Neocloud services surge as AI workload growth strains traditional datacenter capacity and economics

Neoclouds are specialized cloud providers offering GPU-as-a-Service (GPUaaS) designed to support compute-intensive use cases like AI training, machine learning inference, blockchain, gaming, and scientific modeling. According to JLL analysis, the global neocloud segment is projected to grow at an 82% compound annual growth rate (CAGR) between 2021 and 2025, as enterprises race to secure GPU capacity for AI workloads. These providers deliver on-demand access to graphics processing units and tailored service models that hyperscalers often cannot provide quickly enough, driven by unprecedented demand from AI infrastructure.

Unlike traditional hyperscalers, neoclouds focus on high-density GPU infrastructure. AI workloads draw over 100 kW per rack and require specialized cooling like immersion and floor loading capacities around 12-15 kN/m², pushing data centers to upgrade. As Muhd Syafiq, director of JLL's data center research in Asia-Pacific, explains, "Unlike traditional hyperscalers, neoclouds focus solely on these high-density needs, offering quicker deployment, tailor-made solutions for AI, and often more competitive pricing." The data center design is optimized for very high power density, with increasingly many designed from the outset for rack-scale systems like Nvidia NVL72. Alexander Harrowell, principal analyst at Omdia, notes that "the internal design of the data centers is optimized for very high power density to support them," and that these are "usually greenfield projects."

The neocloud category exists primarily because hyperscale infrastructure cannot currently keep pace with AI demand. "The biggest bottlenecks are having enough electricity and advanced cooling for powerful GPU hardware," Syafiq notes. Neocloud providers can get new sites up and running in months, offering a more cost-effective way to meet urgent demand than hyperscalers, which usually need years to build new capacity. Harrowell indicates that much investment comes from GPU vendors or AI labs themselves, explaining that "to some extent they can be seen as a kind of vendor financing for the GPUs." However, "even with fast deployment, finding and securing sites with the necessary high power, advanced cooling, and structural capacity requires deep local expertise," Syafiq adds.

Neoclouds offer significant cost advantages, with some reports indicating up to a 66% cost reduction for GPU instances compared to major hyperscalers. Their typical contracts run for two to five years, compared to the 10 to 15-year leases often seen with traditional data center clients. "Startups and research teams with unpredictable workloads often pick neoclouds for these significant savings and flexible terms," Syafiq explains. Rather than displacing hyperscalers, neoclouds will play a complementary role, focused on AI-heavy workloads. "By offering this dedicated and scalable GPU power, neoclouds overcome the limitations of general-purpose setups, giving AI teams rapid access to specialized resources that enable peak performance," Syafiq says. For investors, neoclouds carry a different risk profile with higher upfront capital requirements and shorter lease terms, but also the potential for higher rental rate premiums.