Goldman Sachs warns of mounting scrutiny over AI returns

Spending on AI infrastructure continues to expand, but companies face growing pressure to show that investments in data centres, computing capacity and advanced chips can generate sustainable profits. Goldman Sachs sees the next 18 months as a crucial period for distinguishing viable business models from commitments that may prove unsustainable. Social Capital CEO Chamath Palihapitiya said measurable earnings gains attributable to AI remain largely unproven.

Specialised cloud providers, known as neoclouds, face particular uncertainty because their financial obligations depend on keeping expensive infrastructure sufficiently utilised. Investors are increasingly favouring stronger projects, developers, leases and tenants. Goldman Sachs leveraged finance executive Miriam Wheeler said developers with limited equity, unproven track records or weaker leases could struggle to secure capital.

Around 30 AI-related deals reached the high-yield debt market over the past 12 months, but activity contracted sharply after September, with no new deals over the last month. Disappointing demand, constrained financing and insufficient power could all slow the investment cycle.

Large technology companies benefit from stronger balance sheets and established businesses. Goldman Sachs estimates that Alphabet, Amazon, Microsoft, Meta, Oracle and SpaceX could achieve returns on invested capital of 15 per cent or more. The outlook points to potential correction and consolidation rather than an immediate collapse.