In northern New Jersey, there is a structure that doesn’t appear to be the hub of anything. No logo illuminated against the skyline, no glass tower. Behind reinforced walls, there were only rows of servers humming, consuming enough power to cause anxiety in the local grid. That’s about where CoreWeave got its start, and it continues to be the kind of unglamorous setting that characterizes what this business truly does. It leases processing power. Massive quantities of it.
For the majority of the past two years, Nvidia and a few well-known companies dominated the AI discourse. CoreWeave remained in the background, mentioned infrequently by regular investors but frequently by analysts. Quietly, then suddenly, that changed. The stock has nearly doubled in the last year and is up more than 60% in 2026, surpassing the S&P 500 by an almost unfair margin. It’s difficult to avoid thinking of the old adage about gold rushes: those who became wealthy weren’t always the ones who dug. The shovels were sold. In this story, CoreWeave constructs the mines.
People are still taken aback by the founding story. The business started out as a cryptocurrency mining operation in 2017, which is the kind of endeavor that most serious investors learned to ignore. Then it made a sharp turn toward GPUs and the kind of infrastructure that AI labs were beginning to yearn for. By the time it made its debut on the Nasdaq in March 2025, the cryptocurrency’s beginnings seemed more like an odd bit of insight than a liability. There’s a feeling that the early bets were carried more by timing than by brilliance. It’s not a critique. That’s how most fortunes operate.
CoreWeave’s offerings are surprisingly straightforward. Businesses want to use Nvidia’s most potent chips without having to invest billions in building their own data centers. They get that access on demand from CoreWeave. AI labs seem to be aware of the technical advantage, which is high-bandwidth networking that enables thousands of GPUs to function as a single, massive machine. To support its Claude models, Anthropic inked a multi-year contract. In March, Meta committed approximately $21 billion. The backlog of contracts has grown to an almost absurd $100 billion.

However, the latest earnings revealed a more nuanced picture. On paper, the first quarter appeared fantastic, with revenue more than doubling year over year and bookings reportedly the highest in the company’s history. However, the following session saw a 11.4% decline in the stock. Investors ignored the growth in favor of focusing on the other factors, which included a still-bleeding bottom line, weaker second-quarter guidance, and rising capital spending plans. Whether that response was one of caution or panic is still unknown.
It’s not a subtle risk. This company is based on promises and debt. Billions of dollars were borrowed to increase capacity in order to fulfill contracts that require almost perfect execution, such as data centers going live on time, demand holding, and customers remaining focused and devoted. A securities lawsuit has already been filed, claiming the company minimized construction delays and exaggerated its capacity to meet demand. That doesn’t prove much on its own because lawsuits follow hot stocks like gulls follow fishing boats. It persists, though.
The potential benefits and risks of CoreWeave are not what make it truly intriguing. It’s that both are abnormally big at the same time. The same balance sheet is being read by both the bull and bear scenarios. Investors seem to think that the AI build-out is nearing its start rather than its finish, and if that’s the case, the business underneath all that processing power has plenty of room to grow. If it isn’t, the leverage that drove the ascent operates just as effectively in reverse if demand declines or if a single delayed data center cascades.
For the time being, servers in New Jersey and other places continue to hum despite the commotion. Anyone claiming certainty in either direction has probably not looked closely enough, and the story is far from settled.
