CoreWeave Stock Surges 11% After Meta and Anthropic Deals — But the Balance Sheet Tells a Different Story

Row after row of Nvidia H100 GPUs sit behind locked cages in a data center somewhere in northern New Jersey, humming at a frequency that indicates significant sums of money are passing through them. The temperature is controlled and the air is cold. There is a huge power draw. Additionally, this infrastructure is pre-purchased, pre-scheduled, and part of a backlog that currently exceeds $87 billion.

It is committed to companies like Meta, Anthropic, and OpenAI for years to come, according to a contract filed with the SEC. When CoreWeave went public fourteen months ago at $40 per share, it signed deals that would have seemed like science fiction. That is the story of CoreWeave in its most tangible form: a company that created a specialized rental empire for the most significant hardware of this decade.

CoreWeave announced a $21 billion extension of its current agreement with Meta Platforms on April 9, 2026, increasing the total amount of money committed to Meta through December 2032 to $35.2 billion. In less than a day, the business announced a multi-year contract with Anthropic to operate Claude at production scale, securing a client from one of the top ten AI model suppliers.

Shares had increased by about 11% by April 11th, trading at $102 on 83 million shares, more than three times the daily average. When two significant announcements are made in a 48-hour period, the market reacts by quickly repricing and then raising questions.

Because it highlights a paradoxical aspect of how the AI economy is structuring itself, the Meta deal is worth considering. In 2026, Meta intends to invest between $115 billion and $135 billion in capital projects. The business runs gigawatt-scale data center campuses throughout Europe and the United States. It has the MTIA, a custom AI accelerator chip, and the engineering depth to create almost anything. And yet.

It recently paid a specialized cloud vendor $35 billion. The most likely explanation isn’t that Meta can’t afford to construct its own GPU clusters—it can, of course—but rather that it lacks the time to do so. Tens of thousands of GPUs in particular configurations at particular times are needed to train the next generation of large language models. Because CoreWeave has already acquired the silicon and constructed the infrastructure, it can deliver that capacity in a matter of months. Timelines for internal capacity are expressed in years. The gap is filled by the rental market.

The situation at Anthropic is structurally similar but slightly different. Early in 2026, the company’s yearly recurring revenue increased from about $9 billion to over $30 billion. Every year, more than 1,000 business clients pay more than $1 million for Claude services. This velocity necessitates inference capacity at a scale Anthropic is unable to construct on its own, and it must be dependable, quick, and accessible on architectures for which its models are optimized. That is precisely what CoreWeave’s Nvidia-native clusters, which include early deployments of the new Vera Rubin platform, provide. Not only does the deal increase revenue, but it also adds revenue that compounds due to production workloads with phased rollout structures and expansion options.

All of this creates real excitement for investors, which immediately leads to a set of financial metrics that are more difficult to feel at ease with. For every dollar of revenue, CoreWeave spends about $2.60. An estimated $30 to $35 billion will be spent on capital projects in 2026. The current debt-to-equity ratio is approximately 4.5. There is a $16.67 negative free cash flow per share. The company recently issued $1.75 billion in senior notes and $3.5 billion in convertible notes at a coupon rate of 9.75%, which indicates what lenders believe the risk profile to be. Despite revenue growing 110% year over year, Q4 2025 earnings fell $0.28 per share short of analyst expectations. In the hope that contracted revenue will eventually surpass the cost of constructing the infrastructure to deliver it, the business model necessitates a significant upfront investment. The wager is currently being played out. However, the debt obligations are real and there is little room for error.

To be honest, the analyst community is dispersed. Cantor Fitzgerald set a price target of $149 to start coverage. $120 is held by Evercore ISI. At $110, Stifel has a hold. Sanford C. Bernstein has an Underperform rating of $56, which suggests that the current price already accounts for results the business hasn’t yet produced. ARK Invest, owned by Cathie Wood, has been purchasing. Executives have been selling; days prior to the Anthropic announcement, COO Sachin Jain sold shares at $92 as part of a prearranged plan, and insider Brian Venturo sold $5.51 million worth of stock on the same day. Because both transactions were prearranged under Rule 10b5-1 plans, they are legally shielded from allegations of insider trading. However, as usual, the timing draws attention.

It’s difficult to ignore the fact that CoreWeave’s stock chart—$40 at IPO, up to $187 by June 2025, down to $33 at the 52-week low, and back above $100 now—tells the tale of a market that is genuinely unsure of the company’s worth. There is no denying the increase in revenue. There is an unprecedented backlog. The client list resembles a who’s who of AI infrastructure expenditures. Whether CoreWeave matters is not the question. It obviously does. The question is whether the current price is predicated on a future that will materialize on time, free from competitive pressure from hyperscalers developing their own capacity, without hardware obsolescence speeding up depreciation, and without the debt load causing financial instability at the wrong time. The next part of that response will be provided during the May 13 earnings call.