Building AI data centers is getting harder and more expensive, which is a problem for developers but a sign of how much money is still flowing into AI.

• AI data centers are turning from simple buildings full of IT gear into factories that produce computing power, and that changes how the business works.
• Demand still looks strong: cloud revenue at Amazon, Microsoft, Alphabet and Oracle is set to grow 20–60% next year, with order backlogs doubling.
• Governments building their own "sovereign AI" systems are becoming a new source of demand.
• Power is the big bottleneck — operators now have to find, finance and manage their own electricity instead of just plugging into the grid.
• The money is the third problem: these projects cost a fortune, so the industry is moving toward equipment loans, compute-backed lending and even turning computing power into tradable securities.

Outlook: Expect the race to focus on sites that already have locked-in power, and new financing tools to spread after 2027.