Nearly 40 countries are now building on NVIDIA

Hi {{first_name|Investor}}

Part 2 of the NVIDIA earnings series. In Part 1, we looked at Vera, the CPU hiding inside a GPU company. This issue is about the customer hiding inside the numbers.

Almost every analyst on Tuesday's call wanted to talk about the same handful of names: the hyperscalers, the roughly $1 trillion of data center spending flowing their way this year. That is the consensus trade, and it is a good one. What stood out to me was how often Jensen Huang steered the conversation toward a completely different buyer.

For years NVIDIA ($NVDA) reported one giant Data Center line and let everyone guess what was inside. This quarter it split the business into two platforms, Data Center and Edge Computing, and then split Data Center again into Hyperscale and ACIE, which stands for AI Cloud, Industrial, and Enterprise. Then it restated nine quarters of history under the new structure and posted it publicly.

Companies do that when a line item gets big enough, and strategic enough, that they want you tracking it on its own. You do not rebuild nine quarters of disclosure for a rounding error.

The split reveals the surprise. Hyperscale came in at $38 billion, about half of data center revenue. ACIE came in right behind it at $37 billion, growing 31% quarter over quarter, with the AI cloud piece inside it more than tripling year over year. Sovereign AI alone grew more than 80% year over year, with NVIDIA infrastructure now deployed across nearly 40 countries that represent $50 trillion of GDP.

Then an analyst asked whether this second category would eventually grow larger than the hyperscalers. Jensen called it a foregone conclusion.

What makes Jensen so certain is who these customers are. Hyperscalers design their own chips and stitch together their own systems, so NVIDIA competes for their business piece by piece. The second category cannot do that. AI-native clouds, enterprises, industrial operators, and governments want to buy a working AI factory and run it, not build one, which is exactly where NVIDIA's full-stack platform turns into a moat. Semi-custom silicon does not apply to a buyer who wants the whole thing to work out of the box. And the base is enormous: where the first category is five or six companies, the second is a vast long tail of buyers, and NVIDIA's customer count there is climbing from hundreds today toward hundreds of thousands.

The part that matters for your portfolio is what this does to risk. The hyperscaler trade is crowded, every analyst already models it, and it rises and falls on the capital-expenditure plans of a handful of giants. The second category runs on its own clock, governments and enterprises building AI for reasons that have little to do with the big four's spending. That makes it the more durable and less crowded end of the same AI story, which is the kind of exposure a portfolio built for work-optional wants.

There is one catch. The purest beneficiary of all this is NVIDIA itself, and you probably already own it. The more direct, smaller-cap exposure sits in a few specific layers around it, and that is the part worth digging into.

The rest of this issue is for Insider members: the three layers you can actually own around NVIDIA, the specific names in each, and how I'm positioning. Upgrade to keep reading. Insider also gets you the full archive and every week's names.