NeoClouds: New Powerhouses of AI Infrastructure

Jim Carroll

6 months ago

Defining the NeoCloud Era

A new class of infrastructure provider—the NeoCloud—has emerged to power the exponential growth of artificial intelligence. These companies sit between hyperscalers and traditional colocation operators, purpose-built to deliver GPU-accelerated compute for AI training, inference, and high-performance computing (HPC).

Instead of broad, multi-service portfolios like AWS, Azure, or Google Cloud, NeoClouds specialize in GPU-as-a-Service (GPUaaS)—bare-metal or near-metal access to the latest NVIDIA and AMD accelerators with InfiniBand or NVLink interconnects and AI-optimized software stacks. Their business models hinge on rapid deployment, energy efficiency, and cost predictability—offering developers queue-free compute at two to seven times lower cost than general-purpose hyperscale cloud instances.

Tremendous Investment and Rapid Build-outs

The rise of NeoClouds mirrors the surge of global AI data-center investment. Dell’Oro Group reports that AI-optimized servers already account for roughly one-third of worldwide data-center spending and could surpass half of all infrastructure investment by 2029. IDC forecasts that cloud infrastructure spending will grow more than 30 % year-over-year in 2025, largely driven by GPU-rich AI configurations.

This explosive trajectory is fueled by tens of billions in new financing—both equity and debt—as NeoClouds rush to secure GPUs, power, and space. Many have transitioned from crypto-mining or rendering backgrounds to become full-scale AI infrastructure specialists. Others are building multi-gigawatt campuses tied to long-term renewable power contracts, betting that AI workloads will demand both density and sustainability.

A Fast-Evolving Market and Business Model

The NeoCloud business model evolves almost monthly. Some players operate as publicly traded data-center REITs, funding massive, energy-integrated campuses. Others concentrate on AI orchestration software, offering managed clusters or LLM-specific tooling atop their GPU infrastructure.

Despite varied strategies, three common pressures shape the field:

Hardware access – early allocation of scarce GPUs (Hopper, Blackwell, Rubin).
Networking and cooling – low-latency fabrics and liquid-cooled designs to sustain cluster performance.
Utilization discipline – balancing rapid capacity growth with stable demand and contract bookings.

A Big List of NeoClouds (October 2025)

The following table provides a ranked snapshot of the major NeoCloud providers by estimated GPU capacity. It highlights their funding, infrastructure footprint, and geographic reach—illustrating how each contributes to the AI compute backbone.

Company	Focus / Stack / Partners	Funding / Status	GPU Capacity & Major Locations*
CoreWeave	GPUaaS for AI training, inference, and VFX; bare-metal with InfiniBand/NVLink; partners: OpenAI, Microsoft, NVIDIA, IBM.	$12.7 B total (equity + debt); NVIDIA strategic investor.	≈ 250 k H100 equiv (claimed); 32 sites across US, UK & EU.
Lambda Labs	GPU cloud + servers; bare-metal clusters to 1 k GPUs; partners: NVIDIA, Meta, Hugging Face.	$500 M+ funding; ≈ $1.5 B valuation.	≈ 50 k GPUs (est.); US (SF, TX).
IREN (Iris Energy)	Renewable-powered GPUaaS; liquid-cooled; partners: NVIDIA, AMD, Poolside AI.	Public (NASDAQ); $674 M GPU capex 2025.	23 k GPUs (disclosed); US (TX), Canada (BC).
Crusoe	Low-carbon GPU DCs using stranded gas and renewables; partners: Oracle, NVIDIA, Microsoft.	$1.26 B+ funding (incl. debt).	≈ 40 k GPUs (est.); US (ND, TX, WY), Iceland planned.
Nebius	EU-centric GPU platform with custom HW for sovereign AI; partners: Microsoft, NVIDIA, Accel.	$700 M+ funding; private.	≈ 30 k GPUs (est.); Finland, France, US, Iceland.
Vultr	Full-stack cloud offering GPUaaS; developer-centric; partners: NVIDIA, StackPath.	$146 M equity; $500 M+ debt.	≈ 20 k GPUs (est.); 32 regions worldwide.
Fermi America	Energy-integrated AI campuses (SMR/wind/solar); partners: Texas Tech, NuScale, NextEra.	Public (IPO Oct 2025); pre-ops.	Pre-deployment; Amarillo TX + planned NM sites.
Nscale	Nordic hydro-powered AI clusters for sovereign compute; partners: OpenAI/Aker, NVIDIA.	$200 M+ funding (2025).	≈ 10 k GPUs (est.); Norway & Sweden.
Paperspace (DigitalOcean)	MLOps + GPU instances within DigitalOcean; partners: NVIDIA, Hugging Face.	Acquired 2023 by DigitalOcean.	≈ 15 k GPUs (est.); US (NJ, CA), EU (AMS).
RunPod	On-demand pods with Secure/Community Clouds; partners: NVIDIA, Stability AI.	$100 M+ funding (2024).	≈ 8 k GPUs (est.); Global (US/EU/Asia).
Voltage Park	Bare-metal deep-learning clusters; partners: NVIDIA, Equinix.	$150 M Series B (2025).	≈ 7 k GPUs (est.); US (WA, TX, VA, UT).
Applied Digital (APLD)	US GPUaaS at high-density ND sites; partner: NVIDIA.	Public (NASDAQ); ≈ $500 M funding.	≈ 5 k GPUs (est.); North Dakota campuses.
Together AI	Open-source LLM platform + GPU cloud; partner: Meta.	$200 M funding.	≈ 4 k GPUs (est.); Global (25+ cities).

*GPU capacities are company-reported or analyst-estimated; independent verification varies.

The Strategic Impact of NeoClouds

NeoClouds have fractured the old model of hyperscale dominance by proving that specialization can outperform scale—especially when the bottleneck is GPUs, not general compute. By focusing on purpose-built interconnects, efficient cooling, and rapid provisioning, they have given AI developers an alternative to long hyperscaler queues and unpredictable availability. In many ways, NeoClouds are serving as the shock absorbers of the AI boom, bridging global demand and constrained chip supply.

Their rise is also catalyzing change upstream and downstream. Upstream, GPU vendors now cultivate strategic relationships with these operators, effectively treating them as “Tier 1” customers alongside hyperscalers. Downstream, enterprises and startups gain access to on-demand, sovereign, or sustainable AI compute—often in regions where hyperscaler infrastructure remains limited. As Dell’Oro Group notes, this diffusion of capital and expertise is accelerating the overall AI build-out, reinforcing demand for high-performance servers, power, and networking fabrics.