The surge in demand for artificial intelligence (AI) and generative AI technologies is expected to see data centre operators globally increase their infrastructure investment to almost half a trillion dollars by 2027.
This was the word from VladGalabov, director of cloud and data centre researchspeaking last week at the Vertiv Driving Innovation 2023:The Prefabricated Modular RevolutionEMEA press conference, in Zagreb, Croatia.
Providing an overview of the global data centre landscape, Galabov explained that the increase in AI adoption is forecast to fuel data centre investments from 2024 to 2027.
While other industries within the ICT sector are expected to see a 20% decline in overall tech investments, the data centre industry is predicted to nearly double its investments, boosted by a wave of AI, physical and IT infrastructure, he noted.
Capex investment in data centre infrastructure solutions has reached an average of $25 billion annually. Next year, this is expected to escalate to $30 billion, mainly due to the unprecedented data centre power capacity required for AI infrastructure, he stated.
With AI becoming infused into almost every application, data centres are being reshaped to accommodate this soaring demand, with the increasing adoption of hardware now specifically designed to handle the high demands of this type of processing.
“The data centre industry has seen a huge gold rush tolaunch large language models to take advantage of the generative AIopportunity. And this has everything to do with the fear of missing out.
“So, we have seen a huge change in spending patterns and most of the budget is spent on building out computing power, as well as physical power. What I believe is happening is that we have entered a five-year period of significant power expansion in the data centre industry,” he said.
This year, many enterprises across the globe have been running AI and generative AI proof-of-concepts, and mostly exploring with use-cases, as well as laying a foundation for their AI strategies.
Next year is expected to become a really “big year for AI” as many enterprises will officially deploy their AI applications across various business units – and this will lead to data centres requiring a lot of computing power, he asserted.
“Data centres are becoming AI factories, and in order to become an AI factory, you need to change your infrastructure − and it has to change quickly. We have seen a huge change in spending patterns and most of the budget is spent on building out computing power, as well as physical power.
“Because the new servers are so much more powerful − they are 10 times more powerful − if we project what this means for the power that’s used by the IT loads in the data centre, we've actually doubled our power capacity for AI. Around 22% of data centre power capacity is now used for AI. And it's important to know that this figure is really only for early AI investments.”
According to Galabov, hyperscale service providers have been among the first to get their hands on the first batch of servers that are specifically focused on AI applications. As more players use these, this will result in total data centre power capacity exceeding 40%.
“We project that this figure will become 41% once everyone is able to get their hands on servers for AI. They are investing in server clusters for training larger language models, and servers that are going to train smaller language models. Data centre operators have also been spending on cooling technologies that are going to enable these servers to better operate.”