The AI infra trade cracks, the 5-year hold dies

THE CARRY

1. Cross-Podcast Themes

The AI Data Center Trade is Showing Its Cracks

A massive wave of private credit is chasing hyperscale yields, but Invesco is walking away. On Capital Allocators, Charlie Rose stated that his $85 billion real estate debt platform absolutely refuses to originate loans on data centers. He argues the underlying collateral suffers from rapid technological obsolescence and zero alternative-use value if the single tenant defaults. Listen to the full episode

Equity sponsors are discovering similar physical limits on their expansion models. Marion Calcine told The Infrastructure Investor Podcast that Ardian actually slashed the expansion pipeline for its Verne data center platform by 30 percent. Her team audited the local Icelandic electrical grid and found that the transmission bottlenecks directly contradicted the seller's aggressive tenant demand projections. Listen to the full episode

The broader municipal infrastructure required to support these facilities is also fracturing. Chris Beall noted on Dry Powder that the massive capital pouring into AI data center capital expenditures is blinding allocators to underlying physical decay. He argues that computing capacity cannot scale without basic municipal resources, pointing out that severe groundwater depletion and regional flight disruptions show acute physical bottlenecks that digital capital expenditures cannot solve. Listen to the full episode

The Standard Five-Year Exit Math is Dead

Investment committees are no longer defaulting to five-year underwriting models. On the Private Equity Funcast, Accordion CFO Jon Apter detailed how the historical private equity hold period has permanently fractured. Sponsors are now forcing either rapid two-to-three-year flips or stretching capital into seven-to-eight-year continuation vehicles. Watch full episode

When those longer holds try to find liquidity, broad enterprise auctions are stalling out on price. Cooley partner Danielle Fortier explained on M&A Science that buyers are bypassing full buyouts to target bilateral corporate carve-outs instead. Sellers are increasingly willing to accept these partial exits to generate liquidity for core operations without taking portfolio-wide valuation discounts. Watch full episode

These extended hold periods are breaking the operational benches that sponsors rely on to sell the business. Ted Bililies argued on Private Equity Spotlight that static management equity plans signed at acquisition lose their alignment as the hold stretches. He insists that sponsors must reset payout structures 12 to 18 months before a sale to keep second-tier commercial leaders from quitting during buyer diligence. Listen to the full episode

AI Code Generation is Collapsing Software Build Timelines

The correlation between raised capital and engineering output has collapsed. Will Robbins told How I Invest that generative tools have bent the historical startup capital ladder into an L-shape. He calculates that a fully featured product no longer requires a year of engineer time, just 50 billion tokens. Watch full episode

Mega-funds are internalizing this velocity to bypass traditional product roadmaps. Katie Yuan explained on the Private Equity Funcast how Hg Capital now drops in-house technical squads directly into its portfolio companies. This intervention shrinks enterprise software build timelines from 18 months down to just 90 days. Watch full episode

2. Best Of the Week

• Capital Allocators: Stephen McLennan revealed that Ontario Teachers' split the chief investment officer seat in two to resolve the inherent conflict between macro liquidity management and bottom-up active dealmaking. Read more.
• Dry Powder: Allison Kingsley unpacked how NOVA Infrastructure quadrupled EBITDA on a 7-ship government fleet by fixing a structural cash trap where 120-day receivables collided with two-week payrolls. Read more.
• How I Invest: Steven Rosenblatt recalled the intense operational pressure of pre-selling $60 million in mobile advertising inventory for an unbuilt Apple iAd product on nothing but Steve Jobs' public promises. Read more.
• M&A Science: Danielle Fortier warned that carve-out deals routinely stall after the letter of intent because buyers expect an 18-month transition services agreement while sellers plan to cut support after 90 days. Read more.
• Private Equity Funcast: Jon Apter discussed taking a 75% pay cut to leave Evercore for a portfolio CFO seat, immediately hiring a controller instead of a VP of FP&A to offset his own modeling background. Read more.
• Private Equity Spotlight: Mike Hollander argued that keeping sell-side diligence confined entirely to the CEO and CFO guarantees operational drift, requiring sponsors to widen the circle to functional leaders. Read more.
• The Infrastructure Investor Podcast: Sam Lissner highlighted that mid-market infrastructure funds delivered a 10-year median net return of 12.3% by avoiding the highly competitive auctions that mega-funds are forced to play in. Read more.

3. Most Quotable

"I'll share with you a controversial position that we've taken. We are not lending on data centers."

Charlie Rose on Capital Allocators · Oct 11, 2026. A rare flash of discipline in a private credit market desperate for AI exposure.

"Our philosophy is that we view leverage as an enhancer of returns but, but never the driver of returns."

Sam Lissner on The Infrastructure Investor Podcast · Oct 11, 2026. The mid-market infra playbook for surviving an era of permanently elevated interest rates.

"The dot-com boom, the reality is that you didn't have balance sheets. Everything was funded on hype and vaporware. Today, you have fortress balance sheets."

Steven Rosenblatt on How I Invest · Oct 11, 2026. Why the current AI capex cycle won't end in a 1999-style immediate operational collapse.

Bottom Line: While mega-funds pile blind capital into AI compute and data centers, the sharpest operators are recognizing that physical grid limits and 90-day software lifecycles are quietly breaking the old underwriting models.