Is AI the Dotcom Bubble of the 1990s for Data Centres?

AI is transforming industries, from healthcare to finance, with its promise of innovation and efficiency. Yet, amid the excitement, we must ask whether AI is heading down the same path as the dotcom bubble of the 1990s - a time when speculative investments outpaced real utility, leading to widespread inefficiencies. Stewart Laing, CEO at Asanti Data Centres, explores how the data centre industry can avoid repeating these mistakes and thrive in this new era.

The AI Hype: Is the Data Centre industry Preparing for the Right Demand? AI’s growth is undeniable, with advancements like generative AI and real-time language translation pushing the boundaries of innovation. However, much of the current investment in AI-focused data centres appears speculative, based more on hype than actual demand. This is reminiscent of the dotcom era when record amounts of venture capital poured into technology start-ups that never reached profitability.

The key this time around is understanding AI’s distinct phases. While training AI models is compute-intensive, it’s also short term. Once deployed, AI requires significantly less infrastructure. Ignoring this distinction risks overbuilding and wasting resources - particularly in power and connectivity, which are already pressing challenges for the data centre sector.

Compounding the issue is the Government’s AI Action Plan. While it aims to promote AI innovation, its focus on building one or two massive data centres fails to address critical nuances.

What Does the Government Mean by AI? A clearer understanding of AI’s infrastructure needs is crucial. AI operates in two phases: the training phase, which places short-term demand on high-compute resources, and the inference phase, with far lower power requirements. The government’s strategy for large-scale, centralised AI data centres seems to overlook this distinction, potentially creating inefficiencies..

A better approach would be to leverage existing data centre ecosystems, enhancing power and connectivity infrastructure, and ensuring that AI growth is driven by practical needs rather than buzzwords and unfounded expectations.

Power Challenges Data centres consume around 2% of global electricity, small in comparison to other sectors but it is a figure that is expected to grow with AI’s rise. In the UK, high energy costs make it hard for data centres to compete in Europe and internationally. Despite progress in renewable energy, inefficiencies in tapping into offshore wind and solar power persist due to regulatory barriers and an outdated national grid.

More direct solutions, such as private wire systems, could significantly reduce costs, but the absence of a national energy strategy leaves data centres struggling with rising expenses. Scotland’s surplus green energy contrasts starkly with power scarcity in the South, where demand outpaces supply.

Fibre Connectivity Beyond power, robust fibre connectivity is essential. The UK’s slow rollout of full-fibre to fibre networks limits new data centre deployments to regions with both power and connectivity - a rare combination. This lack of coordination hinders AI’s growth and the broader data-driven economy.

A Smarter Path Forward The government should collaborate with the UK’s existing data centre industry to adopt a distributed model. Smaller, regionally located facilities would support services more efficiently, address infrastructure gaps, and spread economic benefits.

The AI revolution is real, but the risks of overhype and speculative-investment are significant. By focusing on power, connectivity, and realistic demand, we can avoid repeating the mistakes of the past and ensure that the appropriate infrastructures are put in place to allow broader support.