Sanjay Govender, Qrent’s Head of GBS/BPO Solutions.
The BPO industry has embraced AI as a technology capable of improving operational efficiency, enhancing customer experiences and supporting business growth. But inside South African call centres, the opposite is quietly happening.
As AI tools become deeply embedded into customer engagement environments, many operators are discovering that the real cost of AI is not the software licence – it's the infrastructure required to run it.
From voice neutralisation software and real-time call assistance to AI-driven first line support and live agent coaching, the processing demands inside modern BPO environments have increased dramatically over the past 18 months.
What many providers underestimated was the backend impact. AI does not run for free. It requires compute power, memory, networking throughput, low latency environments and increasingly expensive infrastructure to support it at scale.
The result is that many BPOs are now facing a difficult and expensive decision. One approach is to run AI workloads directly on endpoint devices. This means moving away from standard workstation deployments towards higher specification machines capable of handling AI assisted applications locally.
In practical terms, this is driving a noticeable shift away from traditional Intel i5 deployments towards growing demand for i7 powered devices on the call centre floor. AI-enhanced workloads are forcing hardware upgrades far earlier than many refresh cycles originally planned for.
The second option is to keep endpoint devices relatively standard while shifting the AI processing burden into the backend environment. In this model, AI applications and workloads are hosted centrally on servers, reducing the processing demand on the user device itself. While this avoids large-scale desktop upgrades, it introduces a different problem – significantly increased server infrastructure requirements.
This is where many BPOs are starting to feel the financial pressure. Backend server environments capable of supporting AI-driven workloads require substantially higher compute density, increased storage performance, more advanced networking and far greater scalability than traditional call centre infrastructure.
The cost of expanding on-premises server stacks to accommodate these workloads is rising rapidly, particularly as demand for AI-capable hardware continues to grow globally.
According to Gartner, worldwide spending on AI-optimised servers is accelerating sharply as organisations race to support enterprise AI workloads, contributing to overall global IT spending reaching $6.15 trillion in 2026.
The third route many organisations are exploring is moving AI infrastructure off premises entirely through hyperscale providers such as Amazon Web Services or colocation environments like Teraco. In this model, the infrastructure is rented rather than owned, with AI workloads hosted externally and delivered to the BPO environment through cloud or hosted platforms.
While this removes the burden of large upfront infrastructure investment, it introduces ongoing rental and operational expenditure costs that must be managed carefully over time. For some BPOs, this creates far greater flexibility. For others, especially those operating at scale with strict latency and compliance requirements, the long-term cost equation becomes more complex.
What is becoming increasingly clear is that AI is fundamentally changing the economics of the BPO industry. For years, cost optimisation in call centres focused largely on labour efficiency. Today, infrastructure efficiency is becoming equally important.
The conversation is shifting from simply how many agents a BPO can support to how much compute power it takes to support them effectively in an AI-enabled environment. This is why the traditional procurement model is coming under pressure. Many operators still attempt to purchase server infrastructure outright through large capital expenditure projects.
But in a market where AI workloads are evolving rapidly, hardware demands are changing constantly and infrastructure pricing remains volatile, locking large amounts of capital into fixed infrastructure is becoming increasingly risky.
A growing number of BPOs are instead exploring leasing and rental models for backend AI infrastructure. Rather than purchasing expensive server environments upfront, providers can deploy infrastructure through operational expenditure models that spread costs over time while maintaining flexibility as AI requirements evolve.
This approach also reduces the risk of over-investing in hardware that may become insufficient or obsolete far sooner than traditional infrastructure cycles allowed for. In an AI-driven environment, scalability and adaptability are becoming more valuable than ownership itself.
The uncomfortable reality is that AI is not automatically reducing operational costs inside BPOs. In many cases, it is increasing them. The difference is that the costs are shifting away from people and moving into infrastructure.
That changes everything, because the next competitive battle in the BPO industry may not be about who has the cheapest labour model. It may be about who can afford to power AI at scale.
