Investing in Physical Bottlenecks Within the AI Infrastructure Build-Out

Original Title: Bloomberg Surveillance TV: June 25th, 2026

The AI infrastructure trade: Why the real bottlenecks are not where you think

The AI trade has moved from speculative hype to a structural race for physical capacity. While markets focus on the volatile margins of software giants, the real competitive advantage sits in the bottleneck economy: the essential infrastructure of power, cooling, and specialized hardware. This shift reveals a simple reality: the companies paying the most today are funding the systems that will define future efficiency. For the long term investor, the advantage is not in guessing the next software breakthrough, but in identifying the physical constraints that make AI possible. Understanding this requires looking past the quarterly noise of hyperscaler margin compression to the multi year, multi trillion dollar build out that the market is mispricing due to its focus on short term cycles.

The bottleneck paradox

The most common mistake market participants make is equating spending with failure. When hyperscalers report lower margins because of massive capital expenditure, the market punishes them. However, as Heath Terry of Citigroup points out, this ignores the basic economics of the build out. These companies are generating stable returns on invested capital of roughly 29 percent.

If you have the ability to borrow at 6 percent, or to issue equity and get a 29 percent return on it, you are supposed to do that regardless of what your stock price is doing.

-- Heath Terry, Citigroup

The system is responding to the force of enterprise demand meeting the limit of inference capacity. The bottleneck is not just a constraint; it is a profit engine. Sebastien Page of T. Rowe Price notes that the trade has expanded from GPUs to the entire data center supply chain, including cooling, electrification, and aerospace. The companies that own these physical bottlenecks are seeing their stock trajectories go vertical. The hidden consequence is that while software companies face disruption, those controlling the physical inputs, like power, storage, and specialized memory, are capturing the value that the software layer cannot function without.

Why immediate pain creates lasting moats

Conventional wisdom suggests that high capital expenditure is a warning sign of over leveraging. But when viewed through a systems lens, this spending is a rational response to a supply constrained environment. The market is currently penalizing the companies paying the bills, the hyperscalers, while rewarding the bottleneck owners, such as Micron.

The implication is that current market volatility is a temporary misalignment. As Page explains, the next leg of this revolution will be the handoff from infrastructure build out to actual user efficiency. The companies currently enduring the pain of high borrowing costs and lower free cash flow are building the infrastructure that will eventually allow them to integrate AI into their own operations. This is a case where immediate discomfort, sacrificing current margins, is the only path to a durable, long term competitive advantage.

You want to be where people are spending money not necessarily with the people who are spending the money.

-- Heath Terry, Citigroup

The Fed reaction function gap

The systemic risk is not just in tech; it is in the central bank inability to articulate a clear reaction function. Former New York Fed President Bill Dudley argues that the Fed reliance on task forces and vague forward guidance is a mistake. By looking to financial markets to set policy, the Fed creates a circular feedback loop: markets price in what they think the Fed will do, and the Fed looks at those market prices to decide what to do.

This creates a hidden cost: the transmission of monetary policy to the real economy becomes less efficient. If the market does not understand the Fed reaction function, the economy remains mispriced. Dudley highlights that financial conditions are currently at their most accommodative since 2021, which contradicts the Fed goal of restrictive policy. The downstream effect is that inflation risk remains underestimated, and the current approach may be delaying the necessary tightening that the real economy requires.

Key action items

  • Shift from software to infrastructure: Over the next 12 to 18 months, prioritize investment in the physical bottleneck providers, such as power, cooling, and semi cap equipment, rather than the end user software companies currently struggling with margin compression.
  • Ignore the margin panic: Recognize that hyperscaler capital expenditure is a rational 29 percent return on invested capital play. Do not conflate current margin compression with long term business failure.
  • Monitor the handoff phase: Watch for the transition point where AI users begin reporting actual margin expansion from internal efficiency gains. This is the next leg of the trade.
  • Account for inflation lags: Anticipate that headline inflation will remain an underestimated risk over the next 3 to 6 months. Position portfolios to hedge against the pent up inflation that current market breakevens are ignoring.
  • Audit Fed signal sensitivity: When evaluating market risk, assume the current policy is less restrictive than it appears. Prepare for potential tightening as the Fed eventually realizes that current financial conditions are too loose to meet their inflation targets.

---
Handpicked links, AI-assisted summaries. Human judgment, machine efficiency.
This content is a personally curated review and synopsis derived from the original podcast episode.