These sprawling, power-hungry facilities have become the beating heart of AI innovation, hosting the computing muscle behind tools like OpenAI’s ChatGPT and Google’s Veo 3. They are where trillions of data points are processed and billions of calculations performed to train and operate the technology driving what many see as a new industrial revolution.
Despite growing concern that the AI surge may be another bubble, the signs so far tell a different story. Nvidia, the Silicon Valley chipmaker powering much of the AI infrastructure, last week became the world’s first $5 trillion company. Microsoft and Apple both hit the $4 trillion valuation mark, while OpenAI’s latest restructuring reportedly places its value at $500 billion, setting the stage for a potential $1 trillion stock listing as early as next year.
Elsewhere, Google’s parent company, Alphabet, posted record revenues of $100 billion in a single quarter, largely driven by demand for AI infrastructure. Apple and Amazon, too, reported strong financial results as the AI tide lifted much of the tech world.
Yet the boom isn’t just visible in corporate earnings or Silicon Valley boardrooms. It’s transforming communities, reshaping economies, and redefining the geography of global industry.
A Welsh City Embraces the Future
In Newport, South Wales, a city once defined by coal and steel, a new kind of revolution is taking shape. On the site of an old radiator factory, Microsoft is constructing a massive datacentre complex — part of its global network designed to meet what the tech sector expects to be exponential growth in AI demand.
For Dimitri Batrouni, the Labour leader of Newport City Council, it represents a generational opportunity. Standing on a newly poured concrete floor soon to be filled with thousands of humming servers, he calls the project “a chance to tap into the economy of the future.”
“With cities like mine, what do you do? Do you worry about the past and try to bring steel back with 10,000 jobs – it’s unlikely. Or do you embrace the future?” he asks.
The Trillion-Dollar Infrastructure Race
The scale of global AI spending is immense. The four biggest players — Amazon, Meta, Google, and Microsoft — are expected to invest over $750 billion in AI-related infrastructure over the next two years alone. The Newport facility, for instance, carries a price tag in the hundreds of millions of dollars, while Equinix, a California-based data giant, recently announced a £4 billion centre in Hertfordshire.
But the scale of investment has prompted warnings from some observers. Joe Tsai, chairman of Chinese e-commerce giant Alibaba, has cautioned that parts of the datacentre market show “the beginning of a bubble,” with some projects seeking funding without confirmed customers.
Globally, there are already more than 11,000 datacentres, a fivefold increase in two decades. Analysts at Morgan Stanley estimate total spending could hit $3 trillion by 2028, though only about half will come from the cashflow of the major tech firms. The remainder — roughly $1.5 trillion — may rely on private credit, a fast-growing but loosely regulated part of the global finance system.
Some warn this could carry systemic risks. Gil Luria, head of technology research at U.S. investment firm DA Davidson, says many speculative projects are being financed with debt that may not be properly risk-assessed.
“If it rises to hundreds of billions of dollars,” he cautions, “it could end up representing structural risk to the overall global economy.”
Lofty Returns and Fragile Foundations
Investment bank Morgan Stanley forecasts that revenues from generative AI — from chatbots and image generators to automated agents — will soar from $45 billion in 2024 to $1 trillion by 2028. The challenge, however, is whether that level of demand will materialise.
A Massachusetts Institute of Technology (MIT) study earlier this year found that 95% of organisations have so far gained no measurable return from their investments in generative AI pilots — a sobering contrast to the trillion-dollar optimism fueling the datacentre build-out.
Industry analysts at the Uptime Institute, which monitors datacentre performance, believe many announced projects may never materialise or will operate well below capacity.
“A lot of this is speculative,” says Andy Lawrence, the institute’s executive director. “Many of the datacentres announced with fanfare will either never be built or will be populated only gradually over a decade.”
Powering the Cloud — and the Future
Even with these doubts, the global expansion continues at breakneck speed. Property group JLL expects work to begin this year on 10 gigawatts of new datacentre capacity — roughly a third of the UK’s total power demand. Another 7GW is expected to come online before year’s end.
Goldman Sachs predicts global capacity will double by 2030, requiring an additional $720 billion in grid investment to support the energy needs of these digital fortresses.
Meanwhile, industry giants are pursuing even more ambitious projects. OpenAI, Oracle, and SoftBank have launched Stargate, a $500 billion plan to build a continent-wide AI datacentre network in the U.S., with a UK version already planned for North Tyneside. Elon Musk’s xAI has built the “Colossus” project in Tennessee, while Microsoft’s new site in Fairview, Wisconsin is being billed as the world’s most powerful AI datacentre.
A New Industrial Landscape
Back in Newport, the transformation is personal. Mike O’Connell, a construction safety specialist who has worked on oil rigs and datacentres worldwide, has returned to his hometown as a consultant on the project. His teenage grandson is now an apprentice electrician at the site.
“I’m looking to stay in the local community,” O’Connell says, reflecting on how datacentres may offer a new generation of skilled work.
For now, optimism reigns. Trillions of dollars, vast infrastructure projects, and a vision of limitless computational power have combined to fuel what may become the defining technological story of the 21st century.
Whether the AI datacentre boom proves to be a durable engine of global growth — or another case of digital overreach — may depend not just on how much is built, but how wisely it is used.