Exclusive: CommScope on how AI is boosting data centre infrastructure

The world is unprepared for the energy and connectivity demands of artificial intelligence - and the warning signs are already flashing red.

According to CommScope's Matias Peluffo, data centres in Ireland now consume a staggering 18% of the country's electricity, while Australia's share could double by 2030.

Singapore sits at 7%, already among the world's highest.

"The arrival of ChatGPT and AI has put a new sense of urgency into data centre deployments and competition," Peluffo explained during a recent interview.

"It's driven a huge increase in power density, infrastructure density - and sustainability requirements, especially across the Asia-Pacific region, where demand is surging in markets like Australia, Singapore and Japan."

Without urgent adaptation, the industry faces an infrastructure bottleneck that could stall the next wave of AI growth.

As governments move to stabilise electricity demand, operators are being forced to rethink how data centres are built, powered, and connected.

"The Australian authorities have introduced this framework - a grid stabilisation rule - to manage surges in electricity demand, especially when AI comes into play," Peluffo explained.

Similar pressures are unfolding across Asia-Pacific, with Singapore navigating strict power and land limitations even as demand intensifies. It's a direct response to how volatile and resource-hungry AI workloads have become.

The scale of transformation is jarring: just a decade ago, five kilowatts per rack was standard. Now, that figure has jumped to 130 kilowatts for AI, and could reach as high as 600 kilowatts in the near future, based on Nvidia's roadmap. Connectivity needs are rising just as fast, with some racks now requiring up to a thousand fibres.

Peluffo says this surge in density will soon leave legacy facilities behind. "It is very critical to ensure that you have a good migration plan for existing data centres, which may not be able to sustain the type of power density and connectivity density that is required."

The urgency of adapting isn't just about capacity - it's also about efficiency and cost control.

Poor planning in passive fibre infrastructure can quickly spiral into excessive energy consumption and budget blowouts. "If you just think for now, you can deploy a lot of cables that maybe you could have done a better job with - by having high-fibre-count trunks and structured cabling instead of point-to-point connectivity," he explained.

That reality is shaping the next generation of infrastructure, including CommScope's new Propel XFrame optical distribution platform. "The Propel XFrame is designed to be AI-ready from the start. While we have other products designed for service providers, Propel XFrame takes the principles we implemented for AI clusters and puts them into a low-profile, operationally efficient optical distribution frame."

Built with feedback from global colocation giants - including customers across Europe and Asia-Pacific - it delivers a 30% improvement in space utilisation and a 40% reduction in installation time, alongside simpler maintenance and faster turnaround for changes - all aimed at cutting cost and downtime.

Environmental metrics are now integral to every new product cycle.

"CommScope has a very strong track record. Since 2019, we reduced emissions - scope 1, 2 and 3 - by 44%," said Peluffo.

"We minimise material waste, simplify logistics, and support energy efficiency by reducing human error, making it intuitive and modular. It's future-ready and prevents the 'rip and replace' approach when a product reaches end of life."

The platform is designed for phased deployment, allowing operators to upgrade existing brownfield sites gradually while keeping operations online.

"Propel XFrame is standards-compliant and modular - it can be built on a phased basis, side by side with legacy products, allowing gradual migration. It is optimised for both greenfield and brownfield installations."

Visibility and control over these increasingly complex environments is critical as workloads grow in scale.

CommScope's iTRACS platform enables real-time digital twin simulation of the entire data centre, giving operators a unified view of racks, power, connectivity and workflows in 3D. "It enables simulation of changes, capacity management with predictive scenarios, and real-time metrics with automation features - providing a single pane of glass for transparency across IT, facilities and security teams."

An integrated AI chatbot, TRACi, offers what-if analysis and real-time support, while tight integration with the Propel XFrame platform allows for seamless, full-lifecycle infrastructure management.

Security is also under pressure, with rack values soaring alongside AI hardware costs. "The value of a rack in the traditional data centre, in terms of cost, is now into multi-million dollars per rack for AI clusters. Not just asset value, but business value is crucial, so integrated security is key," Peluffo said.

Solutions like CommScope's GigaREACH XL cabling, which supports Power over Ethernet (PoE) up to 250 metres, are helping operators extend secure coverage across wider footprints.

In Asia-Pacific, CommScope is scaling up its presence to meet regional demand. "We're seeing great growth across Asia-Pacific," Peluffo said.

"Australia is a key market for us, but we're also supporting expansion across Singapore, Japan and other emerging hubs in the region. Data centre is the area we want to grow faster than we have so far - we're working on a number of different opportunities to grow the business there." With a dedicated team and improved local supply chain, the company is positioning itself to support the coming boom.

For Peluffo, the message is clear: AI's momentum isn't slowing - it's accelerating.

"There's no sign of [AI] abating. On the contrary, it's progressing multiple... we're moving from AI clusters to AI factories - massive, hundreds of megawatt deployments for data centres."

The stakes are no longer theoretical. The global race to support AI at scale is on - and only the most adaptable infrastructure will survive.