Microsoft eliminated 4,800 roles this week, roughly 2% of its global workforce, even as it raised AI infrastructure spending to $190 billion for the year — a jump of more than 60%. The cuts landed in the same quarter that the average US domestic stock fund returned 14.8%, its best performance since the post-pandemic rebound of 2020, powered by a rally that has spread from chipmakers to power-generation equipment and fossil-fuel producers. The twin developments expose a tension running through the AI investment cycle: the technology is enriching shareholders while forcing companies to trim headcount to fund it.

The mechanism is becoming clearer as pilot projects give way to large-scale deployment. Uber deployed an AI coding assistant to 5,000 engineers and saw adoption reach 95%, with AI involved in roughly 70% of new code. Yet the annual budget for AI tools was exhausted in four months because providers charge by the token, not per user. Uber’s president acknowledged the company cannot yet link the surge in AI spending to visible business improvements. Across the industry, the shift from licence-based to consumption-based pricing is upending software economics, leaving finance chiefs to question whether the productivity gains justify the outlay.

Big technology firms are responding by sending thousands of engineers directly into client organisations. Microsoft, OpenAI and Amazon have each created deployment units staffed by specialists who redesign workflows and build bespoke AI capabilities on site. The model, known as forward-deployed engineering, is an implicit admission that the bottleneck is not model performance but organisational readiness. In London, FTSE Russell’s global head of investment research noted that “AI is the only topic on the agenda” in markets worldwide. Yet the layoffs at Microsoft, Meta, Amazon and Google show that even as AI creates demand for infrastructure, it is reshaping the labour that remains, compressing roles in sales and gaming while demanding new skills from those who stay.

Viewed from Washington, the IMF’s latest World Economic Outlook tempers the enthusiasm. Global growth is projected to slow to 3% in 2026 and 3.4% in 2027, below the 3.5% average of the previous two years, and the benefits of the AI investment boom are concentrated in economies tied to high-tech supply chains. The White House has published an AI action plan, but the immediate question is whether the capital flooding into data centres and deployment teams will produce returns that justify the cost. The next test arrives with quarterly earnings from the hyperscale cloud providers, where executives must demonstrate that the billions spent on AI infrastructure are beginning to generate revenue beyond the layoffs that helped pay for them.