Token Demand Surges Beyond Expectations, Driving Compute and Networking Growth

The demand for token processing has far exceeded initial expectations, according to industry analysts. This surge in inference demand is driving growth in various sectors, including compute, networking, and power companies.

Nvidia's Vera Rubin racks are designed to deliver dramatically higher token throughput per GW, with up to 10X more agentic throughput per unit of energy compared to Blackwell. The company also expects its Vera Rubin + Groq 3 LPX deployments to deliver up to 35X higher token throughput per MW.

As data centers move towards Nvidia's more advanced systems, margin benefits are expected to increase significantly. According to Morgan Stanley Research, Blackwell-based data centers generate approximately 58% net margins, while Rubin-based data centers approach 78%, and Feynman-based data centers reach roughly 90%.

The agentic AI shift is also driving up networking demand, with estimates suggesting a 15X increase in tokens per user. OpenClaw consumed 603 billion tokens from 7.6 million API calls in one month for an estimated $1.3 million in spend. Cisco estimates that performing tasks using AI agents increases WAN traffic by 450% compared to humans.

Optical networking is emerging as a critical AI infrastructure layer, with Lumentum's sales rising by 90.1% YOY to $808.4 million in its latest quarter. The company expects co-packaged optics (CPO) and near-packaged optics (NPO) market growth from $100 million in 2025 to $39 billion by 2030, a staggering 230% CAGR.