AI spending growth seen slowing sharply by 2028, marking shift towards software: Morgan Stanley

The artificial intelligence (AI) investment cycle could move from a hardware-led phase towards software and applications by 2028, as growth in spending by hyperscalers and cloud providers is expected to slow sharply, Morgan Stanley Research said in a report.

Morgan Stanley forecasts year-on-year growth in hyperscaler and neocloud capital expenditure to fall from 93 per cent in 2026 to 61 per cent in 2027 and 14 per cent in 2028. The brokerage said this points to a shift from heavy spending on chips and data-centre infrastructure towards software and AI applications.

"This indicates we could soon see a transition from hardware to software/services in 2028, and we expect this shift heading into next year as this is priced in," the report said.


The report added that the transition does not mean the end of the infrastructure buildout. Instead, AI-related demand is expected to remain strong because the industry continues to face shortages of power, labour and regulatory capacity.

Power is identified as the biggest constraint on faster data-centre development. Morgan Stanley estimates a 57 gigawatt power shortfall between 2026 and 2028, equivalent to a 30-40 per cent shortfall. It also said local opposition has delayed or cancelled USD 286 billion of US data-centre projects since 2025.

"We continue to believe the demand for compute is likely to significantly exceed supply for many years to come," the report said, referring to the computing capacity required to run AI models and applications.

Morgan Stanley expects the next 6-12 months to be a transition period, with opportunities broadening beyond semiconductor and infrastructure companies to early software providers and businesses using AI to improve productivity and revenue.

The report said AI adopters are already reporting measurable benefits. Companies across sectors reported an average 9.6 per cent net productivity increase over the past year, while Morgan Stanley expects around 100 basis points of net margin expansion through 2027 to be linked to AI adoption.

The brokerage expects software leadership to develop first in infrastructure and cybersecurity before spreading to applications, while services companies are likely to benefit later once businesses establish a more repeatable return on investment from AI.

Overall, Morgan Stanley said the AI cycle is likely to broaden rather than follow a simple handover from semiconductors to infrastructure and then software, as power and other bottlenecks continue to extend the infrastructure phase.