The AI HyperScaler: CoreWeave

The Crypto Miners Who Became the Backbone of the AI Era

In 2017, two commodities traders whose hedge fund had just collapsed bought a stack of graphics cards on a whim, set up a mining rig in a New Jersey garage, and started producing Ethereum. Eight years later, their company went public at a $35 billion valuation, count Microsoft, OpenAI, Meta, and Anthropic among their largest customers, and operate a contracted backlog of $66.8 billion. That company is CoreWeave, and its rise tells you almost everything about why the AI revolution is so much harder than it looks.

The problem CoreWeave solves is invisible to most executives but existential for anyone building modern AI. Training a large language model requires tens of thousands of specialized chips, mostly Nvidia GPUs, wired together with extreme-bandwidth networking, fed with prodigious amounts of power and cooling, and orchestrated by software designed for this exact workload. The big cloud providers, Amazon Web Services, Microsoft Azure, and Google Cloud, were built fifteen years ago to run general business software. Their architectures, billing systems, and data centers were designed for millions of small workloads, not a single training job that occupies thousands of chips for months.

That mismatch left a trillion-dollar hole in the market. The hyperscalers had the customers but not the right architecture. Specialized labs had the architecture but not the scale or the capital. CoreWeave built a third thing, a cloud platform purpose-built for one purpose only, running massive AI workloads at full speed with software, networking, and data centers redesigned around the chip rather than the other way around. It is a different category of infrastructure altogether. The market calls it a #neocloud, and CoreWeave invented it.

How It Started

The story begins with a hedge fund that did not work out. Michael Intrator and Brian Venturo ran Hudson Ridge Asset Management together, an energy-focused fund where they used a machine learning model to pick investments. Their data partner was a young analyst named Brannin McBee. When the U.S. fracking boom flattened the energy markets they traded, Hudson Ridge closed and the three found themselves with what Venturo later described to TechCrunch as "a lot of time on our hands."

They became fascinated by cryptocurrency and bought a single GPU to understand how mining worked. It paid for itself in days. They bought another, then another, then thousands. Operating from Venturo's grandfather's garage, they incorporated Atlantic Crypto in August 2017. When #Ethereum prices crashed in 2018 and the company nearly went under, they pivoted to renting their GPU clusters to visual effects studios for film rendering. In 2019, Peter Salanki, who comes from a software engineering and infrastructure background joined to build and scale the technical platform and they raised a $1.2 million seed round. In 2021 they renamed the company CoreWeave, meaning the weaving together of compute cores into a fabric.

The breakthrough came almost by accident. In 2022, CoreWeave gave free GPU access to EleutherAI, an open-source AI research group, hoping to learn how machine learning infrastructure worked. As Venturo put it, "we thought we were just going to learn how the infrastructure worked." Instead, EleutherAI introduced them to hundreds of AI startups. Stability AI became a paying customer. Then ChatGPT launched in November 2022, and the entire technology industry suddenly needed exactly what CoreWeave had built.

The AI Driven Innovation

The simplest way to understand CoreWeave is this. The big cloud providers were designed to run thousands of small computing tasks for businesses, like email, websites, payroll, and customer databases. Modern AI is the opposite kind of workload, a single enormous job that ties together thousands of specialized chips for weeks at a time, demanding extreme speed and uninterrupted communication between every chip. Trying to run an AI training job on a general-purpose cloud is like trying to run a Formula 1 race on a city street. It works, but slowly and inefficiently.

CoreWeave built a cloud designed only for that race. Its data centers are configured for the high power density and liquid cooling that AI chips require, its networks deliver chip-to-chip communication far faster than general clouds, and its software is built around managing massive GPU clusters rather than virtualized servers. The strategic implication is significant. A model that takes a month to train on AWS may take a quarter the time on CoreWeave at meaningfully lower cost, which is why the most demanding AI customers in the world signed multi-year contracts to run their most important work on it.

Before CoreWeave existed, building a frontier AI model meant queueing for limited capacity at the hyperscalers, accepting performance penalties from infrastructure built for a different purpose, or assembling your own data centers, which only Microsoft, Google, Meta, and Amazon could afford. With CoreWeave, even an OpenAI or an Anthropic could rent specialized capacity at frontier scale. That single shift, from infrastructure being a constraint to infrastructure being a service, is one of the reasons AI development accelerated so dramatically between 2023 and 2026. CoreWeave did not just sell faster computers. It removed the infrastructure ceiling that had kept frontier AI confined to a handful of companies.

The Strategic Landscape

The numbers tell the story. CoreWeave's revenue grew from $16 million in 2022 to $229 million in 2023 to $1.92 billion in 2024, a 737 percent jump in a single year. Trailing twelve-month revenue at the end of 2025 hit $5.13 billion, and the company guided to $12 to $13 billion in revenue for 2026, with a contracted revenue backlog of $66.8 billion. CoreWeave went public on Nasdaq in March 2025, the largest U.S. tech IPO since 2021, and as of late April 2026 trades at a market capitalization of roughly $57 billion. The customer roster is the punchline. Microsoft accounted for over 60 percent of 2024 revenue. OpenAI signed contracts totaling $22.4 billion, Meta added $14.2 billion, and in April 2026 Anthropic signed a multi-year deal alongside another $21 billion expansion of CoreWeave's Meta partnership.

The strategic position is rare and possibly unique. CoreWeave is not really competing with AWS, Azure, and Google Cloud. It is competing for a slice of demand the hyperscalers cannot fully serve themselves, even as Microsoft remains its largest customer. Its true competitors are other specialized GPU clouds, including Lambda, Crusoe, and Nebius. What CoreWeave has that competitors do not is a deep relationship with Nvidia, which has invested directly in the company multiple times, including a reported $2 billion in January 2026, and consistently ships its newest chips to CoreWeave first. CoreWeave was the first cloud provider to deploy Nvidia's GB200 NVL72 chips in February 2025 and the GB300 NVL72 in July 2025. That preferential allocation, combined with software, data center expertise, and a four-year head start, is the moat.

CoreWeave operates more than 32 data centers and is investing $30 to $35 billion in 2026 alone to more than double its active power capacity. Each gigawatt of new capacity creates demand for electricians, mechanical engineers, network specialists, liquid cooling technicians, and AI infrastructure operators, jobs that barely existed in their current form three years ago. The neocloud category that CoreWeave created now supports an entire ecosystem of AI startups that simply could not have existed without it.

What’s Next

Next week, we may look at another AI-native company that is not just optimizing an old workflow but expanding what can be built in the first place.

Send a company you think belongs in that category, or share this with a founder building where the market is headed, not where it has been.