CoreWeave Q2 financial report: Success doesn't have to be mine To understand CoreWeave's financial report, the first step is to understand its business logic. Simply put, it is to borrow money to buy cards and rent out computing power, "2x long token ARR". CoreWeave is borrowing on one hand: raising funds through debt, equity financing, and customer prepayments, and buying cards on the other hand: purchasing Nvidia GPUs, building data centers, and connecting power, and then leasing computing power to OpenAI, Meta, Microsoft, and other AI companies through long-term contracts. So CoreWeave wants to make money: to borrow money at the lowest possible cost, buy as many cards as possible, deploy as much computing power as possible, and then rent out higher prices as much as possible. So the financial indicators that such companies need to consider are revenue and backlog, GPU rental prices, capacity construction speed, capital expenditures, debt size, and financing costs. From these indicators, CoreWeave Q2 revenue reached $2.575 billion, a year-on-year increase of 112% and a month on month increase of 24%; The revenue backlog reached $104.2 billion, a year-on-year increase of 246%, and an increase of $4.8 billion compared to Q1. Considering that $2.575 billion in revenue has been confirmed for this quarter, it can be roughly estimated that the number of new contracts added in Q2 will be approximately $7.4 billion ($7.4-48 billion), which is nearly 2.9 times the Book to Bill ratio. This indicates that the main factor limiting revenue growth at present is still capacity delivery. Demand is not a problem, the problem is that capacity cannot keep up. Looking at the sales side again: The management stated that the expected contribution profit margin of the newly signed contracts in Q2 is 5 to 10 percentage points higher than in previous quarters, with a large portion coming from Vera Rubin products. The company also raised the prices of multiple SKUs by about 25% in July. The order prices are also improving. In terms of expansion: Q2 company's capital expenditure reached 9.352 billion US dollars, and Q3 is expected to further reach 11.5 billion to 13.5 billion US dollars; The annual capital expenditure guidance has been raised from $31 billion to $35 billion to $35 billion to $39 billion. Based on the midpoint of the annual revenue guidance of $12.8 billion and the midpoint of capital expenditures of $37 billion, this year's capital expenditures are close to 2.9 times the revenue. This is a very strong demand signal for the industry chain, but for CoreWeave shareholders, the company still needs continuous financing to maintain growth. Let's take another look at the debt side: Q2 cash flow simplification to the end: operating cash flow+679 million US dollars; Investment cash flow -7.166 billion US dollars; Financing cash flow+10.071 billion US dollars; Net cash increased by $3.584 billion. The company still needs a significant amount of external financing to support capital expenditures. As long as the capital market continues to recognize GPU residual value and customer contracts, the company can obtain funding at a lower cost and expand ARR; If credit spreads rise, lending institutions lower GPU collateral ratios, or customer contract quality declines, CoreWeave may face increased financing costs, diluted equity, or slower capacity building. So ultimately, the profitability of this model is still very fragile, relying heavily on the external financing environment. Therefore, Coreweave is starting to transform towards Token Factor, not just relying on renting GPUs to make money: CoreWeave's Managed Inference has contracted ARR to grow from approximately $1 million to over $100 million within a few months, and the company expects to reach at least $250 million by the end of 2026. This means that the company is gradually expanding from renting GPUs by the hour to providing inference services based on tokens. Reasoning and software services typically have higher profit margins, can enhance customer stickiness, and can also absorb old GPUs after long-term contracts expire, extending the commercial lifespan of devices. The advantage of doing so is that the profit margin of selling services is obviously higher than when renting GPUs. Lastly, I consider CoreWeave as one of the strongest observational indicators of AI infrastructure landscape. This financial report indicates that the demand for AI computing power still exceeds the supply, with strong demand for Rubin and no significant price collapse for old GPUs such as A100 and Hopper. The capital market is still willing to consider GPUs and long-term computing power contracts as assets that can be financed. Among them, the most direct beneficiary is still Nvidia. CoreWeave requires the purchase of more Nvidia GPUs, networks, and software for each increase in Backlog and Active Power. CoreWeave's own profitability is also affected by electricity costs, construction progress, debt interest rates, and customer prices, while Nvidia typically recognizes revenue at the time of GPU delivery. For industry chain companies such as optical interconnection, power equipment, liquid cooling, HBM, SSD, etc., this financial report also provides positive demand signals, but specific investments still need to consider supplier share, expansion speed, delivery capacity, price changes, and technological routes. CoreWeave can verify the overall industry demand, but cannot directly prove that every supplier can achieve the same revenue and profit growth. Big enough. Nebius needs to prove that it can convert industry prosperity into better shareholder returns through a healthier balance sheet, lower financing costs, and a higher proportion of software revenue.