Exclusive: Virtuozzo sees GPU clouds reshape AI infrastructure

AI infrastructure is entering a new phase as cloud providers increasingly look beyond traditional infrastructure services and towards GPU-as-a-service offerings.

According to Virtuozzo CEO Kurt Daniel, there is a shift as organisations face rising demand for generative AI, inference workloads, and emerging agentic AI applications, while also managing rising infrastructure costs and continued pressure on GPU availability.

Industry examples highlighted by Virtuozzo suggest GPU cloud services are becoming a larger revenue source for cloud providers as AI workloads move into production.

Daniel said GPU-as-a-service remains a relatively new market segment, but one that is attracting significant interest from cloud providers seeking higher-margin opportunities.

"I'd say it's newer in nature. I think a lot of the companies that we're talking to, there's a lot of interest in adopting this now," said Daniel.

He pointed to CloudPe, an Asia-Pacific cloud provider that historically focused on infrastructure-as-a-service before expanding into GPU-as-a-service.

A recent CloudPe case study published by Virtuozzo found that GPU services have grown to account for up to 25 per cent of the provider's revenue as demand for AI applications and large language models (LLMs).

CloudPe's GPU cloud platform supports AI and machine learning infrastructure, serving AI-native startups, SaaS providers and development teams running LLMs, inference workloads and open-source AI applications. Customers can deploy open-source models and use frameworks such as VLLM to expose AI services via APIs, according to the report.

The second use case at CloudPe is GPU-powered virtual desktop infrastructure, in which cloud-hosted desktops provide GPU acceleration for graphics-intensive workloads such as animation, engineering, computer-aided design, and digital content creation.

While AI discussions often centre on models and applications, Daniel believes infrastructure efficiency remains underappreciated.

Cloud providers face pressure from several directions simultaneously. From GPU acquisition costs to ongoing supply constraints and rising virtualisation costs following changes in the VMware market.

Virtuozzo positions its software stack as a way to improve utilisation of hardware resources through an optimised Linux operating system, orchestration tools, automation and management capabilities.

"Organisations, while they want to take advantage of AI and all the opportunities that AI brings, they need better efficiency in any way, and particularly from software at the infrastructure level," said Daniel.

Security, compliance and sovereignty requirements are also expected to influence how GPU cloud infrastructure develops.

"The eventual users and end customers, whether they're governments, healthcare companies or tech companies, they're going to care about the sovereign nature," said Daniel. "It's making sure the infrastructure and the hardware and the data centre, wherever it is, can help them with their sovereign goals, wherever it is around the world."

Looking ahead, Daniel expects AI infrastructure platforms to become more automated, more secure and more deeply integrated into customer portfolios.

"In addition to the efficiency, management is important in terms of system-level capabilities. Automation is important in terms of billing, payments and related areas, and protection is paramount in terms of security, sovereign requirements and compliance. So AI brings great opportunities, but organisations must move forward with the right infrastructure to fully take advantage in the right way," said Daniel.