The future of data centres in the age of AI

Alongside the ongoing expansion of cloud computing, acceleration of data generation and now the exponential increase in AI adoption this past year is growing demand for data centres as the backbone of digital infrastructure. Consequently, Australia’s data centre market is set to nearly double to $40 billion over the next four years as businesses continue investing in AI platforms, according to CBRE.

The AI boom is also cited as the culprit for increased energy consumption, starting with data centres. Morgan Stanley estimated that power demand from generative AI will grow at an annual average rate of 70% from now to 2027, primarily due to the growth of data centres. This puts pressure on data centre providers, in their significant role supporting companies to reap the benefits of AI solutions and keep their own energy consumption and carbon footprint in line with their sustainability goals.

So, is balance possible in this new AI-driven era?

Already existing energy management technologies are a critical piece of this puzzle. Data centre providers are showing widespread adoption of renewable energy and integrated energy management systems and solutions that are improving energy efficiency. For those wanting a competitive edge with seamless connectivity in this, the new era of AI, four emerging trends will likely shape their future.

Grid balancing and collaboration

There’s a growing effort to better align forecasts and trends between data centres and utilities, especially for optimising power usage. In the future, as utilities and data centres share information, AI will play a key role in making data centres an integral part of the utility power ecosystem, enabling them to choose the correct power profile and know when to go off-grid to use their back-up sources.

Closer collaboration between utilities and data centres will continue to gain momentum in the next 12 months due to two main factors: ongoing power shortages and the need to stabilise renewable energy sources, like wind and solar, and the addition of BESS (battery energy storage systems) to data centres. Finding the balance between grid management and data centre contractual service level agreements will be an ongoing priority in 2025 and beyond. Such collaboration is likely to be more formalised as the industry moves forward.

Building capacity for internet giants

Data centre operators and co-location companies are now building accelerated computing capacity primarily for AI providers reducing the construction of data centres to host enterprise or cloud applications. The demand for accelerated computing and AI-specific data centres has grown so rapidly that it is leading to intense competition for power resources as companies negotiate with utilities to secure energy before moving forward with permits.

What's particularly notable is the number of commercial real estate companies entering the data centre space that haven’t previously ever been involved. Australia’s most prominent property company, Goodman, has already shifted growth strategies by building more data centres than warehouses — its traditional focus — amid a global thirst for data storage.

The rise of inference

AI is divided into two functional types: training and inference. Training builds models, while inference is the working part of AI, used for decision-making, content generation and, eventually, full automation. The role of AI inference in data centres has progressed significantly, particularly with the continued buzz around the need for edge computing to process real-time data closer to the source to enhance operational efficiency.

However, what’s happening now differs slightly from what was initially expected in the AI space. Large companies that build massive training clusters for AI models — such as large language models (LLMs) — already have significantly accelerated computing capacity. When these clusters are not used for training, they’re repurposed for inferencing, decision-making and content generation.

Initially, many industry experts predicted that once models were trained, smaller, more efficient inference clusters would be built closer to the user, enabling edge AI. But instead of creating new smaller vertical edge AI systems near the data source, AI providers still rely on their large, centralised training clusters for inferencing. This has led to the rise of what’s now being called “data centre inferencing”. Although these training clusters are overpowered for inferencing tasks, they’re being used this way simply because they’re available, even though inference applications typically don’t require high computing power.

Looking ahead, the shift towards leveraging edge computing for inferencing will gain traction as edge devices can operate more efficiently, have lower latency and higher data security, and can be custom-tailored to the application. However, it will likely be a gradual transition as companies adapt their infrastructure to meet the growing demands of real-time AI applications. Until then, data centre inferencing will remain the go-to solution, even if it means using oversized resources for smaller tasks.

Talent shortage

The social and economic dependence on this critical digital infrastructure will fuel industry growth rates of between 25% and 45% to 2030. This has given rise to a shortage of the specialised skills required to complete critical projects. Demand for skills in areas of project management, electrical and mechanical engineering, procurement, and contract management, to name just a few, is lagging well behind supply in the talent pool.

Consequently, we will see an ongoing priority in developing talent pools within organisations, which will see different business models arise in the data centre ecosystem as the skills shortage is tackled.

AI’s long-term impact

AI will continue to drive transformative changes within the data centre industry, especially this coming year. It will also lead to significant data centre reconfigurations over time. The changes that lie ahead for data centres have the potential to be monumental. There is no doubt we are in a cycle of intense innovation to ensure that AI is delivered with the lightest sustainable and most cost-effective footprint.

These shifts include achieving carbon and water neutrality, leveraging nearly 100% green materials, adopting cutting-edge liquid cooling solutions, and leveraging AI in data centre design, maintenance, power management automation with utilities, back-up power control and cooling control.

With every technological leap, data centres of the future are adapting to meet ever-changing demands, ensuring a steadfast increase of their bearing on citizens and society, as well as organisations and economies.

*Joe Craparotta is a 30-year veteran of the IT industry who has been fortunate enough to have experience in many of the technological convergence shifts seen in the industry to date. Joe has been a member of Schneider Electric’s Pacific executive team for the past 10 years, taking on key leadership roles including vice president for the energy business prior to his current role as vice president for the IT business.

Top image credit: iStock.com/onurdongel