Why Building AI Data Centres Isn’t Working Anymore

0:00 – 4:39

The $650 Billion Gamble That Isn't Paying Off

Picture a stack of $100 bills tall enough to clear the International Space Station by 300 km. That's $650 billion, the amount four tech companies planned to spend on data centers in 2026 alone. The goal was to push that figure to $9 trillion by 2030. Yet nearly half of the 140 data centers scheduled to open in the US this year have been delayed or cancelled outright. According to Bloomberg, only about a third of these projects representing 12 gigawatts of computing power (enough to run 9 million homes) are actually being built. The rest exist purely as press releases and satellite imagery tells a different story than corporate announcements. Microsoft, for example, hasn't completed a single data center announced since 2023. The Fairwater facility, claimed to be fully operational, sits less than half finished. When ChatGPT exploded into public view in late 2022, the race to build AI infrastructure became frenzied. Last year, an astonishing 92% of US GDP growth came from data center spending alone. Remove AI infrastructure and the rest of the economy grew just 0.1%. That disconnect explains why stock market gains felt so surreal to ordinary people. The shoe company Allbirds recently pivoted to leasing AI data center equipment, and their stock surged 580% despite the move making no strategic sense. More money has been committed to data centers in six years than the combined cost of the Marshall Plan, the Manhattan Project, the Apollo program, and the International Space Station, with $120 billion left over. All of this for a business model that remains unproven. As one investor bluntly put it, the tech giants are spending more cash than they have available, effectively throwing away money at scale for infrastructure that may never deliver returns.