The numbers are dizzying. CoreWeave, fresh off its 2025 IPO, reported revenue growth that makes traditional cloud providers look like dial-up. Nebius, the reincarnation of Yandex's international AI assets, followed suit with a NASDAQ listing and a promise to pump billions into AI infrastructure. But here's the truth the headlines don't scream: both companies are bleeding cash, and their growth is a debt-fueled sprint that could end in a crash.
Context: The AI Infrastructure Gold Rush
The AI boom has created a new class of infrastructure play: GPU cloud providers that don't build chips but rent them out like a high-stakes Airbnb. CoreWeave and Nebius are the poster children. They buy thousands of NVIDIA H100/H200 GPUs, bundle them with high-speed RDMA networking, Kubernetes scheduling, and liquid cooling, then sell access to AI startups and hyperscalers starved for compute. The pitch is simple: "We get you GPUs faster than AWS." And in a market where speed is the only hedge, that pitch works.
But the model is heavier than it looks. These are not software companies with 80% margins. They are capital-intensive beasts that need to borrow billions, build data centers, and pray the GPU utilization rate stays above 80%. The ledger does not lie, but the CEOs do—they talk about "demand" and "growth" while hiding the debt-to-equity ratios.
Core: The Numbers Under the Hood
Let's break down what the PR spin doesn't say. CoreWeave's revenue growth is real, but it's driven by concentration: a handful of large customers (likely including Microsoft, which has a strategic partnership) account for over 50% of revenue. If one of them decides to build its own GPU cluster or switches to a competitor, the impact is catastrophic. The same applies to Nebius, which is smaller and more exposed to European AI sovereignty projects—a market that's politically driven, not purely commercial.
Both companies are loss-making, but not because demand is weak. The losses come from depreciation, interest payments, and massive capex. In my years tracking crypto and AI infrastructure, I've seen this pattern before: the block explorer reveals what the headline hides. Here, the hidden ledger is the debt-to-equity ratio. CoreWeave's debt load is likely 3-4x its equity, and with interest rates still elevated, every rate hike squeezes their margins. The talk of "strong demand" masks the reality that GPU utilization is the only metric that matters. If occupancy drops below 70%, the economics break.
Speed is the only hedge in a zero-latency market. CoreWeave and Nebius have used speed—faster GPU delivery, quicker deployment—to win contracts. But speed is a fleeting advantage. Once AWS, Azure, and GCP get their GPU supply chains sorted, they can undercut on price and win on ecosystem. The question is not whether these companies will grow, but whether they can turn growth into profit before the music stops.
Contrarian: The Unreported Risks
Everyone is bullish on AI infrastructure. But the risks are being ignored. First, customer concentration is a ticking bomb. If CoreWeave's top client walks away, the stock could halve overnight. Second, the business is interest-rate sensitive. These companies rely on debt and leasing to fund expansion. If credit markets tighten, their expansion plans stall. Third, the power bottleneck is real. In Virginia and Texas, the grid can't keep up with data center demand. Transformer lead times are 2-3 years. That's a physical constraint on growth that no amount of capital can fix.
The ledger does not lie, but the CEOs do. They talk about "AI-native" infrastructure, but the reality is that they are NVIDIA's distribution channel. NVIDIA controls the GPU supply and the CUDA ecosystem. If NVIDIA decides to prioritize its own cloud partnerships or push customers to other providers, CoreWeave and Nebius lose their edge. The idea that they have a moat is a fantasy. The real moat is the cost of capital and the power contracts they've locked in.
Another blind spot: the rise of custom AI chips. Google TPU, AWS Trainium, and AMD MI series are slowly eating into NVIDIA's dominance. If the market moves toward custom chips, these GPU cloud players lose their raison d'être. They are betting everything on NVIDIA's continued dominance—a risky bet when the tech landscape shifts every 18 months.
Volatility is the price of admission, not the exit. For investors, the high growth is tempting, but the volatility is baked in. These stocks will swing wildly on every earnings report, every rate decision, every NVIDIA announcement. The only way to win is to time the cycle perfectly, and that's a game few can play.
Takeaway: The Next Watch
Short-term, the AI arms race continues. CoreWeave and Nebius will grow revenue, sign more contracts, and build more data centers. But the medium-term outlook is grim. When demand growth slows—and it will—the industry will face a shakeout. The players with the lowest cost of capital and the highest utilization rates will survive. Everyone else will be acquired or bankrupt.
Watch for three things: GPU utilization rates (if they start disclosing them), free cash flow trajectory, and customer diversification. If those numbers improve, the story changes. Until then, this is a debt-fueled growth story dressed in AI hype. The ledger tells the truth—you just have to look past the headline.