Particle.news
Download on the App Store

Microsoft Launches In‑House AI Models at Build 2026 to Shift Off Third‑Party LLMs

The company intends to reduce reliance on external models by scaling proprietary models and agent products while heavy data‑center and GPU investment tests whether that spending will produce lasting profits.

Overview

  • At Build 2026, Microsoft unveiled seven proprietary models including MAI‑Thinking‑1 and MAI‑Code‑1 and introduced Microsoft Scout, signaling a public move away from reliance on outside LLM suppliers.
  • The product push follows a strong fiscal Q3 showing in which Microsoft said AI annual recurring revenue grew 123% to more than $37 billion and quarterly EPS beat expectations.
  • Microsoft disclosed large infrastructure plans, routing GPU capacity into internal R&D and signaling roughly $190 billion in calendar‑year data‑center CapEx to support model training and cloud services.
  • Wall Street reacted positively with many analysts raising price targets and maintaining Buy ratings while some executives sold shares recently, prompting investor attention to timing and governance questions.
  • Microsoft also announced Majorana 2 topological quantum‑chip progress and a target for scalable quantum systems by 2029, reflecting a parallel long‑term bet on new compute architectures.