TOOL UPDATES

Microsoft Built Its Own AI to Replace OpenAI Inside GitHub Copilot

R Ryan Matsuda Jun 4, 2026 2 min read
Engine Score 8/10 — Important

Microsoft's decision to replace OpenAI's model in GitHub Copilot with its own homegrown AI, Project Polaris, marks a significant strategic pivot. This move signals Microsoft's increasing independence in AI development and profoundly impacts the competitive landscape for AI models and cloud services.

Editorial illustration for: Microsoft Built Its Own AI to Replace OpenAI Inside GitHub Copilot

Microsoft CEO Satya Nadella unveiled Project Polaris at Build 2026 (June 2-3) — a homegrown AI coding model that will replace GPT-4 Turbo as the default for all GitHub Copilot subscribers starting August 2026. Polaris runs on Microsoft’s custom Maia 200 accelerators inside Azure, bypassing OpenAI entirely.

The move is the clearest signal yet that Microsoft is building post-OpenAI infrastructure, even as the two companies remain commercial partners.

What Project Polaris replaces, and when

Polaris becomes the Copilot default in August 2026, with automatic migration and a three-month fallback period for users who want to stay on the prior model. The default-swap matters: most Copilot users never change model settings, so the default is the product.

Element Before After (Aug 2026)
Copilot default model GPT-4 Turbo (OpenAI) Project Polaris (Microsoft)
Inference hardware OpenAI/Azure stack Maia 200 accelerators
Dependency on OpenAI Core Bypassed for Copilot

The Maia 200 chip angle

Polaris runs on Microsoft’s Maia 200 custom AI accelerators. By owning both the model and the silicon, Microsoft removes two external dependencies at once — the model vendor and, partially, the merchant-GPU supply chain. Vertical integration is the strategy every hyperscaler is now pursuing.

MAI-Thinking-1 targets the reasoning tier

Microsoft also announced MAI-Thinking-1, its flagship reasoning model, which the company says “matches leading models on key software engineering benchmarks and reaches human preference parity with Sonnet 4.6.” Claiming parity with Anthropic’s Sonnet line is a direct shot at a specific competitor, not a vague benchmark boast.

The pricing pressure is mounting from multiple directions — Anthropic just shipped a 3x-cheaper Fast Mode in Claude Opus 4.8, and Microsoft’s AI chief has argued that rival frontier models are too expensive.

A .69 billion DoD contract

Microsoft also disclosed a $9.69 billion Department of Defense contract — its largest government deal ever — covering Microsoft 365, Azure, and Copilot for 2.1 million service members. Embedding Copilot across the DoD gives Polaris an immediate, captive deployment at national scale.

Why this is the divorce signal

Microsoft and OpenAI still have a deep partnership, but Polaris changes the dependency math. When the largest distributor of OpenAI’s technology builds an in-house replacement for its flagship developer product, it is hedging against the day the partnership terms change.

For developers, the practical question arrives in August: test Polaris against your current Copilot workflows during the three-month fallback, because the default will switch whether or not you opt in.

Related Reading

Share

Enjoyed this story?

Get articles like this delivered daily. The Engine Room — free AI intelligence newsletter.

Join 500+ AI professionals · No spam · Unsubscribe anytime