The Softcat News Team
The economics of GPU infrastructure are unforgiving. H100 GPUs cost 14 times as much as standard on-demand VMs per hour. Reserved pricing widens this to 17 times, as VMs benefit from deeper discounts. Yet without predictive FinOps capabilities, organisations cannot confidently commit to reserved GPU pricing. They remain trapped, paying premium on-demand rates.
Production AI systems achieve less than 50% GPU utilisation even with active load. Most organisations waste 60-70% of their GPU budget on idle resources. A 100-GPU H100 cluster running at 60% utilisation effectively wastes £1 million annually at on-demand rates. Reserved instances could reduce this by 30-50%, but commitment risk paralyses decision-making.
The scale of investment makes this difficult to ignore. IDC projects global AI infrastructure spending will reach $758 billion by 2029. More than half of organisations abandon AI efforts because of cost-related missteps. For UK IT leaders in the world's third-largest AI economy, getting cost management right is not optional.
The Integrated Approach to AI Cost Control
IBM has assembled what Forrester describes as the most complete full-stack cloud cost management solution. The integrated portfolio enables organisations to accurately predict GPU usage patterns, unlocking savings from reserved instances that would otherwise remain inaccessible. The suite combines visibility, optimisation, and automated commitment purchasing.
IBM Cloudability serves as the financial system of record. It provides multi-cloud cost visibility across AWS, Azure, GCP, and Oracle Cloud. The November 2025 Cloudability Governance release enables shift-left cost control, estimating infrastructure costs before deployment. Its recommendation engine uses models backed by over a trillion hours of usage data, providing explicit scores for both savings potential and commitment risk.
IBM Turbonomic delivers the operational optimisation layer with capabilities specifically engineered for GPU workloads. The platform models IT environments as a marketplace, continuously analysing applications and infrastructure to execute policy-driven actions. For AI infrastructure, Turbonomic provides GPU-specific capabilities, including automated instance rightsizing and MIG-aware scaling for NVIDIA workloads. IBM's own Big AI Models team demonstrated the impact: 5.3x increase in idle GPU availability whilst requiring 13 fewer GPUs overall.
IBM Kubecost provides container-level visibility that cloud-provider billing cannot. The Kubecost 3.0 release introduced advanced GPU monitoring powered by NVIDIA's DCGM exporter. This enabled organisations to track GPU utilisation and allocate costs to actual usage. IBM Instana completes the picture with GenAI Observability capabilities, including LLM workflow tracing and token cost attribution by workload and tenant.
Customer outcomes validate the approach. Hyland Software saved $3 million and deferred 60 server purchases. Forrester's Total Economic Impact study documents 247% ROI over three years with 35% cloud savings. BBC Studios deployed Turbonomic across 1,000+ virtual machines, reclaiming 228 GB of memory in a single month whilst eliminating downtime events.
UK Procurement and Regulatory Alignment
UK organisations gain additional advantages through regulatory alignment. IBM Cloudability and Turbonomic both appear on the UK Government's G-Cloud 14 Digital Marketplace and AWS Marketplace. This provides both public and private sector organisations with flexible procurement routes. The platform addresses the FCA FG16/5 guidance requirements for cloud monitoring and governance. Something that is increasingly important as the Critical Third Parties regime takes shape.
Where Softcat Fits
This is where Softcat's role as strategic integrator becomes critical. We are the exclusive supplier of IBM Cloud in the UK under the OCRE 2024 framework. We combine IBM's technology leadership with hands-on technical expertise from our own specialist FinOps teams.
Most organisations approaching FinOps for AI workloads need help to understand which components of the suite apply to their environment. Cloudability, Turbonomic, Kubecost, and Instana each address different layers of the cost problem. Our technical specialists assess your current estate, identify cost visibility gaps, and size the right combination of tools to match your workload profile. That scoping work prevents over-investment in capabilities you do not yet need.
Making the Shift
GPU-heavy environments demand FinOps practices tailored to their unique characteristics. The technology exists to unlock reserved instance savings of 30-50% over on-demand rates. Predictive analysis removes commitment risk. Automated purchasing removes manual burden. Container-level visibility confirms utilisation. Reserved instance commitment becomes a confident decision rather than a gamble.
The strategic question is no longer whether to commit to reserved GPU instances. The question is how quickly you can implement the FinOps capabilities that make commitment viable. Every month, on-demand pricing represents unrealised savings that could fund additional AI innovation.