Cloud is rebalancing toward specialists
The market is shifting away from a pure hyperscaler oligopoly toward specialist AI clouds and service operators that offer rapid access to scarce GPU capacity and bespoke services. (siliconangle.com) That trend is creating a permanent multi‑venue layer—self‑built campuses, neoclouds and colocation—that makes a hybrid control plane more valuable if it can actually broker across those environments. (globaldatacenterhub.com)
The old cloud story was simple: three giant companies built giant regions, and everyone else rented space inside them. In April 2026, Nutanix used its.NEXT conference in Chicago to argue that customers now want a mix of infrastructure rental and managed artificial intelligence services, and that service providers are gaining ground in that gap. (siliconangle.com) That shift is showing up because the scarce part of artificial intelligence is no longer just software. Nvidia’s Grace Blackwell rack packages 72 Blackwell graphics processors in one liquid-cooled system, which pushes cloud competition toward whoever can secure chips, power, cooling, and floor space fastest. (nvidia.com) The specialist clouds built around that bottleneck are no longer tiny side players. CoreWeave told investors its revenue jumped from $229 million in 2023 to $1.9 billion in 2024, then to $5.13 billion in 2025, which is the kind of growth you get when customers care more about getting graphics processors next month than about buying every service from one vendor. (sec.gov, stockanalysis.com) This is why “neocloud” became a real category instead of a buzzword. Nutanix said on April 7 that it is extending its Agentic Artificial Intelligence software so these providers can sell graphics-processing-as-a-service, Kubernetes container orchestration, and managed artificial intelligence platforms instead of only raw servers. (nutanix.com) The backdrop is not that the hyperscalers are fading. Jones Lang LaSalle said in its 2026 outlook that nearly 100 gigawatts of new data center capacity will be added between 2026 and 2030, while CBRE said European hyperscaler self-build capacity is still expected to grow 24% in 2026. (jll.com, cbre.co.uk) What is changing is the shape of the market around them. Global Data Center Hub wrote on April 10 that self-build will not kill colocation, but will split the market between commodity space that gets squeezed and premium capacity that still earns strong pricing. (globaldatacenterhub.com) That creates a permanent multi-venue map for computing. One company may train a model in a self-built campus, burst into a specialist graphics processor cloud for a launch, and keep regulated data in a colocation site in another country because sovereignty and latency rules do not disappear just because one vendor has a nice console. (globaldatacenterhub.com) The awkward part is that every venue speaks a slightly different operational language. Nutanix spent.NEXT 2026 pitching a control layer that spans private infrastructure, public cloud, Kubernetes clusters, and new artificial intelligence services, because the valuable product is increasingly the broker that can move workloads across all of them without breaking security or budgets. (virtualizationreview.com, computerweekly.com) The specialist operators still have hard limits. Data Center Knowledge reported this week that some artificial intelligence cloud providers are being turned away from data center deals even with 15-year leases and upfront cash, because landlords now worry about credit risk as much as price. (datacenterknowledge.com) So the new cloud order looks less like one mall and more like a logistics network. The hyperscalers still own huge highways, but the fastest-growing opportunities are in the depots, brokers, and specialist carriers that can find power, secure graphics processors, and stitch together self-build campuses, neoclouds, and colocation into one working system. (siliconangle.com, globaldatacenterhub.com, jll.com)
