● Their success has been built on three major assets: datacenters that were already operational, cheap power contracts and technical expertise in the management and cooling of high-density computing infrastructure.
● Leveraging special partnerships with Nvidia, they now play a critical role in the training of AI systems, but doubts remain about the long-term sustainability of their business model.
A massive upsurge in the global need for computational power prompted by the boom in AI has come as a godsend to a new generation of fast-growing companies: neocloud providers, some of which started out in the crypto-mining industry, are transforming the market with offers of specialist solutions that are more flexible than those offered by hyperscalers. As Orange Business cloud expert Nicolas Bugault explains “these companies have taken advantage of their high-density computing infrastructure to provide unprecedented access to GPU hardware with an agility that has called into question established business models.” Neocloud providers are distinguished by their capacity to combine technical specialisation and rapid deployment with an innovative economic model. It’s a streamlined approach that is optimised for the most demanding computing workloads. And it is this difference that explains why players like Mistral and Anthropic call on neoclouds to provide the computing power they need to train their AI models.
The transition was supported by datacenters that were already operational, high-capacity power contracts and expertise in the management of high-density computation infrastructure.
From crypto mining to AI
“Many of these players have roots in crypto mining,” points out Nicolas Bugault. In response to volatility and dwindling profitability in that industry, they looked to Ai to make a fresh start. “The transition was supported by three pillars: datacenters that were already operational, high-capacity power contracts in regions with cheap energy, and most importantly of all, cutting-edge expertise in the management of high-density computation infrastructure.” French HPC services company Sesterce is an apt illustration of this transition. In February 2025, the company announced the investment of 500 million dollars in a datacenter in France. “While traditional hyperscalers require a minimum of six months to commission a new datacenter, neocloud companies can get the job done in just a few weeks.“ Underpinning this agility is a mastery of advanced cooling technologies, notably direct-to-chip liquid cooling and immersion cooling, which they initially developed while running mining farms. Temperature control provided by both techniques is and will remain a necessity for current and future GPUs (Blackwell, Rubin, Feynman). Whereas AWS, Google Cloud and Microsoft Azure target a diverse clientele with a wide-ranging portfolio of AI solutions, while neoclouds focus exclusively on the most demanding workloads: training large language models and high-performance computing applications.
Special partnerships with Nvidia
Privileged relationships with Nvidia have also played a key role in the success of neoclouds. “Collectively neoclouds account for annual sales of close to a million GPUs, which makes them the second largest market segment just after the hyperscalers,” points out Nicolas Bugault. This buying power commands them special access to the manufacturer’s latest hardware innovations: not only does Nvidia grant them priority access to the most recent graphics accelerators such as the B200, but it also collaborates with neoclouds on the development of optimised software solutions. “It is much more than a simple commercial relationship. Nvidia provides them with advanced virtualization layers that allow them to fully exploit the potential of these GPUs, which is critically important to the profitability of their massive investment in hardware.” In Europe, this dynamic has led to the emergence of compute farms housing 20,000 to 50,000 GPUs, which are operated by a select group of players such as Fluidstack and CoreWeave. This infrastructure has enabled neoclouds to develop optimised computing environments, which fully maximise the energy efficiency and computing power of components. “Neoclouds also serve as a testbed for new technologies, which enables Nvidia to validate new hardware with real-world stress testing.”
A high-risk but strategic business model
There are inherent risks in the neocloud business model: “ Some of these players rely on a single major client for over 50% of their business, but they are able to present investors with long-term contracts with key players like Microsoft or OpenAI, which considerably reassures the financial markets. There are already early signs of oversupply in the market for access to previous generations of GPUs, which has led neoclouds to diversify their customer base and target the B2B market.” In the long term, this shift could democratise access to high-performance computing infrastructure, which is currently beyond the reach of medium sized companies that are unable to pay for prohibitively expensive hyperscale solutions. “Whatever happens we can say with certainty that these players have already changed the rules of the game. The real challenge they now face will be to transform this initial disruption into a sustainable business model,” concludes Nicolas Bugault.
Nicolas Bugault
