The continued growth of AI is reshaping the data center market, driving unprecedented demand for neocloud services that complement hyperscaler infrastructure, a new study reveals.
According to JLL analysis, the global ‘neocloud’ segment is projected to grow at a staggering 82% compound annual growth rate (CAGR) between 2021 and 2025, as enterprises race to secure GPU capacity for AI workloads.
Neoclouds are specialized cloud providers offering GPU-as-a-Service (GPUaaS) to enterprise customers. They are designed to support compute-intensive use cases like AI training, machine learning inference, blockchain, gaming, and scientific modeling.
They deliver on-demand access to graphics processing units (GPUs) and tailored service models that hyperscalers often cannot provide quickly enough.
Muhd Syafiq, director of JLL’s data center research in Asia-Pacific, says neoclouds provide flexible, cost-effective access to high-performance GPUs, specifically for intensive AI, scientific, and blockchain applications.
He adds that AI workloads, drawing over 100 kW per rack, require specialized cooling like immersion and floor loading capacities around 12-15 kN/m², which pushes data centers to upgrade.
“Unlike traditional hyperscalers, neoclouds focus solely on these high-density needs, offering quicker deployment, tailor-made solutions for AI, and often more competitive pricing,” Syafiq says.
Alexander Harrowell, principal analyst, advanced computing at Omdia, said neocloud providers are disrupting the industry as it grapples with supply and demand imbalances.
“They are attracting a lot of investment, although much of it comes from GPU vendors or AI labs themselves,” he told Data Center Knowledge. “To some extent they can be seen as a kind of vendor financing for the GPUs.”
Harrowell explains that the category of neocloud data centers exists primarily because of the demand for AI training or inference infrastructure.
“The product offering is heavy on GPUs, and the internal design of the data centers is optimized for very high power density to support them,” he says.
He adds that increasingly, they will need to be designed from the outset for rack-scale systems like Nvidia NVL72.
“It helps that they are usually greenfield projects,” Harrowell says.
GPU-specialized neoclouds offer cost savings compared to major hyperscalers, with flexible contracts ranging from two to five years instead of decade-long commitments.
GPU Availability Bottlenecks
The JLL report highlights that hyperscale infrastructure cannot currently keep pace with AI demand, creating bottlenecks in availability.
“The biggest bottlenecks are having enough electricity and advanced cooling for powerful GPU hardware,” Syafiq explains.
He says 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.
“However, even with fast deployment, finding and securing sites with the necessary high power, advanced cooling, and structural capacity requires deep local expertise,” he adds.
While neocloud adoption is accelerating, Syafiq notes that the rise of these specialized providers is not expected to displace hyperscalers.
Instead, they will play a complementary role, focused on AI-heavy workloads, while hyperscalers continue to provide the broad mix of compute and storage services preferred by many enterprises.
Specialized Infrastructure Driving Cost Efficiency
High-density GPU infrastructure, central to Neocloud’s GPUaaS model, efficiently provides the extreme processing power needed for intensive AI projects at a more cost-effective rate.
“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.
Neoclouds generally offer lower prices, with some reports indicating up to a 66% cost reduction for GPU instances compared to major hyperscalers. They also offer shorter, more flexible contracts.
Read more of the latest next-gen data center news
Their typical contracts run for two to five years, offering greater agility compared to the up to 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.