Particle.news

Download on the App Store

AI Scale-Up Quickens as OpenAI Hits $1 Billion July and Companies Roll Out Agents

New data show AI entering mainstream business use in Mexico, signaling a turn from pilots to production.

Escultura de 2010 del artista italiano Maurizio Cattelan, titulada «L.O.V.E» («Il Dito»), aparece frente al Palazzo Mezzanotte, el edificio de la bolsa de valores de Italia, el 6 de junio de 2023 en Milán. El nombre L.O.V.E. es el acrónimo de «Libertà, Odio, Vendetta, Eternità» («Libertad, Odio, Venganza, Eternidad»). 
GABRIEL BOUYS / AFP
Las herramientas de IA podrían facilitarles el trabajo a los estafadores
Tras la actualización de ChatGPT a GPT-5, miles de usuarios reportan que los chatbots han perdido la personalidad empática con la que habían formado lazos emocionales, sintiéndose ahora más fríos y distantes/Foto: Canva
Image

Overview

  • OpenAI logged roughly $1 billion in July revenue and projects to top $12.7 billion this year, driven by GPT‑5 usage, 700 million weekly active users, and plans for heavy infrastructure spending alongside a reported $6 billion secondary share sale at a $300 billion valuation.
  • An AWS study reports 495,000 Mexican firms adopted AI in the past 12 months, lifting national adoption to 38%, with most use focused on efficiency gains but clear bottlenecks in skills, costs and governance.
  • Agentic systems are moving into operations: Dow uses an autonomous billing‑accuracy agent, Fujitsu reports a 67% sales‑productivity lift, and Grupo Bimbo has built thousands of apps and hundreds of agents, though analysts warn high cancellation rates for poorly scoped projects.
  • Safety risks are surfacing as Google acknowledged AI Overview search results were exploited to show a scam phone number, and researchers showed an agentic browser could complete phishing actions and transactions on fake sites before fixes and safeguards were applied.
  • Google published per‑query resource figures for Gemini—about 0.24 Wh of electricity and 0.26 ml of water—highlighting environmental costs at scale, with training workloads consuming far more than inference.