I tried to find which AI infrastructure sector hasn’t already been fully repriced

I’ve been trying to look at the AI trade from a slightly different angle.

A lot of the obvious AI infrastructure winners have already had huge moves. Optical networking, data center construction, cooling, and power equipment have all been bid up hard. So instead of asking “what is the next AI stock,” I wanted to ask a more boring question:

Which AI-linked sector has lagged the broader AI/growth trade over the last two years?

This is not meant to be a “best stocks to buy” list. It is just a relative-performance screen to find areas worth doing more research on.

How I built the baskets

I grouped public AI-linked companies into sector baskets: AI power supply, power distribution, compute, semiconductors, cooling, networking, data center construction, materials, critical minerals, defense AI, healthcare AI, etc.

The ticker selection was thesis-first. I tried to include companies where revenue could plausibly benefit from AI infrastructure capex or AI adoption, not just companies that mention “AI” in a press release.

For example, the AI Power Supply basket included names like:

SMR, NNE, FCEL, PLUG, NPWR, ORA, FLNC, EOSE, GWH

I excluded the obvious mega-cap AI winners because I was specifically trying to find parts of the value chain that may not have fully run yet.

How I calculated it

I pulled adjusted price data from 2024-04-29 to 2026-04-28 and calculated each ticker’s 2-year return. Then I compared every ticker against QQQ, since QQQ is a reasonable proxy for the AI/growth trade.

Over that same period, QQQ returned about +53.6%.

For each sector, I used the median return, not the average. I did that because one monster stock running 500–1000% can make an entire sector look hot, even if most of the basket did not participate.

Then I calculated:

Sector normalized return = sector median 2-year return minus QQQ 2-year return

Main results

| Rank | Sector | Vs QQQ |

|—:|—|—:|

| 1 | Health AI | -103.5% |

| 2 | Defense AI | -86.1% |

| 3 | AI Power | -27.4% |

| 4 | Industrial AI | +5.2% |

| 5 | Semi/HW | +14.8% |

| 6 | Compute | +18.7% |

| 7 | Power Dist. | +40.3% |

| 8 | Telecom | +48.9% |

| 9 | Minerals | +62.7% |

| 10 | Components | +84.7% |

| 11 | Materials. | +96.7% |

| 12 | Auto AI | +127.4% |

| 13 | Cooling | +139.0% |

| 14 | DC Buildout. | +273.6% |

| 15 | Networking | +568.0% |

The part that stood out to me:

Healthcare AI and Defense AI were the most underperforming baskets overall, but they are more AI application sectors than core AI infrastructure sectors.

For core AI infrastructure, the cleanest laggard was:

AI Power Supply

That basket had a median 2-year return of about +26%, while QQQ was up +54%. More than half the names in that basket were still below QQQ.

Meanwhile, networking, data center construction, and cooling have already had massive moves. That does not mean they cannot keep going, but the market has clearly already found those bottlenecks.

My takeaway

The AI trade has already aggressively repriced the obvious infrastructure bottlenecks: networking, cooling, and data center buildout.

But the power side still looks relatively under-ran on a 2-year normalized basis.

Curious if anyone else is looking at the AI power bottleneck the same way, or if there are better ways to build the basket.