Three quarters of organisations expected to use specialised AI cloud services

Three quarters of organisations are expected to use specialised accelerated cloud computing services by 2026 to support artificial intelligence (AI) and other demanding applications, a forecast by research firm IDC continues to uphold, although the figure may prove conservative. Dave McCarthy, group vice-president of cloud and data centre infrastructure at IDC, said the prediction, first published in October 2024, remained valid as businesses expanded their AI capabilities and sought alternatives to traditional cloud providers. The forecast formed part of IDC’s FutureScape Worldwide Cloud 2025 Predictions and reflects organisations at different stages of adopting AI, rather than companies moving 75 per cent of their workloads to the technology. McCarthy attributed the expected uptake to three main factors, namely application modernisation, the growing demands of AI workloads and businesses’ confidence in cloud providers as strategic technology partners. Companies that have spent years migrating systems to the cloud are increasingly revisiting their infrastructure decisions and updating applications, a process that now extends beyond moving legacy systems out of physical data centres. AI has accelerated this shift because it requires different computing capabilities from conventional business applications. “The type of services that you would use for an AI application look very different than the ones that you would use for a more traditional general purpose workload,” McCarthy said. However, adoption has not followed a single path. While public cloud remains a major destination because it offers easy access to the latest hardware and software, concerns about data privacy and sensitive corporate information have pushed some AI workloads into private cloud environments. The market has also expanded beyond the three major hyperscale cloud providers, as specialised providers have developed the tools and support needed by enterprise customers. Initially, many of these companies supplied hardware to customers with the expertise to operate it themselves, limiting their appeal beyond major AI laboratories. They have since strengthened their services, while capacity shortages at established providers have encouraged businesses to consider alternatives. IDC research published in July 2026 found that 42 per cent of European respondents cited high performance for specialised workloads as their main reason for using or considering neocloud services, based on a survey of 493 participants. Providers including CoreWeave, Nebius, Vultr, Lambda and Crusoe each attracted mid-single-digit shares when organisations identified suppliers important to their cloud strategies over the following two years, although established hyperscalers remained dominant. Neoclouds generally offer lower prices for graphics processing units (GPUs), but have fewer data centre locations and are less suitable for conventional IT workloads, IDC said in June 2026 guidance. “I still feel the number is fairly confident. Maybe it’s a little bit low in some ways,” McCarthy said of the 75 per cent forecast. IDC’s September 2026 research found that the global public cloud infrastructure-as-a-service market grew by 24.6 per cent to $172 billion in 2025, as AI workloads entered production and increased demand for scalable computing infrastructure. Looking ahead, McCarthy expects spending to shift increasingly towards AI inference, the process of running trained models to generate responses and perform tasks. “When you hear about AI agents, that’s all inference,” he said. Inference requires infrastructure closer to users and the locations where data is generated, encouraging investment in specialised chips and distributed computing capacity. Some companies previously associated with content delivery networks are also repositioning themselves as AI inference cloud providers. IDC expects sovereignty and control over AI infrastructure to remain important, with earlier forecasts suggesting that 50 per cent of organisations would seek network infrastructure providers offering sovereign controls by 2026, while 80 per cent of chief information officers would rely on cloud providers’ edge services for generative AI inference by 2027. Finally, McCarthy advised businesses to design their infrastructure for flexibility and regularly reassess their options, as the cloud choices made five or six years ago may no longer suit their requirements.