CoreWeave's Q2: The Illusion of Profit in AI's GPU Landlord

CoreWeave reported $2.58 billion in revenue for Q2 2025. Growth of 112%. Adjusted EBITDA of $1.51 billion. A margin of 58.5%. The market cheered. The logic, however, was a lie. The company still lost $626 million on a GAAP basis. Net loss. Not a typo. Revenue tripled, but cash flow remained negative. This is not a profitable company. It is a capital-intensive machine burning cash to build infrastructure. The adjusted EBITDA number is a mirage, a non-GAAP construct that ignores depreciation, interest, and stock-based compensation. The code of the balance sheet said one thing. The market chose to hear another. CoreWeave is not a cloud giant. It is a GPU wholesaler. It buys NVIDIA's H100 and B200 chips in bulk, builds data centers, and leases compute power to AI companies. Its IPO in early 2025 at ~$40/share was a bet on the AI infrastructure boom. The company's moat is not technology—it is supply chain relationships with NVIDIA and the ability to secure power and real estate. The backlog of $104 billion suggests long-term demand, but the growth rate of that backlog (4.6% quarter-over-quarter) raises questions. New contract wins are not accelerating. Let's deconstruct the earnings from first principles. Revenue growth is undeniable. But the cost of that growth is staggering. Q2 revenue was $2.58B. Cost of revenue? Not disclosed, but with gross margins likely around 60-70% (typical for cloud), operating expenses plus depreciation eat the rest. The net loss of $626M is an improvement over prior quarters, but it still means the company burns cash. Capital expenditure guidance was raised to $35-39 billion. That is more than 3x annualized revenue. The company is spending heavily to build out capacity, hoping future demand will justify it. But the backlog growth is slowing. The backlog increased by only $4.6% quarter-over-quarter, implying that new contracts signed barely outpaced revenue recognized. The demand is not accelerating; it is steady. That is a problem when you are building capacity at a frantic pace. Using my experience auditing DeFi protocols, I've learned that non-GAAP metrics are the equivalent of 'circulating supply' claims—they mask the true state of liquidity. Adjusted EBITDA is like a protocol's 'total value locked'—it looks impressive but ignores the cost of the underlying assets. In CoreWeave's case, the underlying assets are depreciating GPUs. The value of those assets depends on continued demand for AI compute. If demand plateaus, the assets become stranded. The CEO mentioned that new contracts signed in Q2 have margins 5-10% higher than previous quarters. This suggests that earlier contracts were underpriced, or that costs have risen. Either way, it indicates that the company is learning to price better, but it also means that the backlog's profitability is not uniform. The average margin across the $104 billion backlog is unknown. Data does not lie, but it does not care about your narrative. The company's reliance on NVIDIA is a single point of failure. If NVIDIA's supply chain tightens or if AMD/Intel catch up, CoreWeave's competitive advantage evaporates. The moat is not code; it is a relationship. And relationships are not immutable. CoreWeave is essentially a leveraged bet on NVIDIA's continued dominance in AI chips. That is a bet that may pay off, but it is not a moat of technology or network effects. Another hidden risk: the procurement commitments in the backlog are likely non-cancellable contracts, but they are also subject to the creditworthiness of the customers. If an AI startup goes bankrupt, the contract is worthless. The AI industry is still nascent, and many of CoreWeave's customers are loss-making themselves. A chain of defaults could cascade. Now, the contrarian angle. The bulls have a point. The demand for AI compute is real. CoreWeave is the purest play on that demand. Its ability to secure NVIDIA GPUs and build data centers faster than competitors is a real advantage. The backlog shows that large customers are willing to commit billions of dollars. The revenue growth trajectory is steep. If the company can maintain its margin profile and eventually convert to GAAP profitability, the stock could be a winner. Trust is a variable you cannot hardcode. And in this market, trust is being tested. But the numbers do not care. The net loss is a fact. The CapEx-to-revenue ratio is a fact. The slowing backlog growth is a fact. The palace is built on a fault line of NVIDIA's dominance and sustained AI hype. They built a palace on a fault line. The question is not whether AI demand will grow, but whether CoreWeave can survive a downturn. The answer is not in the adjusted EBITDA. It is in the cash flow statement. And right now, that statement is a warning. In a sideways market, chop is for positioning. CoreWeave's stock may rally on the Q2 headline, but the underlying metrics tell a different story. The company is a high-leverage play on a single narrative. When the narrative shifts, the structural weaknesses will be exposed. The code spoke, but the logic was a lie. The truth is in the cash flow, and it is not pretty.