Stewart Laing, CEO, Asanti Data Centres, asks if the hype around AI echoes the dotcom bubble of the 1990s.
AI is widely seen as the next transformative technology, but could the hype around AI echo the dotcom bubble of the 1990s, where inflated expectations led to speculative investments and wasted resources?
AI has become synonymous with progress, with applications like GenAI and machine learning offering groundbreaking solutions.
However, from what evidence I have seen, many investments in AI-focused data centres are based on speculation rather than demand. This is reminiscent of the dotcom bubble, where investments were made in businesses that could not deliver on their promises.
The root of the problem may be the failure to understand AI’s two distinct stages: training and inference. Training AI models demands enormous computational power, but for a limited time only – around 3-9 months on average.
Once deployed, AI applications require vastly less computational power and infrastructure. Without recognising this, there’s a risk of building huge high tech data centres that are underutilised – a particular concern given existing challenges around power and network connectivity.
AI’s phases are clear: the high-compute training stage and the far less demanding inference phase.
Yet the government’s proposal for a few large-scale AI data centres seems to miss this distinction. The risk? Wasted resources.
Perhaps a distributed model, which leverages regional and edge facilities, could be more effective. These smaller data centres can be used to host the short-term training phase, as well as host AI enabled applications closer to the end user, supporting local economies while avoiding centralisation’s pitfalls. This approach is particularly beneficial for critical sectors such as healthcare and education, ensuring AI infrastructure is built where it is needed most.
Power is a growing concern for all, including data centres, with global electricity consumption currently sitting at around 2%. While that figure is small in comparison to other sectors, it is expected to grow, fuelled by the rise in AI’s. In the UK, energy costs make data centres less competitive then some of their European counterparts. Renewables offer potential, but inefficiencies in accessing green energy due to grid reliance remain a significant hurdle.
Scotland’s excess renewable power contrasts with the South’s scarcity, where demand far outstrips supply. Private wire options could slash costs by up to 60%, but regulatory and logistical barriers prevent their widespread use.
Power alone isn’t enough – connectivity is equally critical. The UK’s slow progress in deploying full-fibre networks restricts data centres to areas with both power and connectivity, making site selection increasingly difficult. This infrastructure bottleneck hinders AI and the wider data centre industry.
To avoid overbuilding and inefficiencies, the government must partner with the UK’s data centre industry to create a distributed, sustainable infrastructure. Smaller, regional data centres can serve end-users more effectively while addressing economic and logistical challenges.
The AI boom presents enormous opportunities, but the risks of overhype and speculative investment are significant. By focusing on power, connectivity, and realistic demand, we can ensure that AI’s transformative potential becomes a reality without repeating the mistakes of the past.
