Krish Prasad, Senior Vice President and General Manager, VCF Division at Broadcom, explains why enterprises can no longer rely on simply adding more hardware to meet the demands of AI.
For many CIOs, the economics of AI are starting to shift. Rising costs and limited access to critical components are making it harder to scale infrastructure in the way organisations have become used to. What has been described as a ‘memory super-cycle’ is a market dynamic that is putting pressure on supply chains, as demand for AI-ready systems outpaces availability.
Hyperscalers have largely secured their position, locking in capacity early. Meanwhile, manufacturers are prioritising high-bandwidth memory for GPUs, reducing supply for everyone else. This is driving up costs, extending lead times and leaving traditional enterprises at a disadvantage.
The old model of scaling by adding more hardware is breaking down. Organisations are reaching a point where simply investing more is no longer enough. The answer lies in using infrastructure more intelligently, with software playing a far greater role in how AI is delivered.
Rethinking the hardware-first approach
Enterprise IT has long relied on adding capacity to address performance challenges, but the current supply crunch is exposing the limitations of this approach. As demand for AI-ready infrastructure accelerates, memory costs have surged – often accounting for more than 50% of total system spend – while supply remains limited. Consequently, simply adding capacity is becoming increasingly expensive and, in many cases, unsustainable.
Enterprises are being forced to rethink their infrastructure strategy and try to do more with less. This is where optimisation comes in: using software to manage resources more intelligently and efficiently. What began as a response to cost pressure now represents a broader transformation in how infrastructure is designed and operated.
In practice, this means rebalancing resource utilisation. In many environments, CPU capacity – the processing power provided by central processing units – remains underused, while workloads are constrained by memory availability. Techniques such as high-speed NVMe memory tiering, which moves less active data from expensive DRAM to cost-effective NVMe storage, can help organisations reduce memory costs and increase VM density.
At the same time, extending the life and value of existing infrastructure has become a priority. Approaches such as intelligent oversubscription, workload balancing and memory optimisation can support higher workload density without compromising performance. Storage efficiency also plays a key role, with data reduction techniques increasing effective capacity while making better use of CPU and memory that may be constrained by rigid configurations.
Together, these software-led strategies are reinforcing the role of private cloud platforms as a control layer for modern infrastructure, giving enterprises greater visibility over how resources are allocated and optimised in response to industry constraints.
The power of smarter software
Today’s supply chain challenges signal a deeper transformation in enterprise IT. Relying on additional hardware to solve performance issues is no longer a viable long-term strategy, particularly as AI workloads place increasing strain on infrastructure. In response, organisations are turning towards software-led approaches that emphasise efficiency, flexibility and smarter use of existing resources.
Focusing on software-driven optimisation within private cloud environments allows organisations to make better use of existing infrastructure, respond more quickly to changing demands and scale without continuous hardware investment. This approach not only reduces dependency on constrained supply chains, but also enables more predictable cost management. Those that adapt early will be better equipped to navigate ongoing constraints, control costs and maintain momentum in their infrastructure plans.
