U.S. Internet: Charts Worth a Look: Revisiting The Unit Economics Of AI Tokens

Research evidence excerpt

U.S. Internet: Charts Worth a Look: Revisiting The Unit Economics Of AI Tokens

Equity Research

5 June 2026

U.S. Internet

Charts Worth a Look: Revisiting

The Unit Economics Of AI Tokens

Frontier AI revenue per token is increasing while models are U.S. Internet

~25% more cost efficient with each new release. Hence AI labs POSITIVE

& hyperscalers are now seeing AI profit margins increase U.S. Internet

Ross Sandler

while capabilities also improve. Selling tokens is proving to +1 415 263 4470

be a profitable business in the age of Agentic AI. ross.sandler@barclays.com

BCI, US

Alex Hughes

We appreciate your 5-star vote in the 2026 Extel All-America Research Survey in the Internet

+1 212 526 3069

Large-Cap and Internet Mid- & Small-Cap categories (Search "Sandler Barclays"). View our alexander.hughes@barclays.com

analysts » Vote 5 Stars for Barclays » BCI, US

The Key Takeaway: The main story in AI thus far in 2026 has been AI lab ARRs surging on the Michael DiSanto

+1 212 526 1054back of rapid adoption of native AI applications, with huge funding rounds to satisfy growing

michael.disanto@barclays.com

compute needs (see AI's S-Curve Is Steepening from earlier this week). However, one important

dynamic that's been happening alongside all this growth is improved unit economics across all

layers of the AI stack. AI model innovation is delivering both gains in capability and cost Owen Clendenin

+1 212 526 7518efficiency, driving a sharp inflection in token economics, and hence expanding inference

owen.clendenin@barclays.com

margins at AI labs. While training models is still very expensive, AI inference is finally emerging

as a profitable business across the stack – from the hyperscalers to model providers.

Alexander Kessinger

+1 212 526 1324

alexander.kessinger@barclays.com