Primer on GPU-as-a-Service: The Infrastructure Layer For The Tokenomics Era

Research evidence excerpt

Primer on GPU-as-a-Service: The Infrastructure Layer For The Tokenomics Era

Technology | IT Hardware & Networking

July 06, 2026

Amit Daryanani Primer on GPU-as-a-Service: The

415-800-0185

amit.daryanani@evercoreisi.com Infrastructure Layer For The Tokenomics

Irvin Liu Era

415-800-0183 ALL YOU NEED TO KNOW: In this report, we examine the business

irvin.liu@evercoreisi.com

model, underlying demand drivers, supply chain backdrop, financing

Victor Santiago

requirements, and competitive landscape of the GPU cloud industry. 212-653-9014

Victor.Santiago@evercoreISI.com Fundamentally, we view GPU clouds as a critical layer of the broader AI

Hannah Liu infrastructure stack that is benefitting from robust demand for AI compute

212-812-2905 (for both model training and inference, measured by token consumption).

hannah.liu@evercoreISI.com The unrelenting demand backdrop for AI compute combined with

Caden Dahl constrained supply (data center powered shell and GPUs/XPUs) should

212-653-9034 lead to multi-year AI infrastructure demand and pricing tailwinds as well.

Caden.Dahl@evercoreisi.com Given that the GPU cloud industry is somewhat nascent, providers are still

in investment mode which means capex as a percentage of total revenue

will remain high, operating margins will be depressed at the consolidated

level as GPU cloud providers absorb ramp-up costs related to new

capacity, and EPS/FCF will likely remain negative. Therefore, we view

backlog/RPO, EBITDA, and capacity (measured in MW, both active and

contracted/developable) as the more important near term metrics to focus

on. Despite limited profitability at the consolidated level near term, it is

important to note multi-year take or pay contracts are underwritten at