IT Hardware/Networking & Data Centers: Introducing The Token Consumption Model +

研报英文原文证据摘录

IT Hardware/Networking & Data Centers: Introducing The Token Consumption Model +

Technology | Communications Infrastructure

May 19, 2026

Irvin Liu IT Hardware/Networking & Data Centers:

415-800-0183

irvin.liu@evercoreisi.com Introducing The Token Consumption Model

Amit Daryanani + Implications for DC Demand & Capex

415-800-0185

ALL YOU NEED TO KNOW: Post the Q1 EPS cycle, companies across amit.daryanani@evercoreisi.com

the broader AI ecosystem (hyperscalers, frontier model builders,

Hannah Liu

neoclouds, AI infrastructure suppliers, data center operators, etc.) have 212-812-2905

hannah.liu@evercoreISI.com made it clear that demand for AI compute capacity is only accelerating.

Victor Santiago Our latest Hyperscale Capex Model now forecasts ~$800B in

212-653-9014 aggregate capex for CY26 (among key hyperscalers), up +83% y/y and

Victor.Santiago@evercoreISI.com growing toward ~$1T by next year, creating tailwinds our IT

Caden Dahl Hardware/Networking & Data Centers coverage. While this top-down

212-653-9034 view provides a directional overview of the broader AI infrastructure

Caden.Dahl@evercoreisi.com spending backdrop, we think it is also important to examine the

bottoms-up underlying drivers of AI infrastructure capacity demand at a

unit level to better understand the durability of the ongoing

infrastructure buildout. We are therefore introducing our Token

Consumption Model to better frame the case for durable AI

infrastructure demand which features three scenarios: 1) Base Case:

In our base case, we see total annual token demand growing from ~100

quadrillion in 2026 to ~4 quintillion in 2030, representing a ~150% 4-

year CAGR. Key assumptions here include ~1.5B AI consumer users

growing to ~3.7B by 2030 and ~50M AI agents growing to ~800M by

2030.