A new Barclays survey of 410 fixed-income investors makes one thing obvious: AI isn’t coming to finance. It’s already here.
About 52% of long-only managers use AI primarily for research. Hedge funds lean on it heavily for market data processing, with 72% reporting daily use. That’s nearly double the rate of long-only managers and almost twice that of asset owners. Trading and execution? Still mostly off-limits. Data security remains the top concern.
Few expect job losses though. Only 7% anticipate meaningful staff cuts. Most predict higher output with stable headcount. So much for the robots taking all the jobs.
But Marc Andreessen sees a different bottleneck on the horizon. His assessment is straightforward: the amount of AI a country can run depends on its air conditioning capacity. AI servers draw enormous power, and cooling them draws even more.
The numbers back him up. The IEA expects data center demand to more than double by 2030, hitting around 945 terawatt hours — roughly Japan’s total current consumption. In the US alone, data centers may soon outpace aluminum, steel, and cement production combined.
Meanwhile, Big Tech isn’t slowing down on spending. Microsoft, Amazon, Alphabet, and Meta have laid out a combined $725 billion in 2026 capital guidance. That’s up 77% from this year.
The takeaway? Demand is real, but the physical infrastructure might be the thing that decides who wins and who gets left behind.
