Shark Tank’s Kevin O’Leary says if he were 25 today, he’d chase these two booming opportunities in the world of AI

Kevin O’Leary says AI growth will be “exponential” and focuses a 25-year-old’s ambition on (1) helping sub-500-employee businesses implement AI tools (36M firms, just under half of U.S. GDP) and (2) building AI data centers. He highlights infrastructure bottlenecks: only ~5 GW of data center capacity is under construction vs. much higher demand, with Goldman Sachs estimating AI will drive data center power demand up 165% by decade-end. His ventures include backing a $70B data center industrial park in Alberta (7.5 GW) and a $100B Utah data center project, both cited as facing local scrutiny and schedule concerns.

Analysis

The investable edge here is not the generic AI enthusiasm; it is the capital-intensity of the buildout. That favors the firms that can monetize compute, storage, and workflow lock-in at scale—MSFT, AMZN, and GOOGL—because they get paid twice: first on infrastructure spend, then on usage and migration stickiness. The bigger second-order winner may be the financing layer, where GS and MS can underwrite project finance, structured lending, and advisory fees tied to datacenter land, power, and M&A.

The weak link is the “implementation” layer for small businesses. That market is likely to fragment into low-ARPU, high-churn services work with limited moat, meaning most of the economic rent should accrue to software platforms and integrators with existing distribution rather than new standalone boutiques. If AI adoption among SMBs is slower than the headline narrative, the near-term effect is not revenue collapse at hyperscalers, but a re-rating risk for the broader software stack that is priced for fast monetization.

The real catalyst path is 1-3 months: next earnings seasons will reveal whether hyperscaler capex keeps rising faster than cloud growth, and whether management still frames AI as capacity-constrained rather than demand-constrained. Over 6-18 months, the key falsifier is unit economics: if datacenter utilization, power costs, or depreciation outpace incremental AI revenue, the market will punish the infrastructure complex despite continued spend. The contrarian view is that this is already consensus at the top of the value chain; the trade is not “AI is good,” but “who captures the spread between capex and monetization.”

Key Decisions for Investors

• Maintain a core long basket in MSFT/AMZN/GOOGL into the next earnings cycle, but add only on pullbacks; thesis is that these names capture both AI infrastructure spend and recurring usage economics. Falsifier: any guidance that implies capex growth is decelerating faster than cloud/AI revenue.
• Use GS and MS as a lower-beta second-order long on data-center financing and project-advisory activity; this is a 6-18 month theme, not a one-quarter catalyst. If deal pipelines do not inflect alongside AI infrastructure announcements, trim.
• Avoid chasing pure-play SMB AI implementation names until there is evidence of recurring software-like margins; if you want exposure, wait for customer retention and gross margin proof rather than first-mover hype. This is a watchlist, not a forced trade.
• Set an alert for hyperscaler capex commentary next earnings; if combined spend growth slows materially, expect multiple compression first in the AI-infrastructure complex and then in adjacent software beneficiaries.