The 5 Names Holding Up the AI Infrastructure Rally — Nvidia, TSM, Micron, Vertiv, SMH

The debate over whether AI is real is over. Microsoft, Google, Amazon, and Meta have already answered — each committing hundreds of billions in AI capital expenditure this year alone. Trillion-dollar companies don't lock in that level of capital to run an experiment. They lock it in because the decision is already made.

That reframes what investors need to do. The question isn't "is AI real?" It's "how do I get onto the path this capex is flowing down?" Follow that path and you don't just see GPUs and software. You see the foundry that etches those chips, the memory that sits on them, the power and cooling equipment inside the data centers — the whole chain scales in lockstep.

Here are the five infrastructure-layer names I actually hold, and why each one earned a spot in the portfolio.

1. Nvidia — The GPU Backbone With No Credible Competitor

Nvidia is the GPU backbone of the entire AI ecosystem. There's no credible competitor at this scale right now. Every data center being built runs on Nvidia.

Early in 2025, the stock plunged 37%. The headlines were brutal — "overvalued," "the AI trade is dead." Nvidia finished that year up 39%. The investors who held, or better yet added into the drawdown, made out exceptionally well. Not because they were lucky, but because they understood what they owned.

Picks and shovels. That single framing is the most accurate lens for Nvidia. It doesn't matter which gold strike pays off — whoever sells the picks and shovels gets paid. Nvidia's GPUs are the picks and shovels of the AI gold rush.

2. TSM (Taiwan Semiconductor) — The Single Foundry Bottleneck

TSM is where those GPUs actually get manufactured. They just raised their 2026 revenue growth outlook to above 30%. A company already growing at 30% just guided the next year higher.

Why? Because AI silicon demand is relentless. Foundries don't get built overnight — a new fab takes years. GPU demand, meanwhile, accelerates every quarter. Demand is a vertical line; supply is a slow ladder. That gap is TSM's pricing power.

Holding Nvidia without TSM is like watching the side of the chain that sells GPUs without watching the side that has to manufacture them. The logic only works if you own both ends.

3. Micron — HBM Is Already Sold Out Through 2026

Every AI chip needs high-bandwidth memory to function. Micron's HBM is already sold out through 2026. Not "selling well" — completely sold out.

The result is in the EPS, which has tripled in two years. Memory is historically a cyclical industry, but HBM is tethered directly to the AI capex cycle, which smooths a lot of that volatility. Supply is booked multiple quarters out; demand grows with every new capex announcement.

Retail investors tend not to take Micron as seriously as Nvidia or TSM. That gap is the opportunity. No AI chip functions without HBM. That alone places this company at the center of the cycle.

4. Vertiv — The Power and Cooling Layer Most People Skip Entirely

This is the name many investors don't even look at. Massive GPU clusters consume enormous power and dump enormous heat. Without the equipment to supply that power and remove that heat, data centers don't run.

Vertiv supplies that exact layer — power management and thermal infrastructure. It isn't as sexy as talking about language models, so it gets less press coverage. But the money flows through here too. As AI data centers scale, Vertiv's backlog builds.

Owning only GPU names gives you exposure to the "compute" story. Adding power-and-cooling names like Vertiv extends that to the "infrastructure" story. Those are different bets.

5. SMH — The Whole Ecosystem Without Single-Name Volatility

For investors who can't stomach single-stock volatility, SMH (the VanEck Semiconductor ETF) is the cleanest basket I know of. You get Nvidia, TSMC, Micron, and the full semiconductor ecosystem in one ticker. It's up over 133% in the last year.

If you want AI exposure but don't want to live through a 37% drawdown in any single name, SMH is close to the right answer. Diversification across multiple names means one earnings miss doesn't shake the whole position, and you still ride the ecosystem up.

Personally, I hold both individual names and SMH. The individual names are where I pursue alpha; SMH provides the beta underneath.

Why I Read These Five Names as One Story

Because they're all different slices of the same capex cycle.

Nvidia designs → TSM manufactures → Micron adds memory → Vertiv supplies the power and cooling to run it → SMH wraps the whole thing. They look like five separate companies, but when a single capex dollar is committed, all five benefit simultaneously.

The most underpriced risk I see is that the market is still discounting this capex cycle as a one- or two-year event. My read is three years minimum, probably longer. The Fortune 500 budget just got set — execution is only beginning.

FAQ

Q: Nvidia has already run a lot. Is it too late to enter? A: "Already run a lot" is only true on a one-year chart. Zoom out to three years and the AI capex cycle is only in its middle innings. How you enter matters. Don't lump-sum — scale in. It's safer psychologically and financially.

Q: Micron is up 2x already. Can it keep going? A: HBM sold out through 2026 means revenue is already booked. There's still runway where pricing shows up in earnings. Remember, though — memory is cyclical. HBM has extended the cycle; it hasn't eliminated it.

Q: Should I invest in Vertiv if I don't know the name? A: Name recognition and return potential are two separate things. Vertiv is a major supplier of data-center power and cooling solutions. If the unfamiliarity is the blocker, consider starting with SMH, which captures this ecosystem in a basket.

Q: Which is better — SMH or SOXX? A: Both are semi ETFs but weighted differently. SMH is concentrated in top names (Nvidia, TSM are big), while SOXX is more diversified. For exposure to the top AI beneficiaries, SMH is tighter; for broader cycle exposure, SOXX is more reasonable.