Jensen Huang says he's found a 'brand new' $200B market for Nvidia

Ronni Holmvig Strøm · 2026-05-21

Angle: Nvidia CEO Jensen Huang is pitching a new $200 billion opportunity: CPUs built for AI agents. The TechCrunch piece frames this as the company's next big market beyond its core AI accelerators, suggesting Nvidia sees agentic systems driving demand for new kinds of compute. For readers

Jensen Huang says he's found a 'brand new' $200B market for Nvidia

On Wednesday's earnings call, after Nvidia posted $81.6 billion in revenue and guided to $91 billion for the next quarter, Jensen Huang told analysts he had found "a brand new $200 billion TAM" for the company. The product carrying that number is Vera, Nvidia's first standalone CPU, which Huang called "the world's first CPU, purpose-built for agentic AI." Huang said Nvidia has already booked $20 billion of Vera sales this year.

The Workload Argument

The case for Vera rests on a workload claim about how agents actually run. Training and inference live on GPUs. The rest of an agent loop does not. Calling a tool, executing code in a sandbox, hitting an API and parsing the response, running an RL trajectory: this is serial, control-flow-heavy, latency-sensitive work that does not vectorize. Nvidia's engineers describe it as a textbook case of Amdahl's law, where the CPU-bound serial portion becomes the bottleneck as GPUs accelerate the parallel portion.

The chip follows the argument. Vera puts 88 custom Olympus cores on a monolithic die, with 1.2 TB/s of memory bandwidth, about three times the per-core bandwidth of a conventional data-center CPU. Nvidia claims 1.5x agentic sandbox performance against x86 competitors and 4x sandbox density per rack.

The Competitive Frame

The $200 billion claim lands in the part of Nvidia's business that Wall Street has been most anxious about. The GPU is Nvidia's. The CPU has historically belonged to Intel and AMD, and is now being contested by hyperscaler silicon: AWS's Graviton, Google's Axion, Meta's internal chips. Last month, AWS announced a contract with Meta for millions of its homegrown AI CPUs, and Andy Jassy has been publicly clear that he thinks AWS can build AI chips, GPUs and CPUs, at least as well as Nvidia, and possibly better.

Huang's $20 billion figure is, per his own framing on the call, standalone Vera CPU sales, not bundled with Rubin. If that number holds up as recurring revenue rather than first-wave placement, it would be the most aggressive standalone server-CPU result Nvidia has ever printed. Whether it does is the line item to watch, because standalone CPU is precisely where hyperscalers have shown the clearest appetite to in-source on their own silicon.

What To Watch

The TAM figure is the kind of number earnings calls produce, and we would not anchor on it. The more durable signal is the posture. Nvidia is no longer treating the CPU as something it tolerates next to the GPU. It is positioning Vera as a second front, with a real workload argument and a coherent piece of silicon behind it. Whether the agentic CPU market converges on Vera or on the next generation of Graviton is the question the next two earnings cycles will start to answer.