AI data-centre boom faces trillion-dollar test of profitability
Hyperscalers could spend nearly US$1.1 trillion on data centres by 2027, but researchers warn the investment depends on rapid revenue and productivity growth.
AI hyperscalers face a potentially historic test of whether soaring infrastructure spending can generate enough revenue to cover its costs. An analysis reported by MIT Technology Review estimates that spending on data centres could reach nearly US$1.1 trillion by 2027, while total AI capital investment by Alphabet, Microsoft, Amazon, Meta and Oracle could exceed US$5 trillion over four years.
The investment case depends on three conditions: sharply higher revenue for the companies building the facilities, broad productivity gains across the economy and continued demand for expensive frontier AI models. Jessica Wachter, a finance professor at the University of Pennsylvania’s Wharton School, and a collaborator estimate that hyperscalers would need to increase their productivity 2.7-fold to break even by 2030, after accounting for capital costs, depreciation and a 15 per cent return.
Current revenues remain far below projected spending. Gary Gensler, a former US Securities and Exchange Commission chair who is now a professor at MIT’s Sloan School, estimates total AI revenue at about US$150 billion to US$200 billion this year, compared with roughly US$750 billion in hyperscaler data-centre spending. Free cash flow for the group is expected to turn negative as borrowing rises; Alphabet reported a US$5.9 billion deficit in its latest quarter.
The scale of the buildout also leaves companies exposed to changing technology and demand. More efficient or cheaper models could reduce the need for computing capacity, while weaker demand, higher borrowing costs and rapid GPU depreciation could leave new facilities as stranded assets. Mihir Kshirsagar of Princeton’s Center for Information Technology Policy said data centres could become “hulks” if operators cannot keep them competitive.
Economy-wide productivity gains have so far been limited. A survey of about 6,000 senior executives in the US, UK, Germany and Australia found that around 90 per cent reported no productivity increase over the previous three years, although respondents expected gains over the next three. Daron Acemoglu, an MIT economist, said sustained investment would require measurable productivity growth.
Financial risks could extend beyond the companies themselves. Morgan Stanley estimates that more than half of US$2.9 trillion in hyperscaler data-centre spending between 2025 and 2028 could be financed through external capital, potentially increasing exposure among lenders, guarantors, private credit funds and other investors. The scale and timing of any retrenchment remain uncertain, but the projections show how the financial outcome of the AI buildout depends on demand and economic benefits that have yet to be established.
