As the first chart shows, investment levels are higher than a few years ago. This matters because these investments must ultimately be depreciated. If revenues from AI-based services take time to materialise, rising depreciation charges could put pressure on operating margins. This risk is particularly relevant for AI infrastructure. Server hardware typically has a shorter replacement cycle than traditional data-centre assets, resulting in higher depreciation rates.
Case in point: Alphabet. Based on consensus estimates, its FY26 capex-to-sales ratio is expected to reach around 44%, while depreciation is projected to be 14% of sales. If depreciation charges rise over time to reflect this higher level of investment, profit margins could come under pressure, creating a headwind for earnings per share. For Microsoft, the equivalent figures are 35% and 14%, while Amazon is expected to post ratios of 24% and 13%, respectively.
During periods of heavy investment, capex typically exceeds depreciation. The AI sector is currently in such a phase. The key questions for investors are whether these investments generate returns above the cost of capital and whether the anticipated revenue growth ultimately materialises.