Morgan Stanley’s technology analysts are urging investors to look beyond headline-sized capital expenditures on artificial intelligence and consider the revenue runway that could make those investments profitable by 2028. In a note to clients dated Oct. 13, Katy Huberty, the firm’s global director of research, highlighted a chart showing a cluster of tightly connected technology companies through which significant capital is flowing, often tied to and emanating from relationships with OpenAI.

“Our team believes the sustainability of the current investment cycle ultimately depends on whether AI generates durable cash flows to support returns on the significant capital being committed,” Huberty wrote. “And their bottom-up analysis suggests they will, as they forecast US$1.1 trillion in AI software revenue in 2028 at typical software margins.” The forecast frames the current capex wave — including spending on data centers, GPUs and software platforms — as an early-stage deployment rather than the peak of the cycle.

Morgan Stanley’s assessment matters because institutional and retail investors have been grappling with whether the shift to generative AI justifies multi-year, capital-intensive buildouts by hyperscalers, chipmakers and enterprise software providers. The bank’s research suggests the scale of expected software revenues could produce returns that offset today’s heavy investment, particularly for firms that capture subscription-like software margins and recurring monetization streams.

The note underscores a concentration risk in the economics of AI: capital and value appear to be aggregating around a small number of platforms and service providers that control model training, hosting, and distribution. That concentration can amplify returns for dominant firms while leaving less-connected companies dependent on partnerships or reselling agreements. Huberty’s chart—circulating widely on investor desks—illustrates how capital is funneled through networks anchored by prominent AI developers and cloud providers.

The implications extend beyond corporate balance sheets. Policymakers face choices about competition policy, subsidies and trade restrictions that affect supply of critical components such as advanced semiconductors. U.S. support for domestic chip production and export controls on high-end AI chips are already reshaping how firms plan capacity and supply chains—factors that feed into the timing and size of returns. Energy use and sustainability are other practical constraints; large-scale model training demands substantial power and specialized cooling infrastructure.

For investors, the near-term question is patience and selection. If Morgan Stanley’s revenue forecast materializes, companies with subscription-like pricing, scalable cloud operations and efficient model deployment could post high incremental margins that justify upfront outlays. By contrast, firms that fail to convert trials into repeatable enterprise spending or are forced into heavy discounting may struggle.

What to watch in the next two to four years are adoption rates among enterprise customers, visible monetization metrics from AI products, and policy moves that alter the cost or availability of chips and data-center capacity. If those trends align with Morgan Stanley’s bottom-up revenue trajectory, the bank’s forecast that capex will pay off by 2028 may be confirmed; if not, investors could be left recalibrating expectations about how soon the AI investment cycle becomes a durable profit engine.