Name: Can the AI Spending Boom Pay Off?
Uploaded: 2026-09-09T22:03:53+00:00
Duration: 307 s

Speakers

Brian Nowak — US Internet Analyst, Morgan Stanley

Summary

Morgan Stanley US Internet analyst Brian Nowak argues that the enormous AI infrastructure spending can pay off. He estimates major cloud providers will spend more than $1.4 trillion next year, with compute capacity potentially quadrupling to 120 gigawatts by 2028. His bottom-up analysis sees 25-50% ROIC across three AI business models: renting compute, integrated AI model developers owning infrastructure, and AI developers renting compute. He also highlights chip and software token-throughput gains as critical to realizing these returns.

Major cloud providers are estimated to spend more than $1.4 trillion on AI buildout next year.
AI compute capacity may quadruple from 2025 to 2028 to roughly 120 gigawatts.
Renting AI compute from next-generation data centers shows about 30% base-case ROIC.
AI model developers owning both models and infrastructure could earn 40%+ ROIC and about 75% incremental margins.
AI developers renting compute still show roughly 25% post-tax return potential.
Token pricing and chip/software-driven token throughput are key to AI unit economics.

Ideas

Brian Nowak US Internet Analyst, Morgan Stanley 1:10

AI infrastructure buildout can pay off.

The massive AI infrastructure buildout should pay off. Bottom-up analysis points to 25-50% returns on invested capital across three emerging AI business models, and the $1.4 trillion-plus cloud capex and quadrupling compute capacity do not by themselves imply poor returns because revenue and profit from serving and inference will determine the payoff.

Brian Nowak US Internet Analyst, Morgan Stanley 1:36

AI compute rental economics are attractive.

Renting AI compute power, the infrastructure layer of the AI economy, has attractive economics: a large next-generation data center can generate roughly 30% base-case ROIC, with scenarios ranging from low 20s to nearly 40% even if rental prices move, so cloud providers building GPU-filled data centers should earn strong returns despite heavy capex.

Brian Nowak US Internet Analyst, Morgan Stanley 2:19

Integrated AI model ownership yields highest returns.

AI labs that own both their model and their underlying infrastructure and sell API access can generate roughly 75% incremental operating margin and 40%+ ROIC, stronger economics than pure compute rental; token pricing and token throughput efficiency determine whether these returns materialize.

Brian Nowak US Internet Analyst, Morgan Stanley 3:26

Chip and software improvements drive AI returns.

Continued improvements in chips and software that increase token throughput are critical to the long-term unit economics of the AI ecosystem, supporting the suppliers and enablers of faster AI compute.

Brian Nowak US Internet Analyst, Morgan Stanley 3:40

AI compute renters still earn attractive returns.

AI developers that rent compute infrastructure instead of owning it still produce attractive unit economics, with roughly 30% incremental operating margin and 25% post-tax return potential, even though the infrastructure provider captures part of the economics.

This Morgan Stanley video, published September 09, 2026, features Brian Nowak discussing AI infrastructure buildout, Cloud providers, AI compute infrastructure, AI model developers with owned infrastructure, AI-SECTOR, AI developers renting compute. 5 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Brian Nowak · Tickers: AI infrastructure buildout, Cloud providers, AI compute infrastructure, AI model developers with owned infrastructure, AI-SECTOR, AI developers renting compute

Can the AI Spending Boom Pay Off?

Summary

Ideas

Up Next