AI spending nears $760 billion as big techs test the economics of scale

211

Big Tech’s AI infrastructure race is moving into a more expensive phase. Alphabet, Amazon, Meta, Microsoft and Oracle are expected to spend about $760 billion on capital expenditure in 2026, according to Visible Alpha estimates reported by The Wall Street Journal, up from roughly $412 billion for the five companies in 2025.

The spending reflects an urgent effort to secure data centres, GPUs, networking equipment and power as demand for AI computing continues to outrun available capacity. Alphabet, for instance, had earlier guided to $175 billion-$185 billion of 2026 capex, with most infrastructure investment directed toward AI and cloud capacity.

The commercial case is becoming clearer, but so is the financial strain. Microsoft said its AI business had surpassed a $37 billion annual revenue run rate, up 123% year-on-year. “Our AI business surpassed an annual revenue run rate of $37 billion,” CEO Satya Nadella said while reporting fiscal third-quarter results.

Investors are increasingly watching cash generation rather than headline AI spending. Reuters analysis of LSEG estimates found hyperscaler capex could grow about $534 billion between 2025 and 2027, versus roughly $340 billion of additional operating cash flow. At that trajectory, capital expenditure could exceed combined free cash flow by 2027.

Accounting may also delay the full impact on reported earnings because data-centre and computing assets are depreciated over several years. David Zion of Zion Research Group warned that analyst depreciation forecasts are “all over the place.”

The AI race is therefore becoming a return-on-capital test. Capacity remains scarce and demand remains strong, but the winners will increasingly be judged on how efficiently billions of dollars of infrastructure translate into durable AI revenue and cash flow.