This panel featuring Peter Hind (Principal Research Analyst at ADAPT),  Peter Alexander (CTO, Defence Digital Group at the Department of Defence), Matt Gurr (Senior Director of Design Management APAC at NTT Global Data Centres, formerly AECOM, AWS, Aurecon) and Adam Gardner (Head of Products at NEXTDC), discuss the growing difficulty of planning and scaling data centres amid rising demand, especially from AI. Matt points out the shift from long-term infrastructure planning to much shorter timeframes, due to the four-year lifecycle of data centre development and rapidly changing requirements like water-cooled racks and AI-heavy loads. Flexibility and modularity now take precedence over fixed infrastructure. As AI and high-density workloads drive infrastructure demand, ADAPT data suggests that 25% of organisations are now repatriating workloads from public to private or hybrid cloud, seeking better cost control, compliance, and operational efficiency. This shift increases complexity, requiring modular infrastructure, architectural interoperability, and tighter alignment with business value. Adam adds that a “Lego-like” approach is being adopted, with components such as generators ordered well in advance. He notes that rack densities have risen from 30kW to as much as 1,000kW, placing significant pressure on power and cooling systems.

Peter outlines the defence sector’s specific challenges, such as forecasting capacity needs for unpredictable future operations. Defence has moved away from building its own facilities and instead relies on commercial providers with better capabilities. He observes that US hyperscalers are securing rural land and investing in their own renewable or nuclear energy sources, creating more sustainable operations than Australia’s fragmented model. Sovereignty and security remain critical for Defence, with strict provider vetting in place. Peter stresses the need for broader public awareness, citing Ukraine’s swift relocation of workloads to hyperscalers during wartime as an example of resilient, sovereign-by-design infrastructure.

The discussion broadens to include data sovereignty, AI acceleration, and changing infrastructure roles. Peter explains that even classified data can now be safely hosted in hyperscale public clouds, protected from foreign threats. He questions why sovereignty is so tightly linked to data when imported military hardware is rarely scrutinised in the same way. He and Adam describe how rapidly advancing AI is solving previously intractable problems, driving up infrastructure demands but also enabling more compact, powerful compute. Matt and Adam highlight the importance of early collaboration across the tech supply chain, particularly with integrated solutions from companies like Nvidia. All three panellists agree that traditional on-premise models are fading, but infrastructure roles remain crucial, albeit higher up the stack and more focused on application delivery and mission-specific outcomes.

Key Takeaways:

• Infrastructure must be flexible and modular: Traditional long-term data centre planning is no longer viable due to rapid shifts in demand from AI and high-density compute loads. Providers are adopting “Lego-like” designs and ordering long-lead-time components early to stay ahead.
• Sovereignty and sustainability are evolving: Defence and hyperscalers alike are prioritising sovereign-by-design approaches, with US providers building self-powered, rural data hubs. Australia’s fragmented model and public misunderstanding of data centres as utilities present ongoing challenges.
• AI is transforming infrastructure roles: As AI accelerates, infrastructure management is moving beyond physical facilities toward higher-value layers like applications and mission-specific capabilities, requiring early supply chain collaboration to meet evolving demands. Yet only 36% of organisations can confidently link cloud spend to business value, highlighting the need for stronger stakeholder engagement, FinOps adoption, and clearer metrics such as cost efficiency and risk reduction.