Why the Category Appeared When It Did

A constraint created this market. In late 2022 and through 2023, demand for accelerated computing ran ahead of supply. The public release of ChatGPT by OpenAI in November 2022 set off a race among model developers, and the race ran on Nvidia hardware. The chip that mattered most in that period was the Nvidia H100, and there were not enough of them. A buyer with a contract and a supply allocation could rent capacity almost as fast as it could install it.

Scarcity does two things to a market. It raises prices, and it rewards whoever can deliver first. The hyperscalers had the money and the customers, but they also had long internal planning cycles, large existing fleets and many priorities competing for the same power and floor space. A smaller operator with one priority could move faster. That speed was the neocloud's first product.

Three inherited assets made the speed possible. The first was operational. Cryptocurrency miners already knew how to buy large quantities of specialised hardware, pack it densely and run it hard. CoreWeave began as an Ethereum mining operation. Crusoe began by capturing gas that oil producers would otherwise have flared and using it to power mining containers in the field, a business it sold to New York Digital Investment Group, or NYDIG, in March 2025 to concentrate on artificial intelligence. The skills transferred; the chips did not.

The second inherited asset was power. Mining operators had signed power contracts and built sites in places where electricity was cheap and available. Some of those sites could be converted to accelerated computing. That is why CoreWeave leased capacity from Core Scientific, a mining company, and why in July 2025 it agreed to buy Core Scientific outright in an all-stock deal. Core Scientific's shareholders voted the deal down on October 30, 2025, and the merger agreement was terminated. The logic of the attempt matters more than the outcome: a neocloud without its own power has to rent it from someone who has it.

The third asset was a supplier with a reason to help. Nvidia sells to the hyperscalers, but the hyperscalers also design their own chips, such as Google's Tensor Processing Unit and Amazon's Trainium. A healthy population of independent clouds that buy only Nvidia hardware diversifies Nvidia's customer base and reduces its dependence on buyers who are also competitors. Nvidia invested in CoreWeave before its initial public offering and anchored the offering itself. It participated in Nscale's funding round in 2025. It has also become a customer of the clouds it supplies, a relationship the next module examines closely.

Put those three together and the category explains itself. A supply shortage created a premium for speed. Mining firms supplied the operating skill and some of the power. Nvidia supplied chips and, at times, capital. And a small number of very large buyers, chiefly the model developers and Microsoft, supplied contracts big enough to finance the whole arrangement.

Notice what the model does not require. It does not require the neocloud to write better software than the hyperscalers, though several claim to. It does not require a new chip. It requires access to three scarce things at once: chips, power and a creditworthy buyer. The neocloud is an assembler of scarcity. Its margin is the rent on having assembled them faster than anyone else.

That also tells you what erodes the margin. If chips become plentiful, speed of access stops commanding a premium. If power becomes the binding constraint, as it has in many markets, the advantage shifts to whoever controls sites and interconnection rights, which is why Crusoe's Abilene development matters. If the large buyers build enough capacity of their own, they need fewer outside suppliers. Each of those shifts is visible in public data: Nvidia's delivery schedules, utility interconnection queues and hyperscaler capital spending guidance.

Faraday's read: the neocloud appeared because one input was scarce and stayed because a second input, power, became scarcer. A durable operator is one that has moved its advantage from the first scarcity to the second before the first one ends.