backlog at $50 billion in September, up from $15 billion in June. About $35 billion of that increase is attributed to a commitment from Anthropic. The Anthropic capacity runs on NVIDIA hardware in a Texas facility operated by Hut8. Strip that single agreement out and the order book is closer to $15 billion, which is a healthy business and a very different story.

Concentration like that is normal at this stage and it is still the whole risk. Anthropic is one of a small number of frontier labs, and its own spending depends on its own funding. If Anthropic slows its buildout, defers a tranche, or renegotiates, the backlog that anchors the valuation moves with it. A valuation built on a single counterparty is a concentrated bet, whether or not the word appears in the term sheet.

The Debt Gets Repaid Whether the Backlog Converts or Not

Capacity buildouts are funded by debt, and lenders have tightened. Lambda added another $1 billion in debt last week on top of the October term loan. The money buys GPUs and energizes sites on a schedule. The revenue arrives later, and only if the customers still want the capacity when the racks power on. CoreWeave and Nebius carry the same structure, which is why their share prices now set the cost of their next buildout.

This is where the AI cloud looks less like software and more like a utility. Capital sits in physical assets with a hard depreciation curve. Utilization decides whether the asset earns. A contract that is signed but not yet delivering is a line on a spreadsheet, not a payment. See how the rack itself changes the math in our piece on NVIDIA rack scale purchasing, and how utilization moves the same economics in our look at token prices.

Three Questions Before You Underwrite a GPU Cloud

Ask what share of the backlog comes from your three largest customers, and whether you have seen the contracts behind the numbers. Ask how much of the debt is secured by customer cash flows that have not started. And ask what the model looks like if the single biggest commitment is deferred by two quarters, which is the version a buyer should stress test before the listing.

A strong backlog and a profitable business are different things. Lambda has real customers and real growth. The round is a bet that one of them keeps paying, and that is the line to read closely when the paperwork becomes public.