Since this outlet began tracking the shifting dynamics between Microsoft and OpenAI — including the White House equity stake talks, Reid Hoffman's exit from Microsoft's board, and the broader corporate AI cost revolt — the relationship between the two companies has quietly been restructuring at the foundation level.

Now Mustafa Suleyman has said it out loud.

Microsoft Wasn't Free to Compete — Until Recently

Suleyman, CEO of Microsoft AI, told VentureBeat in an exclusive interview at Microsoft Build 2026 that a contractual renegotiation with OpenAI — completed roughly six months ago — gave Microsoft's division formal permission to pursue superintelligence using its own researchers, its own data, and its own silicon.

"We were only sort of set free from our contract with OpenAI about six months ago to formally pursue superintelligence," Suleyman said. "So this is very early days."

The world's most valuable public company — which has poured more than $13 billion into OpenAI — was contractually restricted from building its own top-tier AI until late 2025.

Seven Models. Built In-House. No Distillation.

Microsoft announced seven new AI models at Build 2026 under the MAI brand, all developed by its internal AI Superintelligence Team. The flagship is MAI-Thinking-1 — a 35-billion-active-parameter reasoning model Microsoft says matches leading competitors in its weight class on software engineering benchmarks and demonstrates advanced mathematical reasoning.

Suleyman made a specific point of emphasizing what was NOT in these models: outputs from competitors.

"We train our reasoning models from scratch," Suleyman wrote in an accompanying blog post. "We don't distill from other labs and we don't rely on unlicensed or opaque data."

This is a direct shot at a widespread industry practice — training cheaper models on outputs from frontier systems like GPT-4 or Claude — without naming names.

The remaining six models fill out a multimodal enterprise portfolio covering code generation, image creation, transcription, and voice synthesis.

A Controlled Separation

Most tech media is framing this as Microsoft "diversifying" or "hedging" its AI bets. What's actually happening is a controlled separation — executed carefully so Microsoft doesn't blow up a $13 billion investment before it has a viable alternative. The MAI model launch is the first concrete proof Microsoft can execute on that alternative.

Suleyman also told Benzinga that Microsoft is now "less concerned" about Google, Meta, and OpenAI as primary competitors — pointing instead to Anthropic as the more relevant rival. OpenAI, the company Microsoft bankrolled into existence as a competitive force, is no longer the benchmark Microsoft is chasing.

The OpenAI Government Stake Adds Another Wrinkle

This independence push lands while the White House is actively exploring a government equity stake in OpenAI — a development the Trump administration confirmed this week, per prior coverage. If that deal materializes, OpenAI's governance and strategic priorities could shift in ways Microsoft has zero control over.

Microsoft betting on its own models is partly about product ambition and partly about risk management. The company cannot afford to have its entire AI future tethered to an organization that may soon have the U.S. government as a partial owner with its own agenda.

The $13 Billion Question

None of this means Microsoft is walking away from OpenAI's models. Copilot still runs on them. Enterprise customers still rely on GPT integrations. The commercial relationship isn't going anywhere soon.

But the strategic direction is no longer aligned. Microsoft is building what it needs to eventually not need OpenAI at the frontier level.

The challenge is execution. Suleyman himself acknowledged "this is very early days." MAI-Thinking-1 matching competitors "in its weight class" is not yet matching GPT-5 or Google's Gemini Ultra. Microsoft is years behind on in-house frontier development — a gap created because the OpenAI contract made internal competition pointless.

Now the clock is running.

What It Means for You

If you're an enterprise customer paying for Microsoft Copilot or Azure OpenAI services, you're about to have more options — and potentially more leverage on pricing. Competition between Microsoft's own models and OpenAI's models, even within the same corporate ecosystem, means neither side can charge whatever they want forever.

For the broader AI market, Microsoft entering the frontier model race in earnest is significant. Google, Anthropic, and Meta now have to account for a competitor with virtually unlimited compute resources, a captive enterprise customer base of hundreds of millions, and a CEO who just publicly said the old contract was holding them back.