Generative artificial intelligence is no longer a futuristic curiosity from the headlines. Today, it is a powerful business tool, actively deployed in companies to optimise processes, create innovative products and gain competitive advantage.
However, this technological revolution has a huge appetite – an insatiable hunger for computing power. It turns out that the biggest inhibitor to the development of AI in-house is not a lack of ideas, but a lack of physical space and energy to implement them.
The problem has become so acute that it is forcing IT departments to make a fundamental change in strategy: planning infrastructure in terms not of months, but of whole years.
The scale of the phenomenon – business is pushing, IT is looking for space
The demand for resources for AI is no longer a grassroots initiative of technology enthusiasts. It’s a strategic priority coming straight from the boardrooms. According to a recent report by Flexential, as many as 90% of IT leaders are actively implementing solutions based on generative AI.
More tellingly, however, the driving force behind these implementations in 81% of organisations is top management. This is a massive increase from the 53% recorded just a year earlier.
When management says ‘act’, IT bosses face a very physical problem: where will the company’s new artificial intelligence ‘reside’? Every model, every application and every query to an AI system requires powerful computing power, cooling and secure space.
This business pressure translates into an immediate and exponentially increasing demand for data centre resources, which, it turns out, are starting to run out.
Market consequences – tight, expensive and no quick fixes
The data centre market, accustomed to stable, predictable growth, was not ready for such a sudden wave of demand. The result is a real supply crisis, which is manifesting itself on three main levels.
Firstly, increasing competition. Getting the right server space nowadays is akin to fighting for luxury city centre real estate. Demand far outstrips supply and companies are desperate to find available resources.
The same Flexential study found that already more than half of businesses are bailing out with colocation services (renting space in external server rooms) to fill their gaps.
Secondly, soaring costs. Where supply fails to keep up with demand, prices go up. According to CBRE’s market analysis, rents for space in key technology hubs have skyrocketed by almost 15% year-on-year. AI is therefore not only expensive to develop, but is becoming more expensive to maintain.
Thirdly, and perhaps most importantly, the lack of quick solutions. It is impossible to build a data centre in a quarter. These are complex engineering projects, requiring huge amounts of money, permitting and time – often several years from the first shovel to getting the servers up and running. When current capacity runs out, there is no contingency plan B that can be implemented overnight.
New strategy – planning for a five-year perspective
Aware of these limitations, IT leaders are abandoning the reactive model in favour of proactive, long-term strategic planning. Waiting until power requirements become critical is no longer an option. The data from the Flexential report perfectly illustrates this shift in thinking:
- 79% of companies are still planning their needs a year in advance – the standard to date.
- However, 62% are already looking further ahead, forecasting demand over a horizon of one to three years.
- Most tellingly, 17% of the most prudent organisations are reserving power and space with a view of three to even five years.
What was once the domain of financial planning is now becoming the norm in IT strategy. We are also seeing geographic diversification, with new hubs growing in importance alongside traditional server basins, attracting investors with land and energy availability.
A look to the future – the next wave is yet to come
If anyone thinks that the current boom is a temporary phenomenon, they are mistaken. Experts agree that this is just the beginning. The current demand is mainly driven by the adaptation of large language models (LLMs) from technology giants. However, the inevitable next evolutionary step will be small language models (SLMs).
These will be specialised, much more powerful models, trained by companies on their own confidential data to perform specific tasks. This will allow the creation of secure, in-house AI tools, tailored to the needs of the organisation.
The growth of SLM will generate a second, massive wave of demand for dedicated infrastructure, which will further intensify the race for data centre space.
The conclusion is simple and inexorable: the race for dominance in the age of artificial intelligence is largely a race for access to computing power. Long-term planning and strategic reserving of resources in data centres is no longer good practice – it has become an absolute requirement for survival and growth.
Companies that ignore the growing ‘power hunger’ for AI today may wake up in a few years’ time with brilliant models and innovative ideas for which they will run out of power and space. Their revolution will be stuck on the drawing board, while competitors who thought five years ahead will already be far ahead.
