As I write this, a little ahead of schedule, Jensen Huang is on a stage in Taipei with his arms stretched wide. Behind him is a wall of server racks spanning the full width of the screen.

He looks tiny against it.

Because he is tiny against it.

What you’re looking at is not really a collection of servers. It’s a visual representation of the scale of the AI buildout now underway. Rack after rack filled with Nvidia (Nasdaq: NVDA) GPUs, CPUs, networking gear, switches, storage, and the infrastructure needed to make it all work.

In other words, you’re looking at AI.

When most of us picture AI, we picture something small.

A little icon on a phone. A blinking cursor. A sentence that appears a couple of seconds after you ask a question.

It feels weightless, as if the answer was already sitting there and the machine just went and fetched it.

That is not what happens in reality.

To produce that one sentence, a building full of hardware powers up, pulls enough wattage to run a small town, and coordinates millions of components to hand you a few hundred words.

The AI in your hand is the size of a coin. The AI behind the scenes is the size of a small city.

The image above, while large compared to Huang, is a tiny part of Nvidia’s AI factories. As Huang describes it himself, Nvidia is now an AI infrastructure company. And that infrastructure has rows upon rows of racks filled with the gear in the image behind him.

I’m of the view that most people don’t comprehend the insane scale of what AI is in its true physical form.

But once you understand the physical scale of what AI really is, you begin to see just how much opportunity there is to own a part of it all.

What one answer actually costs

“Every company will be an agent company. Every company will have agents running inside .”

These are not my words; these are Huang’s words on the stage at GTC on Monday.

He also said, “Compute is revenue now. Compute is profit.”

 That perhaps sums up the entire thesis behind the AI buildout, or as I call it, the Intelligence Revolution.

Of course, the big deal at this event was the release of several new revenue opportunities for Nvidia. The Vera Rubin platform is the granddaddy of them all.

And it’s not one chip. It’s seven different chip types, co-designed to behave as a single machine, spread across five kinds of rack tray. A single Vera Rubin NVL72 rack ties together 36 Vera CPUs and 72 Rubin GPUs, runs fully liquid-cooled, and reportedly sells for around US$8.8 million on its own.

Now scale it up. A full Vera Rubin pod is 40 of those racks bolted together. That works out to 1,152 GPUs, roughly 1.2 quadrillion transistors across close to 20,000 individual silicon dies, all behaving as one computer.

That’s what a “chip” looks like today.

Huang has described the broader launch as involving nearly two million parts and around 150 ecosystem partners.

He name-dropped and announced partnerships with everyone from Cadence (Nasdaq: CDNS), to CoreWeave (Nasdaq: CRWV), Nebius (Nasdaq: NBIS), Microsoft (Nasdaq: MSFT), and many more. It was a long list, and the activity in those stocks this week has shown exactly why Nvidia runs this show.

In fact, a day later Huang went on stage with the CEO of Marvell Technology (Nasdaq: MRVL) and said they would be “the next trillion-dollar company”.

The stock jumped 32% during trading and another 8% in after-hours trading.

In simple terms, Nvidia moves markets.

And not just its own stock price. It is now capable of adding tens of billions of dollars in market value to other companies with little more than a mention during a GTC keynote.

When Huang talks about the shift towards more CPOs (co-packaged optics), investors listen. The entire optics sector responded, with names such as Lumentum (Nasdaq: LITE) and Coherent (Nasdaq: COHR) jumping double digits.

That is the level of influence Nvidia now has.

It doesn’t just participate in the AI buildout. It effectively sets the direction of travel for the entire ecosystem around it.

The scale is the runway

When you appreciate that these Nvidia “chips” are actually vast arrays of different hardware and technologies working together, you begin to see why this opportunity has such a long runway.

Just above, I’ve named seven companies that shot higher following GTC Taipei on Monday. There were another dozen that ripped higher as well.

This is what markets do as they begin to understand the significance and scale of what’s happening.

You cannot conjure two million parts overnight.

You cannot wish 20,000 silicon dies into existence, or pour the foundations for a gigawatt-scale data centre and have it humming next quarter.

Every layer has a lead time measured in years.

Huang also unveiled new laptops and PCs that Nvidia has been developing with Microsoft for the better part of three years. Consumer AI chips arriving today that were being designed years ago.

So bear in mind that Nvidia is developing technology today that may not reach the market for another two, three, or even five years. And when it does, it will likely be even more important to how the world works, and the value it creates will be even greater.

That is the runway.

When something this physically enormous is being built, the process cannot stop and start on a whim. Huang believes orders for the current and next generation of these systems will top a trillion dollars through 2027. Even allowing for the fact that he’s talking up his own company, the sheer scale of it suggests this story runs for years, perhaps decades, not quarters.

When you break the machine into its component parts and look at who makes them, that’s where the opportunity lies.

The memory makers. The foundries. The packaging specialists. The power and cooling suppliers. The optical and networking firms. The materials companies feeding all of them.

Every one of those 150 partners gets paid as long as the racks keep going up.

And the racks will keep going up.

None of this is risk-free, and I’d be doing you a disservice to pretend otherwise.

The customer base for these machines remains tiny, a handful of hyperscalers capable of writing cheques for eight-figure racks. But when Google is raising US$80 billion, including US$10 billion from Berkshire Hathaway, you get a sense that there is still plenty of capital flowing into this buildout.

So next time an answer appears a few seconds after you ask for it on your smartphone, picture the wall of racks Jensen Huang was standing in front of.

That is the actual machine.

For investors, the job is to stop focusing on how agreeable ChatGPT is and start counting the parts that make it possible.

There are close to two million of them in a single one of these systems, and somebody, somewhere, makes and sells every one.

That is a very long list of companies, and a very good place to start when looking for your next investment opportunity.

Until next time,

Sam Volkering
Investment Director, Southbank Investment Research

PS Two million parts. 150 partners. Trillions of dollars of expected spending. Most investors will focus on Nvidia. But like I just explained, I’m far more interested in the companies supplying the memory, power, cooling, networking and infrastructure that make it all possible. And James Altucher agrees.

Here’s where he is seeing the next wave of winners.