Nvidia CEO Jensen Huang is arguably the best salesman in Silicon Valley. Maybe ever. He’s got that relentless optimism, the kind Marc Benioff wears like a badge of honor, but here is the thing that makes it dangerous: Jensen actually delivers. Quarter after quarter, the numbers back the hype. So when he claims there is a brand new $200 billion total addressable market (TAM) sitting right under our noses, you don’t roll your eyes. You take notes.
He dropped this bomb on the latest earnings call. Nvidia just posted another record-smashing quarter, raking in $81.6 billion and forecasting $91 billion for the upcoming period. Good timing, right? Huang pitched the Vera CPU not as an upgrade, but as a paradigm shift. He calls it transformative. The early sales figures say otherwise isn’t the case—they’re promising.
“Vera opens a brand new $80 billion TAM for Nvidia… The world is rebuilding computing for agentic AI.”
Jensen Huang
Wall Street is jittery, though. Everyone wants to know what’s next on the chopping block. Nvidia rules the GPU world, sure. But the CPU? That’s Intel and AMD’s backyard. Or at least, it used to be. Then came the hyperscalers. Last month, Amazon Web Services flexed a massive deal with Meta for its homegrown AI chips. Andy Jassy has been loud and clear: AWS can build chips at least as good as Nvidia’s, maybe better. It’s a real threat.
Then Vera showed up.
Sold alone. Bundled with the Rubin GPU. Huang insists this isn’t just another processor. It’s the world’s first CPU purpose-built for agentic AI. He believes every major system maker and hyperscaler is already lining up to deploy it.
Here is the logic, or what passes for it. When an AI model is thinking, heavy lifting, the GPUs are screaming. But agents? They run on CPUs. They do tasks. They execute. Huang predicts agents will eventually have their own CPU-driven equivalents of personal computers.
Vera handles this by prioritizing one thing: processing tokens as fast as humanly possible. Standard cloud CPUs? They’re designed for cores. For running multiple app instances efficiently. Vera doesn’t care about that. It cares about speed. Tokens. That distinction sounds logical on a slide deck, but the landscape is crowded. Startups and giants alike are pouring billions into their own silicon. Why would anyone trust Nvidia to lead this new charge?
Huang’s answer is simple.
Money talks. He says Nvidia has already sold $20 billion worth of standalone Vera CPUs. And it is only March.
“The world has a billion human users,” Huang noted, drawing the comparison. “My sense is the world will have billions of agents. We will have billions of agents using tools, and those tools are going to be like the PCs we use today.”
So we need more CPUs. Not just any CPUs. The right ones. Huang seems certain Vera is that chip. The $20 billion figure is hard to argue with, yet the question remains: does having a billion agents really mean buying a billion Vidas? Or is this just the start of a new, incredibly expensive hardware cycle where the rules haven’t fully settled yet.
Only time tells if agents really will want their own computers. Right now, they just want to be told they need one.
