On Nvidia’s earnings call on May 20, 2026, Jensen Huang dropped a number nobody was modeling: Vera, the CPU built for agentic AI, opens a brand-new $200 billion market for the company. “Vera opens a brand new $200 billion TAM for Nvidia, a market we have never addressed before,” he said — and standalone Vera CPU sales have already reached $20 billion this year. Vera was only introduced at GTC in March. That ramp means it is no longer a slide in a roadmap; it is a shipping product with real bookings behind it.
The context is that the front line of the AI chip war is moving. GPUs do the model’s “thinking,” but agents execute their tasks on CPUs — and when the agent population starts numbering in the billions, the CPU, historically Intel’s and AMD’s territory, suddenly becomes AI’s newest battlefield.
The Earnings-Call Claim
Nvidia calls Vera “the world’s first CPU, purpose-built for agentic AI.” It sells standalone and bundled with the Rubin GPU. The call itself was another record: $81.6 billion in quarterly revenue, $91 billion forecast for the next quarter. But Huang deliberately steered the conversation from GPUs to CPUs: “The world is rebuilding computing for agentic AI and robotic physical AI,” and “we’re going to need a lot more CPUs.”
He made the market math explicit. Today roughly a billion humans use computing; agents will eventually number in the billions, each running what he predicts will be its own PC-like setup. The ratio between human users and agents is the ceiling on CPU demand. On distribution, his claim was that “every major hyperscaler and system maker is partnering with us to deploy it.”
Why Agents Need a Different CPU
Huang’s argument rests on a difference in workloads. Traditional cloud CPUs are designed around cores: pack many application instances onto one machine, and sell density and utilization. Vera is designed around token throughput — every tool call an agent makes, every chunk of context it reads, every decision it takes is a flow of tokens. The bottleneck is not how many processes run concurrently; it is how fast tokens move.
That is the substance behind “purpose-built”: if the agent’s unit of work is a token rather than a process, a general-purpose CPU is no longer the optimal part. For Nvidia the framing does double strategic duty. It repositions the company from “GPU vendor” to full-stack supplier of AI computing — CPUs, GPUs, and networking sold as one package — and a standalone Vera SKU turns customers who never bought an Nvidia GPU into direct Nvidia customers for the first time.
Enemies on the CPU Battlefield
CPUs are Intel and AMD’s home turf, and the threats don’t stop with the incumbents. In April 2026, Amazon signed Meta to a contract for millions of its homegrown AI CPUs — a deal TechCrunch called “another wild turn for AI chips.” The same month, Amazon CEO Andy Jassy used his annual shareholder letter to take aim at Nvidia directly, suggesting AWS can build AI chips as well as, and possibly better than, the incumbent. Wall Street has started to price in scenarios where Nvidia’s dominance erodes.
Huang’s answer is to widen the war. Rather than defend the GPU, take the fight to the CPU incumbents’ home market and make “who supplies the AI CPU?” a live procurement question. TechCrunch’s verdict captures the dynamic — Huang may be the industry’s most relentless hype man, but he “delivers on the hype, quarter after quarter.” A $200 billion TAM is marketing language. Twenty billion dollars of standalone sales is not.
What It Means for Developers and Infrastructure Teams
Three practical effects. First, agent cost structures will shift: if agent execution moves from borrowed general-purpose CPUs to purpose-built ones, per-task inference cost and latency have room to fall, and agent product teams should put CPU SKUs and pricing into their cost models instead of staring at GPU unit prices alone. Second, CPU sourcing becomes a new variable in cloud selection — when hyperscalers sell their own silicon while also deploying Vera, cross-cloud and hybrid agent deployment strategies deserve a fresh look at lock-in risk. Third, the supply chain will reorganize around compute purpose-built for agents: the ecosystem spent the last three years rebuilding around GPUs, and CPU-to-GPU ratios, memory bandwidth, rack design, and procurement cycles will now re-sort around agent workloads. For hardware planners the question is no longer “do we need GPUs?” but “what ratio of GPU to CPU do we need?”
Sources
- Jensen Huang says he’s found a ‘brand new’ $200B market for Nvidia — TechCrunch
- NVIDIA launches Vera CPU, purpose-built for agentic AI — NVIDIA
- Meta signs deal for millions of Amazon AI CPUs — TechCrunch
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.