Nvidia just posted $81.6 billion in a single quarter of revenue — and its CEO is already pointing at a market the company has never touched before. That’s either the most confident thing you’ll hear in tech this year, or the most audacious. Possibly both.
Jensen Huang, Nvidia’s founder and CEO, used the company’s latest earnings call to unveil what he calls “a brand new $200 billion TAM” — that’s Total Addressable Market, the total potential revenue a company could chase if it captured every customer in a given space. His claim? Nvidia’s new Vera CPU has unlocked it. To understand why that matters, we need to back up a little.
What Is It?
You’ve almost certainly heard of the GPU (Graphics Processing Unit) — the chip that originally powered video games but turned out to be extraordinarily good at training AI models. Nvidia dominates that market. What’s less talked about is the CPU (Central Processing Unit) — the chip that handles general computing tasks, the “brain” coordinating everything a computer does. Intel and AMD have historically owned that space.
Nvidia has dipped into CPUs before, but it was never the main event. That changes, Huang says, with Vera — a CPU Nvidia introduced in March 2026 and describes as “the world’s first CPU purpose-built for agentic AI.”
So what is agentic AI? Think of traditional AI as a very smart answering machine — you ask it something, it responds. Agentic AI goes further: it takes goals, breaks them into tasks, and executes them semi-autonomously, using tools and making decisions along the way. An AI agent might book your travel, monitor your inbox, coordinate with other agents, and flag anything that needs your attention — all without you prompting each step. It’s AI that acts, not just responds. You can dig deeper into how these systems relate to each other in this overview of the subsets of AI.
Vera is the chip Nvidia is betting these agents will run on. It’s also sold bundled with Nvidia’s Rubin GPU, so customers can pair the “thinking” power of a GPU with the “doing” power of Vera in one package.
How Does It Work?
Here’s where the technical distinction gets genuinely interesting. Traditional CPUs are built around cores — think of them as workers who are really good at juggling multiple jobs simultaneously. A server CPU might have dozens of cores, each capable of running separate applications at the same time. That’s perfect for classic cloud computing, where you need to serve millions of users doing different things at once.
Vera is designed differently. Instead of optimizing for parallel multitasking, it’s optimized for processing tokens as fast as possible. A token is the basic unit of text (or data) that AI models work with — roughly a word or part of a word. When an AI agent is executing a task, it’s constantly reading, generating, and acting on tokens. Speed here translates directly to how fast an agent can think and move.
Think of it like the difference between a busy restaurant kitchen and a Formula 1 pit crew. A traditional CPU is the kitchen — lots of chefs, all cooking different dishes simultaneously. Vera is the pit crew — every single person laser-focused on one car, executing one sequence of tasks as fast as physically possible. Different problem, different tool.
Huang’s argument is that as the AI infrastructure race intensifies, the world will need billions of agents, each running on chips like Vera. “The world has a billion users, human users,” he said on the earnings call. “My sense is that the world is going to have billions of agents… and those billions of agents will all use tools. And those tools are going to be like PCs.”
Why Does It Matter?
The stakes here extend well beyond Nvidia’s balance sheet. If Huang is right that agentic AI will become as ubiquitous as personal computers, then the chip powering those agents becomes one of the most strategically important pieces of hardware on the planet.
That’s not a given. Amazon Web Services recently signed a major contract with Meta for millions of Amazon’s own homegrown AI CPUs. AWS CEO Andy Jassy has been explicit that he believes his company can build AI chips — both GPUs and CPUs — to match or beat Nvidia. Startups are pursuing the same prize. The competitive pressure is real.
But Huang’s counter-argument is already backed by numbers: he reports that Nvidia has already sold $20 billion worth of standalone Vera CPUs in 2026 alone, with every major hyperscaler (the term for massive cloud providers like AWS, Google Cloud, and Microsoft Azure) reportedly partnering to deploy it. That’s a meaningful head start.
For everyday users, the implications are more subtle but significant. Faster, cheaper, more efficient agentic AI chips mean AI agents become more capable and more affordable to run at scale — which accelerates their deployment across industries. We’re already seeing early versions of this in hiring, healthcare, and beyond. AI agents are beginning to replace human recruiters, and deep learning is automating complex medical classification tasks — both trends that depend on exactly the kind of compute Vera is designed to provide.
More compute capacity for agents also raises harder questions about economic disruption and who benefits from these transitions — concerns that aren’t going away anytime soon.
Common Misconceptions
1. “GPUs are being replaced by CPUs”
Not quite. Huang’s own framing is clear: GPUs handle the “thinking” — the heavy model training and inference — while CPUs like Vera handle the “doing,” the task execution once an agent is deployed. These chips are complementary, not competing. That’s why Vera is sold both standalone and bundled with the Rubin GPU.
2. “A $200 billion market claim means Nvidia will capture $200 billion”
TAM is the total theoretical size of a market, not a revenue forecast. It’s the ceiling, not the floor. Nvidia will compete with Amazon, Google, AMD, Intel, and a wave of AI chip startups for a slice of that opportunity. The number signals the scale of the opportunity, not a guaranteed outcome.
3. “Agentic AI is just a buzzword for chatbots”
This one’s worth pushing back on firmly. Chatbots respond to prompts. Agents initiate, plan, and execute multi-step tasks — often interacting with external tools, APIs, and other agents — with minimal human intervention. It’s a meaningful architectural shift, not a rebranding exercise, and it changes what hardware you actually need to run AI workloads efficiently.
Where to Learn More
- Nvidia’s Investor Relations pages (investor.nvidia.com) publish full earnings call transcripts where Huang explains Vera and Rubin in detail.
- Anthropic’s and OpenAI’s research blogs have accessible explainers on agentic AI architecture and how agents differ from standard language models.
- The Chip Letter (newsletter) covers semiconductor industry dynamics, including the CPU/GPU competitive landscape, in reader-friendly depth.
- What experts say you should know about AI is a solid grounding piece if you want broader context before going deep on the hardware side.
- IEEE Spectrum regularly publishes technical but accessible explainers on chip architecture for readers who want to understand the engineering underneath the hype.
Key Takeaways
- Nvidia’s Vera CPU is designed specifically for agentic AI — AI that executes tasks autonomously — rather than general-purpose cloud computing, marking a genuine architectural shift.
- Huang claims Vera opens a $200 billion Total Addressable Market that Nvidia has never competed in before, backed by a reported $20 billion in Vera CPU sales already in 2026.
- The key technical difference: Vera is optimized for processing AI tokens at maximum speed, while traditional CPUs prioritise running many tasks simultaneously across multiple cores.
- Major cloud providers including Amazon are building their own competing AI CPUs, so Nvidia’s dominance in this new market is contested, not guaranteed.
- If billions of AI agents do become as common as personal computers — as Huang predicts — the chip powering them becomes one of the most strategically important pieces of hardware of the next decade.
