Hunting the Next Marvel? Jensen Huang Already Shared Clues on One Slide

The following guest post comes from Ziven.io, a public markets intelligence platform delivering data on companies exposed to bitcoin mining, artificial intelligence, and crypto treasury strategies. Originally published on June 18, 2026, by Cindy Feng.

Since Jensen Huang stood on the Computex stage and called Marvell “the next trillion-dollar company,” MRVL hasn’t looked back. A stock that traded between $50-$100 as recently as April now sits around $300, with an ATH around $316 and a gain of roughly 241% YTD. One sentence from Jensen, and a company re-rated by a quarter-trillion dollars.

Not surprising, a new exercise has begun: comb through everything Jensen says, find the next name he’ll bless and get rich.

I understand the impulse, but what’s clear from listening to Jensen’s whole keynote is that most people are watching the wrong thing. Jensen didn’t just drop a hot name, he laid out a full map of how an AI factory actually gets built, layer by layer, company by company. That map is the part worth knowing, because it still works long after the hype fades. I’m going to walk you through that specific slide, but first let’s start with the part that confused a lot of people.

RTX, DGX, DSX: worker, team, factory

Jensen split NVIDIA’s brands into three layers, each a bigger unit than the last:

RTX is the GPU, the worker. The chip that does the actual computing. One pair of hands. DGX is the system, the team. Wire a pile of those chips into a single machine and you’ve got a DGX. A crew acting as one unit. DSX is the infrastructure, the factory. The building those teams work in, plus the power, cooling, network, and software to keep thousands of them running around the clock.

RTX and DGX you’ve probably heard of. DSX is the new one, and it’s the one worth understanding, because it’s where NVIDIA stops selling you a chip and starts selling you a way to build the entire plant.

What DSX actually is

In Jensen’s words, DSX is “a blueprint, a reference design for building and operating AI factories at maximum efficiency and profitability”.

In plainer terms, it’s a recipe and a toolkit for booting up a gigawatt of compute and keeping it profitable. NVIDIA even named the toolkit’s parts: a digital twin to design and test the whole factory before a single rack ships (DSXSim), an operating system to run it once it’s live (DSX OS), and tools to pack more GPUs into the same power budget and flex with the grid (DSX Max LPS, DSX FLEX). The pitch is that 100 gigawatts of these factories come online before the decade is out, and that DSX-built ones run cheaper and lean on the grid more gently.

That all sounds like something NVIDIA would sell you by itself. It’s actually not the case.

No single company can build an entire AI factory

A one-gigawatt AI factory is now a $30-100 billion project, according to Jensen. At that scale it stops being a server room and becomes infrastructure on the order of a refinery or a power station.

NVIDIA can’t build that alone. It doesn’t pour concrete, run high-voltage lines, manufacture chillers, or negotiate with the local utility. And you can’t bolt those pieces on one at a time, because the chips, racks, network, power, and cooling all have to be designed together from day one. Every hour the factory sits idle is revenue lost, so a build this expensive has to work the first time.

Therefore NVIDIA did the sensible thing: it published the blueprint and assembled a coalition of partners to cover every layer it doesn’t do itself. That coalition has a name, the AI Factory Ecosystem, and Jensen put the entire roster on a single slide. That slide is the map.

The map: who actually builds an AI factory

Most of those company are private or listed overseas, but still a plenty of U.S. listed ones. I made a table to list all publicly traded names from the map. The last column is my rough read of how much of each business truly rides on the AI build-out, because being on the slide (could lean on marketing purpose) and being moved by it are two very different things.

Please note OTC or foreign listed names are excluded from the table. If you want the complete CSV list, just drop me a message and I’ll send it over. Also a few names are still private with upcoming IPOs, such as Lambda (US), Nscale (UK), Firmus (Australia) and Yotta (India).

Important Note

One has to realize that a logo display tells you a company is involved but it doesn’t tell you whether the involvement is material. For CoreWeave or Vertiv, AI-factory demand is essentially the entire story. For Caterpillar or National Grid, it’s a sliver of a far bigger business that will barely move the stock. The “High” rows give you torque and volatility in equal measure. The “Low” rowsgive you a steadier company with only a thin thread tied to AI built-out trade.

Final Thoughts

Maybe one of these names becomes the next Marvell, maybe none do. That’s not a call I can make from a slide, and chasing whichever logo you hope Jensen blesses next is closer to a guessing game than a strategy.

The durable value here is the map, plus a sharper question to take into it. For any name on this chart, how much of its business actually rides on the AI build-out?How much pricing power does its layer hold? Pure-plays, diversified incumbents and commodity definitely have different leverages and risk profiles.

Here’s what doesn’t change: Every hyperscaler deal you’ll read about, every “X-gigawatt data center” headline, quietly depends on this entire stack to happen. Someone designs it, someone builds it, someone powers it, someone cools it, someone racks the servers, someone runs it. This chart is the cast list. Pick a layer that interests you and weigh its exposure against how much pricing power it holds. That’s where the real work starts. The map won’t tell you what to buy, but it’s a framework you can refer to.