CoreWeave is the arms dealer

In 48 hours this week, CoreWeave signed a $21 billion expansion with Meta and a multibillion-dollar contract with Anthropic. With those two deals, nine of the ten leading AI model providers now run on CoreWeave's infrastructure. The company isn't building the models everyone's talking about. It's selling the compute those models need to exist. That distinction matters more than it seems.

The week that made it undeniable

On April 9, CoreWeave and Meta announced a $21 billion expanded agreement for AI cloud capacity through December 2032. The deal builds on a $14.2 billion arrangement from September 2025, bringing Meta's total committed spend with CoreWeave to roughly $35 billion. Meta will use the capacity to scale inference workloads, with some of the earliest deployments of NVIDIA's Vera Rubin platform. The very next day, CoreWeave announced a multi-year agreement with Anthropic to power the development and deployment of Claude. Financial terms weren't disclosed, but CoreWeave CEO Michael Intrator called it a "multibillion-dollar contract" in a Bloomberg interview. The stock jumped over 11%. Two landmark deals in two days, with two of the most important players in AI. That's not a coincidence. It's a signal.

Not a cloud company, a utility

The traditional framing for what CoreWeave does is "picks and shovels." Every gold rush has its tool sellers, and all that. But the scale here makes the analogy feel quaint. CoreWeave generated $5.13 billion in revenue in 2025, up from $1.92 billion in 2024, which was itself up from $228.9 million in 2023. That's not a startup selling shovels at the general store. That's an infrastructure utility being wired into the foundation of the AI economy. The company now sits on a $66.8 billion revenue backlog of non-cancellable contracts. It operates 1.6 gigawatts of contracted power capacity across its data centers. It provides GPU cloud infrastructure to almost every major AI lab in the world. When nine of ten frontier model providers depend on the same platform, you stop being a vendor. You start being a utility.

The AWS parallel

This playbook has precedent. In the early 2000s, Amazon built internal tools to manage its own computing infrastructure. It then realized other companies needed the same thing and launched AWS in 2006. The bet wasn't on any specific application. It was on the idea that compute itself would become the bottleneck, and whoever solved the infrastructure problem would capture enormous value. AWS is now four times larger than Amazon's original shopping business. The apps that run on AWS have come and gone, pivoted, merged, and died. The infrastructure layer endured. CoreWeave is making a similar bet, but in a narrower, higher-stakes corridor. Where AWS bet on general cloud compute, CoreWeave is betting specifically on GPU-accelerated AI workloads, the most capital-intensive and fastest-growing segment of computing. The thesis is the same: infrastructure players win long-term, not the apps.

The origin story nobody expected

CoreWeave's path to this moment is one of the stranger pivots in recent tech history. The company was founded in 2017 as an Ethereum mining operation. When Ethereum moved to proof-of-stake in 2022, rendering GPU mining unprofitable, CoreWeave repurposed its entire GPU fleet toward AI workloads. The timing was extraordinary. ChatGPT launched weeks later, and demand for AI compute exploded. CoreWeave went from mining cryptocurrency to powering the world's most advanced AI models in less than two years. It IPO'd on the Nasdaq in March 2025 at $40 per share, raising $1.5 billion in the largest U.S. tech offering since 2021. Sometimes the best business strategies are forced pivots that happen to land exactly where the world is headed.

Scaling is expensive

Intrator said it plainly on CNBC this week: "Scaling is expensive." And he's not wrong. CoreWeave posted a net loss of $1.2 billion in 2025 despite its explosive revenue growth. The company carries significant debt and faces massive capital expenditure requirements to fulfill its contracted commitments. S&P has CoreWeave rated B+ with a positive outlook, acknowledging the revenue visibility from its backlog while flagging the risk that "continued aggressive growth could delay improvement in credit metrics." This is the fundamental tension of the infrastructure play. The demand is clearly there, but meeting it requires spending enormous amounts of capital upfront, with payoff stretched over years. CoreWeave is essentially financing the AI revolution on credit, betting that the contracts in hand are worth more than the debt required to fulfill them. When AI margins eventually compress, and they will, the question becomes who absorbs the cost. If model providers start negotiating harder on compute pricing, CoreWeave's margins get squeezed from above. If GPU costs don't come down fast enough, they get squeezed from below.

The concentration problem

The most interesting risk in CoreWeave's model isn't technical. It's structural. Microsoft accounted for approximately 67% of CoreWeave's 2025 revenue. Meta, with its $35 billion in total commitments, is likely the second-largest customer. Add Anthropic and the picture becomes clear: CoreWeave's fortunes depend heavily on a small number of hyperscale buyers. This concentration cuts both ways. Right now, with GPU supply still constrained and AI demand accelerating, CoreWeave has leverage. These companies need compute faster than they can build it themselves. CoreWeave fills the gap. But over time, as GPU availability improves and these customers expand their own data center capacity, the leverage could shift. Microsoft is already building its own AI infrastructure at massive scale. Meta's entire AI strategy depends on eventually owning more of its compute stack. The risk isn't that CoreWeave loses these customers overnight. It's that it gradually transitions from strategic partner to supplemental provider. There's also a systemic angle. When one infrastructure company powers nearly every major AI lab, any disruption, whether technical, financial, or regulatory, ripples across the entire ecosystem. That kind of concentration creates fragility that no single contract can hedge against.

Fragile or fortress?

The bull case for CoreWeave is straightforward: AI compute demand will continue growing faster than anyone can build, and CoreWeave has locked in $66.8 billion in non-cancellable contracts that provide years of revenue visibility. The company is building on NVIDIA's latest platforms, acquiring companies like Weights & Biases to move up the stack, and diversifying its customer base with every new deal. The bear case is equally clear: CoreWeave is a capital-intensive, debt-heavy business with extreme customer concentration, operating in a market where its largest customers are also its most likely future competitors. The non-cancellable contracts provide revenue floors but also capex obligations. And the entire model depends on continued GPU scarcity, a condition that won't last forever. The truth is probably somewhere in between. CoreWeave has built something genuinely impressive, a GPU cloud platform trusted by almost every frontier AI lab in the world. But the durability of that position depends on whether the company can evolve fast enough to remain essential as the market matures.

The real lesson

What CoreWeave's week really illustrates is a pattern that keeps repeating in technology: distribution and infrastructure matter more than product. The companies building AI models get the headlines. They demo impressive capabilities, attract the talent, and generate the hype. But underneath all of it, someone has to provide the raw compute. Someone has to build and maintain the data centers, negotiate the power contracts, rack the GPUs, and keep the systems running at scale. That someone increasingly is CoreWeave. Not because it's the cheapest or the most innovative, but because it showed up early, scaled fast, and made itself hard to replace. In every platform shift, the infrastructure layer tends to consolidate before the application layer does. The arms dealers often do better than the armies they supply. CoreWeave is betting its entire business on that historical pattern holding true one more time. Given the numbers this week, it's a bet worth watching.