The way out of the AI Capex conundrum may be commoditization
AI demand is real, but the capital required to meet it is growing faster than profits. What are the possible ways out?
Hyperscalers are investing large amounts of capital in AI infrastructure. We understand that AI demand is real and growing. The question is whether it is becoming profitable fast enough to justify the amount of capital required to serve it.
So far, our findings suggest it is not. Return on capital across the hyperscaler group has deteriorated since the beginning of the latest AI investment cycle, with the accounting effect naturally appearing with some delay. Profits continue to grow, but the amount of capital required to generate those profits is growing faster.
This is increasingly visible to the market, which appears to follow a relatively simple equation: more Capex is positive for infrastructure providers supplying chips, servers, data centers and power equipment, but increasingly negative for the hyperscalers funding the investment. Conversely, a slowdown in AI Capex would improve cash generation at the hyperscalers but raise questions about future growth across the infrastructure ecosystem.
There appear to be three possible ways out.
