Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
IndustryInsight · On RadarGuardian→ original

CoreWeave Owes $47 Billion on Nvidia Chips and Lost $709 Million in One Quarter

CoreWeave, a former crypto miner, buys Nvidia chips on credit and rents them to OpenAI and Anthropic. Its debt already stands at $47 billion, with $10 billion due to be repaid next year. In the latest quarter the company lost $709 million on $3.6 billion in revenue. The model holds only while customers keep paying and Nvidia keeps financing them. That is why teams should not tie their product to a single GPU provider.

AI-processed from Guardian; edited by Hamidun News
CoreWeave Owes $47 Billion on Nvidia Chips and Lost $709 Million in One Quarter
Source: Guardian. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceGuardian ↗
Editor-in-Chief
Key metric #1Interest on the debt is expected to exceed $1 billion per quarter
Key metric #2Two-thirds of the cash inflow comes from customer prepayments, not profit
Key metric #3The rate on GPU-backed loans fell from 15% in 2024 to 9%

Imagine you rent GPUs to train a model or run agents. The provider you rely on may be sitting on a mountain of debt. That is the situation at CoreWeave, a former crypto miner that the Guardian calls the "face of the AI bubble." It owns Nvidia chips, rents them to the largest AI labs, and builds all of this with borrowed money. Its debt has reached $47 billion, and more than $10 billion must be repaid as early as next year. In the latest quarter the company lost $709 million on $3.6 billion in revenue.

How the model works

CoreWeave belongs to the category of "neoclouds." OpenAI and other labs want to serve billions of users but do not want to invest hundreds of billions in data centers themselves. Nvidia wants to sell as many chips as possible. CoreWeave sits in the middle: it buys the chips, installs them in data centers (by its own count there are 52, from Texas to Scotland, many still under construction), and rents out the capacity.

Nvidia owns about 11% of CoreWeave. There is also a $6.3 billion deal under which Nvidia committed to buy back capacity that CoreWeave cannot sell. Nvidia also invests in CoreWeave's customers, including OpenAI and Anthropic. The result is a loop: a dollar from Nvidia becomes a customer's money, then CoreWeave's revenue, and after the chip purchase, Nvidia's revenue again. The Guardian compares it to a sandwich: no slice of bread wants to build trillion-dollar "semiconductor sheds," so the risk falls on the filling.

Debt backed by chips

Roughly $25 billion has been raised through loans secured by the GPUs themselves, backed by customer contracts. Lenders are shown both the product and the customers who pay for it. The idea looked dubious: unlike real estate, chips lose value. In October, The Verge called the company a "horror from the abyss" with no clear path to profitability. But Crusoe, Nscale and Lambda adopted the same structure, and the market got used to it. The first such loan in 2024 carried a 15% annual rate; later loans came in at 9%. In March, $8.5 billion of this debt received an investment-grade rating and was sold to institutional investors, including insurers and pension funds.

The disputed point is the chips' useful life. CoreWeave assumes they lose value over six years; skeptics think it happens faster. If so, the collateral would be worth less than the loan. Jordan Chaflin, an analyst at CreditSights, is optimistic: the accelerator market is very tight, used chips are in demand, and customers have signed long contracts. Besides GPU-backed loans, the company also has unsecured bonds. If it went bankrupt, their holders would in theory be repaid first, but CoreWeave has few assets beyond its chips, and it leases many of its buildings.

An SEC filing shows that over the summer the company repaid 18 types of debt totaling about $35 billion, and the total has since grown by almost $12 billion. Interest is expected to exceed $1 billion per quarter. In the autumn, creditors put the probability of default within five years above 60%.

Where the weak points are

Brian Hopkins, an analyst at Forrester, points to prepayments. CoreWeave takes large advances on multi-year contracts, so roughly two-thirds of the cash it "generates" comes from customers who paid in advance, not from profit on services already delivered. That means the company must keep finding new buyers. The customers themselves, including OpenAI and Anthropic, are deeply loss-making and pay for compute with borrowed money, some of it from Nvidia.

The Guardian also reports on a scandal around Nvidia CEO Jensen Huang, which, according to the paper, saw him acknowledge a relationship with an alleged Chinese spy and step down from his post. We cannot verify this independently. The paper's logic is clear, though: if Nvidia steps back from its role as the AI industry's "central bank" and stops financing customers, the strain would fall on the entire chain.

A CoreWeave representative disagrees: customers sign multi-year commitments for older GPUs, for longer than their accounting useful life and at prices above the original cost, and financing is becoming more flexible and cheaper.

What practitioners should do

For developers and product managers this is not an abstract financial drama. If the chain breaks, availability and compute prices will suffer, and tenants face the risk of outages. It is sensible not to tie a product to a single GPU provider, to keep a fallback option, and to factor the providers' debt load into vendor selection. Which version is right will become clear within the next year, when CoreWeave has to repay $10 billion.

Story

Key Questions Answered

How the model works?

CoreWeave belongs to the category of "neoclouds." OpenAI and other labs want to serve billions of users but do not want to invest hundreds of billions in data centers themselves. Nvidia wants to sell as many chips as possible. CoreWeave sits in the middle: it buys the chips, installs them in data centers (by its own count there are 52, from Texas to Scotland, many still under construction), and rents out the capacity. Nvidia owns about 11% of CoreWeave.

What are the key details on Debt backed by chips?

Roughly $25 billion has been raised through loans secured by the GPUs themselves, backed by customer contracts. Lenders are shown both the product and the customers who pay for it. The idea looked dubious: unlike real estate, chips lose value. In October, The Verge called the company a "horror from the abyss" with no clear path to profitability. But Crusoe, Nscale and Lambda adopted the same structure, and the market got used to it. The first such loan in 2024 carried a 15% annual rate; later loans came in at 9%.

Where the weak points are?

Brian Hopkins, an analyst at Forrester, points to prepayments. CoreWeave takes large advances on multi-year contracts, so roughly two-thirds of the cash it "generates" comes from customers who paid in advance, not from profit on services already delivered. That means the company must keep finding new buyers. The customers themselves, including OpenAI and Anthropic, are deeply loss-making and pay for compute with borrowed money, some of it from Nvidia.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…