NVDA quietly became a CPU company 🤫

Hi {{first_name|Investor}}

This is Part 1 of a three-part series taking apart NVIDIA's earnings call from May 20, 2026. The headlines all went to the same place: record revenue, Blackwell everywhere, the trillion-dollar pipeline. That story is real, and it is also the story everyone already owns.

I want to start somewhere quieter. On the call, Jensen Huang kept returning to one product, naming it four separate times, and most of the coverage treated it as a footnote.

It is called Vera, NVIDIA's ($NVDA) first CPU built specifically for agentic AI. For years the company has been the GPU business, the chip that does the thinking. Vera is the other half of the machine, the processor that runs everything around the thinking: the orchestration, the memory, the tool calls, the actual work an agent does when it browses, runs code, or queries a database.

The economics of computing are shifting, and that is what makes Vera matter now rather than someday. The cloud era was priced in dollars per core, because people rented cores. The AI era is priced in tokens per dollar, because what gets sold is intelligence, measured in output. Agents do not rent cores. They want work done fast and cheap. Vera was designed for exactly that world, and Jensen framed it as a market NVIDIA has never addressed before: a $200 billion TAM, with visibility to roughly $20 billion in standalone CPU revenue this year alone.

A company that already prints record GPU numbers just told you it is walking into a brand-new $200 billion market next door, and the room mostly shrugged.

That gap between what gets said and what gets covered is usually where the research pays off.

The reason Vera opens a market this large is that it shows up in four different roles, and only one of them sits inside the trillion-dollar pipeline analysts already model.

The first role is the familiar one: Vera as the host processor in Vera Rubin racks, paired two-to-one with Rubin GPUs. That piece is already counted. The other three are additive on top. Vera standalone runs agentic inference and orchestration, where Jensen says it delivers up to 1.5x the performance per core, 2x the performance per watt, and 4x the density per rack of an x86 chip. Vera paired with the CX9 networking stack becomes the control plane for how data moves through storage. And Vera again, with a confidential-computing stack, becomes the secure boundary for regulated AI in finance, healthcare, and government.

When an analyst asked what sits above the $1 trillion Blackwell-and-Rubin number, Jensen named standalone Vera as the second largest source of upside, behind only continued share gains in frontier AI models. That is management telling you where it expects the next leg to come from.

That covers what Vera does. What it takes to actually build it is the harder question, and the answer starts with one name.

One line is easy to skim past: Jensen expects Vera to be supply constrained throughout its entire life. Read as a warning, that sounds like a risk. Read as a map, it tells you demand is not the question. The inputs are, and a Vera chip needs a short list of them to exist at all.

Start with the design. Vera runs on custom ARM cores, which Jensen confirmed directly on the call. ARM Holdings ($ARM) owns the CPU architecture NVIDIA licenses to build those cores, and the choice is deliberate: ARM's efficiency is a big part of why Vera can claim up to 2x the performance per watt of an x86 chip. ARM never touches a wafer. It earns a royalty on every chip that ships using its architecture, on top of the upfront license, so with NVIDIA forecasting millions of Vera units that royalty stream scales one-for-one with Vera's volume while ARM carries almost none of the manufacturing cost.

ARM is the cleanest single expression of the Vera trade, and probably the one you would land on yourself. The names that take real digging sit further down the chain: the software Vera is designed in, the one foundry that can actually build it, and the memory that has to scale with every unit.

The rest of this issue is for Insider members: the suppliers behind ARM and the rest of the Vera Rubin rack, why each one is critical, and how I'm positioning around it. Upgrade to keep reading. Insider also gets you the full archive and every week's names.