The model race hasn’t ended. But the infrastructure race has started pulling capital that used to queue behind frontier model announcements.
That’s the claim at the center of a thematic analysis Morningstar reportedly published this week. According to Wire reporting, the analysis frames AI infrastructure’s physical layer, specifically power equipment, copper wiring, and electrical switchgear, as the primary investment bottleneck of what analysts are calling the agentic era. The specific report was not independently accessed at publication. Key figures are attributed to Morningstar as reported and should be treated as analyst estimates pending independent verification of the underlying methodology.
What makes this framing worth examining is not the Morningstar report itself. It’s that the verified infrastructure data from multiple sources already assembled in this hub is telling the same story the analyst thesis describes.
Section 1: What the Verified Data Shows
Start with power. Vertiv raised its 2026 revenue guidance to $14 billion and reported an 80% year-over-year surge in its AI data center backlog as of its most recent earnings. Vertiv makes the power distribution and thermal management equipment that sits between the grid and the GPU, the physical layer that determines whether a data center can actually run. An 80% backlog surge doesn’t reflect ordering enthusiasm. It reflects a supply chain that cannot keep pace with deployment demand.
The power draw behind that demand is documented by the IEA. Data center electricity use surged 17% in 2025, according to IEA data reported by this hub in April, driven by accelerating AI agent workloads. That figure covers one calendar year. The trajectory it implies for 2026 and 2027, as inference scales and agentic workflows move from pilot to production, is what’s motivating the current investment thesis.
Site selection decisions reflect this shift in real time. The relevant question for operators choosing where to build AI infrastructure is no longer primarily about network latency to end users. It’s about proximity to power. Developers are routing data center capital to locations with available grid capacity: Brazil and orbital compute deployments emerged as targets precisely because domestic grid access in the US and Europe faces constraints that take years to resolve. That’s a site-selection logic built around power scarcity, not bandwidth optimization.
Section 2: The Industrial Supply Chain Layer
The Morningstar framing adds specificity to a pattern already visible in the data. The analyst thesis reportedly highlights not just power demand in aggregate, but the industrial equipment that enables it: switchgear (the high-voltage switching apparatus required for large-scale power distribution), transformers, and copper wiring.
These components have long lead times under normal demand conditions. Under the current AI infrastructure build cycle, the Morningstar analysis reportedly characterizes switchgear delivery backlogs as extending well into 2027. This figure has not been independently confirmed from the specific report, it’s attributed to Morningstar as reported, but it is consistent with what Vertiv’s backlog data and general industrial supply chain reporting indicate about power equipment availability.
The investment implication the analysis reportedly draws: exposure to industrial infrastructure companies in the power equipment segment may represent an indirect AI trade with lower volatility than direct model company exposure. The Morningstar analysis reportedly estimated global AI infrastructure build-out costs at $85 trillion. That figure requires prominent qualification. If accurate, it would approach the scale of current global GDP, an extraordinary magnitude. No methodology, time horizon, or geographic scope for this estimate was accessible at publication. Treat it as a reported analyst projection, not a verified market consensus figure, and apply independent judgment before using it in investment analysis.
Section 3: Opposition and Constraints
The infrastructure investment thesis has a constraint layer the bullish framing tends to underweight. The Data Center Coalition map documented this week shows organized opposition from building trades unions to new AI data center construction in several US markets. Permitting delays, labor disputes, and community opposition are not theoretical risks – they’re active friction points in multiple planned build-outs.
Power procurement itself faces regulatory timelines that equipment backlogs compound. A utility interconnection request filed today in a constrained US grid region may not resolve for 18 to 36 months. That timeline sits upstream of the switchgear and transformer backlog. The bottleneck is sequential, not parallel.
Investors treating the infrastructure thesis as a straightforward demand story should price in the regulatory and opposition surface area. The opportunity is real. The timeline uncertainty is also real.
Section 4: What the Investment Thesis Actually Requires
The Morningstar framing reportedly positions “speed to power” as the new primary site-selection criterion, framing it as a replacement for fiber latency in the previous era. This characterization deserves scrutiny as analyst framing rather than established industry practice. Some workloads remain latency-sensitive. Inference at the edge, real- time agent interactions, and regulated-data applications all have location requirements that pure power-proximity logic doesn’t fully address.
The more precise formulation: power availability has become a necessary condition for large-scale AI infrastructure deployment in a way that it was not three years ago. It doesn’t eliminate other site-selection criteria. It now competes with them at the top of the decision hierarchy.
For investors, the practical implication of the industrial layer thesis, regardless of whether the $85T aggregate figure proves accurate, is that supply chain bottlenecks in power equipment create pricing power for manufacturers with existing production capacity and order books. The 80% YoY backlog surge at Vertiv is evidence of that pricing power in action.
Section 5: The Limits of the Thesis
The infrastructure investment thesis has a failure mode: if AI adoption plateaus before the current build-out completes, the assets being financed today become stranded capacity. That’s not a prediction, it’s a risk that the bullish framing tends to underweight.
The $85T estimate, to the extent it’s directionally useful at all, implies a build-out timeline measured in decades. Infrastructure investment theses with multi-decade time horizons have historically produced returns, but they’ve also produced the world’s most spectacular stranded-asset write-downs. The IEA data, the Vertiv backlog, and the coalition opposition map together describe a market that is genuinely constrained right now. Whether that constraint persists long enough to validate the full scope of current capital allocation is the question the data cannot yet answer.
What to watch: Vertiv’s next earnings guidance update is the most accessible real-time indicator of whether the backlog is clearing or extending. Interconnection queue processing times in PJM and ERCOT grid regions are early-warning signals for whether the power procurement bottleneck is structural or solvable within the current build cycle. The Morningstar report itself, once a URL is confirmed and the methodology is accessible – would significantly sharpen or weaken the investment thesis this brief outlines.
The TJS synthesis: The infrastructure bottleneck thesis doesn’t require the $85T figure to be actionable. The Vertiv backlog data and IEA demand trajectory are enough to establish that physical supply constraints are real and near-term. The analyst framing adds interpretive structure; the verified data provides the foundation. Investors evaluating industrial-layer AI exposure should weight the verified numbers over the aggregate estimate, and verify the Morningstar report directly before treating its projections as investment-grade analysis.