Jensen Huang wants to own the entire AI hardware stack, and he’s not being subtle about it. The Nvidia CEO is betting that the company’s new Vera CPU line will capture a massive share of what he projects to be a $200 billion total addressable market built around AI agents.

What Vera actually is

The Vera CPU packs 88 custom Olympus cores and delivers memory bandwidth of up to 1.2 TB/s. Dell’s PowerEdge XE9812 system, built on Vera architecture, reportedly delivers up to 10x lower cost-per-token for large-scale AI inferencing compared to previous generations.

OpenAI and Anthropic are already early adopters of Vera-based systems. Huang’s ambition extends well beyond CPUs. He’s aiming to manage the entire AI factory stack, including GPUs, networking hardware, and now these purpose-built processors. Nvidia projects its AI infrastructure revenue will exceed $1 trillion from 2025 to 2027.

Why crypto should pay attention

Decentralized compute networks like Akash, Render, and io.net have positioned themselves as alternatives to centralized cloud providers for AI workloads. Nvidia tightening its grip on the full hardware stack is a double-edged sword for these projects. On one hand, cheaper and more powerful chips could lower the barrier for node operators to offer competitive AI compute. On the other, Nvidia’s vertically integrated approach makes it harder for anyone, centralized or decentralized, to compete without using Nvidia hardware.

What this means for investors

AMD and Intel have been trying to claw back relevance in AI chips, but Nvidia’s strategy of controlling CPUs, GPUs, and networking simultaneously makes it harder for rivals to compete on any single component. The $1 trillion revenue projection from 2025 to 2027 suggests Huang expects most of the industry to operate within that stack.

Watch for how quickly Vera-based systems actually reach production scale and whether the 10x cost-per-token improvement holds up under real-world workloads.