CoreWeave's AI Gold Rush Is Getting InsaneCoreWeave, Inc. Class ABATS:CRWVmoonyptoCoreWeave is probably the clearest example of the neocloud model today.. The company buys NVIDIA GPUs, puts them into data centers, and rents that computing power to major customers like OpenAI, Microsoft, and Meta The interesting part is that most of that capacity is already committed. 98% of CoreWeave’s Q2 revenue came from contracted business, while on demand usage accounted for only 2% Revenue surged 112% year over year to $2.6 billion, but gross margin fell 8 percentage points to 66%. The reason is pretty straightforward eh.. data center leases, electricity, and other costs tied to the rapid expansion are rising quickly CoreWeave also reported a $49 million operating loss and a $626 million net loss, with $640 million in interest expense playing a major role. Much of that financing is tied to debt backed by its GPU infrastructure And the scale of investment is even more striking when you look at CapEx. CoreWeave spent $9.4 billion in Q2 alone, more than three times its quarterly revenue What does it all mean MoonMaster? 📚 Demand is still far ahead of supply: Backlog reached $104 billion, up 246% year over year, and CoreWeave added another $25 billion in customer commitments early in Q3. In the near term, available capacity is basically sold out 💰 Margins could start improving: Gross margin took a hit as CoreWeave rushed to bring new capacity online, but adjusted operating margin improved from 1% to 5% sequentially. The company says newer Q2 contracts should generate contribution margins 5 to 10 points higher than recent deals, even before the roughly 25% price increase introduced in July 🧠 The business is becoming more diversified: Storage, CPU, networking, and software now generate more than $400 million in ARR. Managed inference is also taking off, with booked ARR jumping from just $1 million to more than $100 million in a single quarter. 🏗️ But scaling this business is insanely expensive: CoreWeave raised its 2026 CapEx forecast to $35 billion to $39 billion as it pushes toward more than 1.85 GW of active power capacity by year end CoreWeave having a $104 billion backlog next to a targeted $19 billion 2026 exit revenue run rate sounds almost ridiculous at first. But the bottleneck isn't demand. It's physical capacity The real investment question is how quickly CoreWeave can turn GPUs and electricity into revenue while keeping its financing costs from eating up the margin gains. If it can execute on that, the growth story gets very interesting. If not, that massive backlog could take much longer to translate into actual profits.