CoreWeave keeps posting bigger losses even as its revenue doubles, and The Information’s latest breakdown makes a sharp point: the headline loss number tells you far less than most people assume. As The Information reports, reading this company by net income alone will lead you to the wrong conclusion in both directions. The business isn’t quietly failing. It’s also not as safe as the revenue chart suggests.
Start with the raw figures. In the second quarter of 2026, CoreWeave pulled in $2.58 billion in revenue, more than double the $1.21 billion from a year earlier. Its net loss widened to $626 million from $290 million. So revenue is booming and losses are growing at the same time. That combination confuses people, and it’s exactly where the real analysis begins.
📉 The losses are mostly accounting, not operations
Most of the red ink comes from two line items tied to how CoreWeave buys its hardware.
- Depreciation: the company ran roughly $1.15 billion of depreciation and amortization in a single quarter as its data center gear started aging on the books.
- Interest: net interest expense hit around $536 million in the same stretch, the cost of financing GPUs with debt.
Strip those out and the operating engine looks healthier than the bottom line. That’s the case bulls make. CoreWeave buys Nvidia chips with borrowed money, signs multi-year contracts, and the accounting front-loads the pain. Fair enough. But the same two line items are where the danger hides.
⚠️ Where the risk actually lives
Three things deserve more attention than the net loss.
- The depreciation clock. CoreWeave spreads GPU costs over roughly six years. Nvidia ships a new architecture every 18 to 24 months. Rival Nebius depreciates over four years, which recognizes the cost faster. If chips lose earning power quicker than six years, CoreWeave’s reported losses are actually understated, not inflated.
- The debt treadmill. Net debt sits near $33 billion against a market cap around $52 billion. Capital spending for 2026 is guided at $35 billion to $39 billion. That structure only works if demand stays hot and financing stays cheap. Any slowdown gets expensive fast.
- Margin compression. Adjusted operating margin fell from about 17% a year ago to roughly 1% as the cost of standing up new capacity outran the revenue it produced.
🔍 The concentration nobody’s pricing
CoreWeave loves to cite its backlog, and it’s a big one: around $99 billion in signed contracts with names like Microsoft, OpenAI, Meta and Nvidia. Impressive on paper. Look closer and the customer base is thinner than it appears.
The top two customers account for about 65% of revenue. And on a look-through basis, much of the Microsoft business exists to serve OpenAI workloads, while OpenAI is also signing directly. So a huge share of CoreWeave’s future rides on one underlying buyer, one that’s openly working to build and control more of its own compute over time.
💡 What this means for you
The CoreWeave story is really the template for the whole “neocloud” model, so the lessons travel.
- If you’re comparing AI infrastructure providers, check the depreciation schedule first. A four-year assumption versus a six-year one can flip a loss into a profit without anything real changing.
- If you’re a business leaning on rented GPU capacity, watch supplier balance sheets. A debt-heavy provider is more fragile if AI demand cools or rates move.
- If you’re tracking this as an investor or operator, the number that matters isn’t net loss. It’s the gap between how fast a GPU pays back its cash cost and how long it’s carried on the books, plus how concentrated the customer list really is underneath the logos.
What stands out here is how much of the AI buildout narrative rests on accounting choices and a handful of mega-customers. CoreWeave’s demand is real. Its financing and its dependence on OpenAI are the variables to watch through the rest of 2026. For the full breakdown of how to read the losses line by line, see the original analysis at The Information.