How Are AI Systems Affecting Cloud Hosting Costs?

Recent research conducted by UK data centre operator Pulsant indicates that organizations operating artificial intelligence (AI) systems are experiencing increased pressure regarding the hosting of these systems. Sectors such as banking, healthcare, and manufacturing are observing elevated costs and greater infrastructure demands as the utilization of AI expands.

According to Pulsant's analysis, global data centre demand is projected to reach between 171 and 219 gigawatts by 2030, up from approximately 60 gigawatts today. Facilities designed to support AI could represent approximately 70% of this demand.

AI systems typically consume more electricity compared to conventional business software. The training of AI models necessitates dedicated graphics processors that operate continuously for extended periods. Furthermore, the constant movement of large data sets across networks increases energy consumption and cooling requirements.

After the training phase, live AI systems require consistent processing speeds, as any delays can impact fraud detection, medical analyses, or production forecasts. These infrastructure challenges have elevated infrastructure decision-making to a board-level priority.

Public Cloud Considerations for AI Deployment

Public cloud platforms are frequently chosen at the inception of AI projects due to their ability to provide quick access to powerful processors and facilitate model testing without the need for purchasing equipment. This is particularly beneficial for short-term projects and trials.

However, issues arise when AI systems require continuous operation. Extended rental of advanced processors significantly increases costs, and data egress from cloud platforms results in additional charges. Pulsant notes that financial teams often reassess strategies as these costs become evident.

Additionally, capacity challenges have emerged because the demand for high-end processors sometimes surpasses the available supply on shared cloud platforms, causing resource allocation delays during peak periods.

Organizations such as banks, insurance companies, and healthcare providers, which manage sensitive data, find that while cloud platforms adhere to stringent digital security standards, data location and audit controls remain outside their direct oversight, potentially complicating compliance efforts.

Advantages of Colocation for AI Systems

Colocation offers organizations the option to deploy their own hardware within specialized data centers designed to handle significant power consumption and advanced cooling requirements. These facilities are equipped to support dense processor clusters operating continuously.

A survey conducted by Pulsant identified high-density power and cooling as the top priorities for hosting AI workloads, cited by 54% of IT leaders. Following closely were direct connections to public cloud platforms at 51%, and support for high-performance computing infrastructure at 49%.

This approach enables organizations to conduct demanding training tasks on stable equipment while maintaining cloud access for additional capacity when necessary. The connectivity between private systems and public platforms is critical in this model.

Cost management differs between the two hosting options. Cloud hosting avoids initial capital expenditure, but long-running AI workloads result in variable monthly costs. In contrast, colocation requires upfront investment, but operating costs remain stable, facilitating easier financial planning.

Stephen Spittal, Technology Director at Pulsant, stated, "AI imposes significantly greater demands on infrastructure than traditional IT systems. Once beyond the initial pilot stage, the continuous need for power, cooling, and connectivity becomes evident."

Spittal further noted, "Colocation provides the necessary capacity to manage these workloads without disruptions, in facilities designed for efficiency. It also assures performance reliability as projects scale."

He concluded by highlighting that "AI is moving towards the Edge, necessitating inference to occur closer to end-users. Our latest research indicates that 87% of UK businesses plan to migrate partially or fully from the public cloud within the next two years."

Based on reporting by techround.co.uk.