13 Macro Themes for AI Infrastructure — And Why Transformers Is the One That Blocks All the Others

Paradis Labs recently published a list of 13 macro themes for AI infrastructure investment. Investor Johan Ekroth mapped each theme against his portfolio and arrived at a conclusion about theme 13 that GridReadiness has been documenting since 2025:

"Grid transformers are a hardware bottleneck. Order backlogs are 2–3 years. No new datacenter gets built without solving the power delivery problem first." — Johan Ekroth, @ekroth

The statement is correct. The implication goes further than most investors have modelled: theme 13 is not one of 13 parallel bets. It is the dependency that determines whether themes 1 through 12 can be deployed at all.

THE 13 THEMES — AND THE DEPENDENCY STRUCTURE

Here are the Paradis Labs themes, mapped against their dependency on resolved power infrastructure:

Themes 1 through 11 are components of AI compute and networking infrastructure. They are essential. They are also entirely dependent on data centers being built and powered. A 800G transceiver that ships on schedule but sits in a warehouse waiting for a powered data center generates zero return.

Theme 12 (fuel cells) and Theme 13 (transformers) are the power delivery layer. Without them, the stack above cannot function. But as we will show, they are not equivalent alternatives — they address different parts of the power problem at different scales.

THEME 13: WHY TRANSFORMERS CONTROL THE ENTIRE STACK

The transformer bottleneck is not a supply chain inconvenience. It is a physical constraint with no workaround at scale. A 100MW AI data center campus requires high-voltage transformers to step transmission-level electricity (typically 63kV–225kV in Europe, 115kV–500kV in the US) down to the voltage levels that server infrastructure can use.

There is no substitute for this function. The physics require it. The lead times are now 48–60 months at major manufacturers, 20–32 months at second-tier European producers. Ekroth cites 2–3 years — consistent with the second-tier range that GridReadiness tracks monthly.

Ekroth identifies $ETN (Eaton) as the "liquid anchor node for grid infrastructure" — a correct characterisation. Eaton manufactures the switchgear, UPS systems and power distribution units that sit downstream of the transformer in the data center power chain. They are a critical part of the infrastructure stack and equally supply-constrained. GridReadiness covers switchgear lead times of 18–36 months as a compound bottleneck that even projects with secured transformer slots still face.

THEME 12: FUEL CELLS — THE OFF-GRID PARTIAL SOLUTION

Ekroth identifies Bloom Energy ($BE) as the clearest gap in his portfolio — "Aschenbrenner's anchor position" — citing fuel cells as the solution for data centers that cannot wait for grid capacity.

This is an important and partially correct framing that deserves precise analysis.

What fuel cells can do

Bloom Energy's solid oxide fuel cells can provide reliable baseload power generation on-site, without waiting for utility grid connection. A data center with Bloom boxes can theoretically operate independently of the transmission grid — drawing natural gas or hydrogen directly, generating its own electricity on-site. For projects in markets where grid connection timelines are 5–10 years (Northern Virginia, parts of Texas), this is genuinely compelling.

What fuel cells cannot do at hyperscale

The economics of fuel cell power at 100–500MW scale are fundamentally different from utility-scale grid power. Current Bloom Energy installations typically serve tens of megawatts. A 500MW hyperscale AI campus using fuel cells would require an extraordinary number of units, significant hydrogen or gas infrastructure, and ongoing fuel cost that makes French nuclear baseload at €50/MWh look extraordinarily competitive by comparison.

Fuel cells are a genuine solution for specific constrained situations — smaller deployments, interim power while grid connection completes, off-grid edge deployments. They are not a structural replacement for grid-connected transformer infrastructure at hyperscale.

THE CORRECT INVESTMENT FRAMEWORK

For investors evaluating the 13 themes, the dependency structure matters more than the list structure. The correct mental model is not 13 parallel bets — it is a dependency tree:

Foundation layer (must resolve first)

Theme 13 (transformers) and Theme 12 (fuel cells as partial alternative) — without resolved power delivery, no other theme generates returns from data center deployment.

Infrastructure layer (unlocked by foundation)

SiC/GaN power (Theme 10), smart grid and metering (adjacent to Theme 13), advanced packaging (Theme 1) — these benefit directly from data center construction activity.

Compute and networking layer (highest leverage once deployed)

HBM/NAND (Theme 2), 800G transceivers (Theme 6), SiPh foundry (Theme 8) — enormous addressable market, but returns require data centers to be built and operational.

WHERE EUROPE SITS IN THIS FRAMEWORK

For each of the 13 themes, Europe's position varies. On the theme that matters most — transformers — Europe has a structural advantage that the US lacks:

• Competitive GOES (Grain-Oriented Electrical Steel) production: ArcelorMittal, ThyssenKrupp — no monopoly equivalent to Cleveland-Cliffs
• Second-tier transformer manufacturers with 20–32 month lead times vs 48–60 months for US OEM equivalents
• France: nuclear baseload at €50/MWh, RTE grid connection 18–36 months on brownfield sites
• Available industrial land with legacy HV infrastructure — brownfield sites that dramatically compress grid connection timelines

For a US investor building a portfolio across all 13 themes, the geographic arbitrage available in European power infrastructure is the highest-conviction asymmetric opportunity in the stack. The market has not priced European grid-connected infrastructure assets to reflect the US transformer and grid connection constraints.

THE GRIDREADINESS COVERAGE MAP

GridReadiness covers the power delivery layer of this stack in depth — the layer that determines whether all other themes generate returns:

• Transformer lead times: monthly tracker across 14 European manufacturers
• GOES supply chain: CLF monopoly analysis, European alternative supply
• Switchgear lead times: the compound bottleneck downstream of transformers
• Grid connection timelines: France vs Germany vs UK vs US by market
• Brownfield site intelligence: grid-ready sites in Europe before they are widely marketed
• Fuel cell vs grid economics: when each solution makes sense at what scale

If you are building a portfolio across the 13 Paradis Labs themes, the question to answer first is whether Theme 13 is resolved for each asset in your pipeline. Everything else follows from there.