My positions: I trade some of these names and watch the rest. Positions get disclosed inline where I hold one. Nothing below is investment advice. It's my read, next to my own tool's read, and you'll see us disagree.
Key points
- One pick in every layer of the AI hardware stack. GPU, memory, storage, CPU, networking, manufacturing, and power.
- Each pick sits next to its grade from our own stock score tool: NVDA 83/A, WDC 81/A, TSM 89/A+, and VRT 78/A agree with the pick. Micron (91/A+) is a partial: I like it, I don't love it here. INTC (36/D) and COHR (55/C) don't agree with the tool at all, and I'm holding both anyway.
- Where I disagree with my own scoreboard I say so and explain why. That's the whole point of this piece.
We built a stock score tool that grades any US stock straight from its SEC filings. No opinions. Just what the paperwork says. Then I picked my favorite stock in every layer of the AI hardware stack and put my picks next to the tool's grades.
Fair warning. We don't always agree. And the disagreements are the most useful part.
If you want the full map of what each layer actually does, Dennis wrote the field guide. This is just the picks.
GPU: Nvidia (NVDA). Tool says 83/A. We agree.
Boring pick. I know. But the tool and I land in the same place for the same reason. The filings are a machine: $176 billion in net income over the last four reported quarters and growing. More cash than debt. Buying back stock. There are cheaper AI stocks. There is no safer engine in the stack.
Memory: Micron (MU). Tool says 91/A+. I like it. I don't love it.
The highest score our tool has ever handed out, and I get why. Profits up eightfold in a year. Fortress balance sheet. Reasonable multiple. But this is exactly the setup where I hold back from a full endorsement. Memory is the most boom-bust business in tech, and earnings this good usually mean you're closer to the top of the cycle than the bottom. The tool grades the last four quarters. It can't see a cycle turning, and it can't price in how fast this stock moves once sentiment shifts. I'm long-term bullish on memory scarcity. I'm just not going to pretend this is a sleep-at-night position the way Nvidia or Western Digital are.
Storage: Western Digital (WDC). Tool says 81/A. We agree.
The unglamorous layer, and one of the tool's cleanest scores. $6.5 billion in trailing net income, growing fast, share count actually shrinking as the company buys back stock, more cash than debt. Storage never leads an AI rally. It follows it, because drives get ordered once a data center is actually filling up with data, not when it's announced. That lag is exactly why I like owning it here instead of chasing whatever layer is loudest this week.
CPU: Intel (INTC). Tool says 36/D. I'm overriding my own scoreboard.
Here's the fun one. My own tool gives Intel a D and the tool is right. Trailing earnings are negative. The share count grew when the government took its stake. Debt is real. Every word of that is in the filings.
I'm picking it anyway, and I've already written up why. The government owns 10% and does not want its own stock going to zero. The former CEO of SK Hynix now runs Intel's packaging business, right as packaging is one of the tightest chokepoints in the stack. And turnarounds always look terrible on trailing numbers. That's what makes them cheap. If Intel's filings looked like Nvidia's it wouldn't trade where it trades. The tool prices what happened. I'm betting on what happens next. One of us will be wrong and you'll get to watch.
Networking and optics: Coherent (COHR). Tool says 55/C. Second disagreement.
Coherent is genuinely profitable, $469 million trailing net income and growing, but the tool dings it hard on two things: the share count is up almost 26% in a year on offering filings, real dilution, and the stock trades near 135 times trailing earnings, priced for a future that hasn't shown up in the filings yet. That's a fair C. My case for owning it anyway is the same one I'd make for any optics name right now: as GPU clusters scale, more of every data center dollar goes to simply wiring the machine together, and Coherent makes the transceivers doing that wiring. The dilution paid for capacity, not for survival, which is a very different kind of dilution than Intel's. This multiple is rich. I know it, you know it, let's move on.
Manufacturing: TSMC (TSM). Tool says 89/A+. We agree.
The second-highest score on this list, and the layer everything else depends on. Doesn't matter whose GPU wins, Nvidia's or AMD's, it still gets built at a TSMC fab. The custom silicon coming out of Google and the other cloud giants goes through the same door. The filings back it up: this is the one company in the stack that wins no matter who wins the design race. My only real hesitation is the one everyone already knows about. Geography. Taiwan sits where it sits, and no 10-K fixes that. I own it anyway. A cleaner map doesn't make a worse business the safer choice.
Power and cooling: Vertiv (VRT). Tool says 78/A. We agree.
Vertiv is the pure play on the layer nobody priced in early, the power and cooling gear that keeps a GPU cluster from melting itself. Profits are growing, debt is manageable, and capital spending is up 208% year over year, paid for out of real profits rather than a cash scramble. That's the same test I ran on Intel's dilution, and Vertiv comes out clean where Intel doesn't. The one soft spot is the chart. Price sits above the 200-day but below the 50-day, so I'm holding the thesis here instead of chasing the stock at any price.
The scoreboard and me
Four out of seven picks are clean agreements with the tool. Nvidia, Western Digital, TSMC, and Vertiv all have the filings to back the story, and those are the sleep-at-night positions. Micron is the in-between one. I like the business and trust the score, but I'm still wary of where we sit in the memory cycle. Then there are the two real disagreements. Intel is a straight bet against the trailing paperwork, priced for a turnaround that hasn't shown up yet. Coherent is a good business at a rich multiple, priced for growth the filings haven't caught up to. Two different reasons to disagree with my own scoreboard, and I'm fine with both. Know exactly which number you're arguing with before you own the stock. That's the honest way to use a scoring tool. It won't predict the future for you. It'll just tell you, bluntly, when your favorite stock has ugly filings.
This is general market commentary and opinion, not investment advice. Scores are computed from public filings and change as new filings land. Markets can go down as well as up, and you can lose money. Always do your own research and consider speaking with a licensed financial professional before making any investment decision.
