The cloud has been steadily tightening its grip on the artificial intelligence (AI) market, as the economics and complexity of training generative AI models have left most enterprises with little choice but to rely on hyperscalers. Large language models demand colossal computing power, built around graphics processing unit (GPU) clusters capable of handling the intense matrix calculations at the heart of neural networks with ultra-low latency. Yet GPUs remain both scarce and staggeringly expensive, with the cost of building and operating dedicated clusters quickly reaching into the millions once power, cooling, networking and specialist staff are factored in. For the majority of firms, establishing such infrastructure in-house is prohibitively costly and operationally complex.

Instead, a new generation of cloud providers, known as Neoclouds, emerged as the practical solution, offering GPU clusters as-a-service models and full AI development platforms on a pay-as-you-go basis. Firms such as CoreWeave, Nebius, and Fluidstack are reshaping the whole data centre landscape and are acting as a buffer between operators and hyperscalers. This evolution brings new challenges, including the need for rapid energisation and access to affordable power. Typically, these providers have leased space in regional hubs, emerging markets and tier 1 locations.

CoreWeave currently leads the field, buoyed by its March IPO. Notably, Europe is prominently represented among the top five Neoclouds, with Nscale and Nebius securing second and third place, respectively. It is also significant that NVIDIA holds a stake in four of the five leading firms, and that major players such as NVIDIA, Oracle, OpenAI, and Microsoft are involved in key partnerships across the sector. Given the rapid pace of industry change, emerging Neoclouds such as Vultr, Together AI, Voltage Park, and Fluidstack, among others, are poised to make an impact in the near future.

Hyperscalers are investing heavily in AI-specific chips, expanding their libraries of foundation models, and eyeing acquisitions of specialist GPU providers to strengthen their dominance. These moves suggest that even as the technology becomes more affordable, they will retain the upper hand through economies of scale and integrated service offerings. Hence, slowly but surely, the cloud is swallowing the AI market, leaving hyperscalers as the gatekeepers of generative intelligence.

This reliance on the cloud raises questions about data sovereignty. Sensitive sectors, such as healthcare and government, may remain cautious, concerned about compliance, sovereignty, and control. As a result, the European Commission’s decision in late October to launch a €180 million sovereign cloud tender seeks to guide the market towards greater alignment with EU standards and values. This will reshuffle the cards between market players. European providers may suddenly benefit from a structural tailwind, as their models already align with EU jurisdiction, open-source frameworks, and transparent supply chains. Conversely, US hyperscalers will need to adapt by multiplying sovereign regions, setting up local joint ventures with EU-based management, and maintaining independent control over encryption keys, all of which add cost and complexity. In the near term, this shift is likely to support capacity growth in established hubs and boost second-tier locations where governments aim to localise workloads.

>>  For more information, please contact our EMEA Data Centre Advisory team