The Bridge Ep. 19: Do You Need to Own AI to Profit From It?

Summary

In this episode, Sonali Basak speaks with Stonepeak Chairman and CEO Mike Dorrell about why infrastructure may be the “boring” but durable way to invest in the AI boom. Dorrell breaks the asset class into digital, power and energy, and transport and logistics, arguing that Stonepeak is intentionally diversified and not chasing only the hottest digital names. He says AI has materially improved the economics of data centers by stretching hyperscaler contracts from roughly 5-7 years pre-AI to 15-20 years now, but he also warns that data-center valuations are elevated and could compress if AI capex slows. The conversation repeatedly returns to power as the binding constraint: power availability, permitting, community pushback, and grid investment are now central to data-center expansion. Dorrell’s central point is that infrastructure investors can capture the AI cycle through contract-backed assets and utilities-like cash flows without having to predict the eventual winners in AI software or models.

Notes

• Stonepeak is presented as a large infrastructure and real-assets manager with AI exposure mainly through physical infrastructure.
• Infrastructure is framed as airports, roads, and other everyday assets people can easily understand.
• The asset class is traced to government fiscal stress and privatization in Australia.
• Macquarie is credited with inventing and exporting the modern infrastructure asset class.
• Dorrell divides infrastructure into digital, power and energy, and transport and logistics.
• Digital is about a third of Stonepeak’s work, and Dorrell says it is the most over-exuberant part of the market.
• Stonepeak’s model is to find predictable trends and collect toll-like cash flows.
• Dorrell links infrastructure duration risk to bond-market risk and higher rates.
• Stonepeak’s “boring” AI exposure starts with contract-backed data-center development.
• Before AI, data-center contracts were typically 5-7 years, which Dorrell says was too short for Stonepeak.
• AI-driven demand shifted bargaining power toward data-center owners and stretched contract duration.
• Dorrell wants a low-to-mid-teens equity return on long-term contracted deals, with a margin over Treasury yields.
• Counterparty credit quality is the core underwriting issue in data centers.
• He sees the strongest counterparties as Microsoft, Nvidia, Meta, and Google, while frontier AI labs are riskier.
• Long-duration government bonds are portrayed as especially dangerous after a rate regime shift.
• Data-center valuations are said to be 20-25 times earnings.
• A slowdown in AI capex would likely compress data-center multiples.
• Data-center development depends on three hard problems: land, power, and hyperscaler trust.
• Power delivery is described as the hardest constraint in data-center siting.
• Grid development is a separate skill set from data-center operations.
• Public opposition and state-level moratoriums are emerging as real obstacles to AI infrastructure in the U.S.
• Despite moratorium chatter, Dorrell says buildout is still strong and contracted capacity remains in demand.
• Community benefit agreements are becoming part of the approval process for data-center projects.
• A 200 MW data center may need 300 MW of power capacity, with 100 MW contributed back to the grid.
• Dorrell says no prior buildout comes to mind at this scale, and AI spending may already equal 3%-4% of GDP.
• He says a report put AI spend at roughly 3%-4% of GDP, implying macro sensitivity if it slows.
• Stonepeak’s broader infrastructure lens also includes food, logistics, and other steady-demand assets.
• Dorrell treats energy investing as a long-term call option on disruption, with income along the way.
• The episode emphasizes that grid and energy assets can be attractive exposure to both AI and the energy transition.
• Transport and logistics are presented as beneficiaries of trade rerouting and less efficient global commerce.