TCS Research and Innovation
TCS Research
TCS Research and Innovation: Inventing for Impact
Highlights
- AI is changing the economics of the data centre, making compute choice a strategic business decision rather than an infrastructure consideration.
- A right-fit compute approach shifts planning from reactive provisioning to evidence-based workload optimisation.
- Production-scale AI demands infrastructure decisions that balance performance, cost, resilience, and sustainability from the outset.
- Competitive advantage will increasingly depend on how effectively enterprises leverage compute into a driver of growth, efficiency and innovation.
On this page
Why does compute choice now matter for AI?
Cloud expanded access to infrastructure. Data platforms made intelligence more usable. Now AI is pushing both into the core of how businesses operate, compete, and create value. That shift is changing not only applications and business models, but also the logic of the data centre itself. Enterprise AI workloads are evolving quickly and demand massive parallel compute, high-speed data movement, and dense accelerator clusters such as Graphics Processing Unit (GPUs) and Tensor Processing Unit (TPUs). This puts pressure on existing infrastructure by driving up power use, heat output, cooling needs, infrastructure costs, and rack space requirements. Manual, fragmented, and reactive compute choices no longer serve growing AI demands, from proof of concept to production lifecycle.
Existing data centre environments already face constraints in supporting high-density GPU deployments. In addition to application performance, compute choice has to support AI growth sustainably and economically. The scale of the challenge is massive. Recent reports have shown that around 88% of AI proofs of concept still fail to reach wide-scale production. With AI investment skyrocketing, an inefficient compute strategy can lead to over-provisioning and a sub-optimal mix of hardware resources. This can result in performance, network, and memory bottlenecks, increased costs and power consumption, limited scalability, and delayed production go-live.
These pressures create a strategic need for robust capacity modelling, cost optimisation and evidence-based planning. This challenge is especially important for business executives focused on return on investment and risk, as well as for CIOs and CTOs responsible for aligning AI scale with enterprise compute strategy. It is essential to choose the right compute environment up front to reduce waste, improve production readiness and scale AI with confidence.
Why move from traditional to right-fit compute?
Traditional compute planning is often reactive. Teams choose infrastructure based on familiarity, available capacity or a narrow view of cost, then try to fix performance, latency or utilisation issues after deployment. That approach can work for predictable enterprise workloads, but it struggles in AI environments where workload behaviour differs widely across cloud, on-premises, edge, high-performance computing and specialised accelerators. The result is fragmented decision-making, over-provisioned resources, missed service levels, and higher run costs.
A right-fit compute approach asks what each workload needs before production begins, instead of asking what infrastructure is already available. It profiles workload behaviour, compares compute options, and models trade-offs across performance, cost, energy use, and reliability. It then recommends the most suitable execution plan in advance.
For example, in a Network-AI use case with multiple agents, a traditional approach may respond to rising demand by simply adding more GPUs, which can increase costs without fully resolving bottlenecks across CPU, memory or network resources. A right-fit compute approach instead models the workload scalability envelope and selects the optimal GPU-CPU configuration to deliver higher concurrency, faster response and cost minimisation under a strict SLA.
The result is a more disciplined and repeatable way to plan AI infrastructure. Profiling, benchmarking, and what-if simulation help teams test scenarios before committing capital, while multi-objective optimisation supports decisions that balance speed, efficiency, and service-level confidence. This next-generation AI-ready data centre compute strategy helps mitigate investment risk while avoiding both over-provisioning and under-provisioning. For sectors such as banking, manufacturing and retail, this creates a stronger basis for scaling AI without wasting scarce compute capacity.
Figure 1: Traditional compute versus right-fit compute
Why does this approach matter now for industries?
Across industries, AI programmes are moving into production, where performance, cost, sustainability, and governance must be managed together. In banking, financial services and insurance, workloads such as risk modelling, fraud detection, payments and AI-assisted service demand low latency, resilience, auditability and cost control. Manufacturing, retail, life sciences, and pharma face similar pressures, as AI increasingly shapes operational efficiency, customer experience, compliance, and growth readiness.
At the same time, workloads now run across cloud, on-premises, HPC, edge, and accelerator-based environments, making infrastructure choices harder to govern through manual methods alone. Without a more structured approach, organisations risk higher spend, underutilised assets, delayed rollout, missed service levels, and unnecessary energy use. Governance is also becoming more important, as enterprises need workload placement decisions that are explainable, repeatable, and aligned with operational controls and compliance needs.
The business impact can be significant. A more evidence-led compute approach can improve utilisation, reduce avoidable waste, strengthen service-level confidence, support energy-efficient operations and give business, engineering and operations teams a clearer basis for decision-making.
What will shape the future AI data centres?
The future AI data centre will be shaped less by how much compute it contains and more by how intelligently that compute is used. Workloads will move across cloud, on-premises, edge, and specialised environments based on business priorities, performance needs, energy efficiency, and governance requirements. Infrastructure decisions that are manual today will become more continuous, predictive, and closely tied to business outcomes.
This will change how enterprises build for scale. Instead of discovering bottlenecks after deployment, organisations will be able to see likely performance, cost and resilience trade-offs earlier and act before they become operational problems. Data centres will be designed not only for capacity, but for adaptability, sustainability, and confidence at scale.
In the years ahead, competitive advantage will depend on turning compute from a hidden limitation into a deliberate engine of growth.