The Weekly Inference: Codex/ChatGPT Work Cross 8M+ WAUs, Frontier Labs Capture 95% of Spend on Vercel

Research evidence excerpt

The Weekly Inference: Codex/ChatGPT Work Cross 8M+ WAUs, Frontier Labs Capture 95% of Spend on Vercel

Truist Securities

Equity Research Report July 15, 2026

TMT: Artificial Intelligence, Digital

Platforms and IT Services The Weekly Inference: Codex/ChatGPT

Work Cross 8M+ WAUs, Frontier Labs

Arvind Ramnani Capture 95% of Spend on Vercel

214-969-5873

Arvind.Ramnani@truist.com Last week, we published The Quarterly Inference: 2Q26, in which we forecast

a $2T annual TAM for AI models and applications by 2030, with $1T enterprise

Wayne Trinh, CFA opportunity (70%+ via consumption). This week, we highlight Anthropic & OpenAI

212-590-0911 continuing to define the frontier despite a slew of new model releases, while Codex &

Wayne.Trinh@truist.com ChatGPT Work crossed 8M active users as OpenAI races to recapture enterprise share

from Anthropic. Vercel data shows open models capturing 29% of token volume, but

31 Page Document frontier labs capturing 95% of spend. The debate around enterprises owning theirown

models grows, AI capex trends, and Anthropic co-founder argues AI's advancing faster

than humans' relative advantages.

Reasons for this report

✓ Sector Update A slew of new model releases, but we continue to see the frontier defined by

the two leading labs, Anthropic and OpenAI. Within the last week, at least half a ✓ Sector Read-Throughs

dozen new models have been released, most notably OpenAI's (private) GPT-5.6 (Sol,

Terra, Luna) and GPT-Live models, SpaceXAI's (SPCX, NR) Grok 4.5, Meta's (META,

Buy, Squali) Muse Spark 1.1, Cognition's (private) SWE-1.7, and Mistral's (private)

Robostral Navigate. We believe the AI frontier continues to be defined by Anthropic and

OpenAI, with their latest models, Fable 5 and GPT-5.6, pushing capabilities forward.