Microsoft chief says the company was cleared by an OpenAI contract change to formally pursue superintelligence with its own AI organization. Mustafa Suleyman, CEO of Microsoft AI, told VentureBeat at Microsoft Build 2026 that the change happened roughly six months ago and gave Microsoft room to use its own researchers, data pipelines, and custom silicon for frontier AI work.
Microsoft chief says the company was cleared by an OpenAI contract change to formally pursue superintelligence with its own AI organization. Mustafa Suleyman, CEO of Microsoft AI, told VentureBeat at Microsoft Build 2026 that the change happened roughly six months ago and gave Microsoft room to use its own researchers, data pipelines, and custom silicon for frontier AI work.
The disclosure marks a shift in how Microsoft describes its artificial intelligence strategy after years in which its OpenAI partnership shaped Copilot, Azure AI services, and much of its enterprise AI story. Suleyman said the effort is still in its early days, while stressing that Microsoft is not walking away from OpenAI.
What Happened
According to VentureBeat, Suleyman said Microsoft was only recently able to formally pursue superintelligence after a contractual change with OpenAI. The partnership, backed by more than $13 billion in Microsoft investment, had given Microsoft early access to OpenAI models and a strong position in enterprise AI.
That arrangement also created strategic dependence. VentureBeat reported that the earlier deal limited Microsofts ability to pursue its own artificial general intelligence research and capped the scale of systems it could train by compute thresholds measured in floating point operations per second, or FLOPS.
The renegotiated agreement, which VentureBeat said was previously reported by Fortune and Axios, removed those limits. Suleyman framed the result as a dual-track strategy: Microsoft can keep working closely with OpenAI while also building its own path toward what he calls humanist superintelligence.
Microsoft Chief Says MAI Models Show The Shift
The most visible sign of that shift came at Build 2026, where Microsoft announced seven new in-house AI models under the MAI name. The models cover reasoning, code generation, image creation, transcription, and voice synthesis, giving Microsoft a broader first-party model portfolio than it previously had.
The flagship model, MAI-Thinking-1, is a 35-billion-active-parameter reasoning model. Microsoft says it matches leading models in its weight class on key software engineering benchmarks and shows advanced mathematical reasoning. Suleyman also emphasized that Microsoft trained its reasoning models from scratch using clean, commercially licensed data rather than distilling them from third-party frontier models.
The rest of the family includes MAI-Code-1-Flash for GitHub Copilot and Visual Studio Code, MAI-Image-2.5 for image generation and editing, MAI-Transcribe-1.5 for transcription across 43 languages, and MAI-Voice-2 for multilingual speech generation. The models ship through Microsoft Foundry, and developers can tune model weights through third-party platforms including OpenRouter, Fireworks, and Baseten.
The Details
Suleyman told VentureBeat that Microsofts goal is to be able to produce state-of-the-art frontier-scale models over the next five years. He described that as a long transition, not a short-term replacement for OpenAI. Microsoft still has access to OpenAI, Anthropic, and thousands of models through Foundry, which Suleyman described as a source of optionality.
Microsoft is also tying the MAI model family to enterprise customization. Frontier Tuning, announced alongside the models, lets customers adapt MAI models using their own proprietary data, workflows, and terminology inside their compliance boundary. Microsoft describes the approach as using reinforcement learning environments where agents can learn from workplace tasks without affecting production systems.
VentureBeat reported that Microsoft cited early examples including an Excel-tuned MAI model that matched GPT 5.4 performance while operating at up to ten times greater efficiency. The company also named enterprise partners using Frontier Tuning, including Mayo Clinic, EY, Land O'Lakes, and Pearson.
Context and Background
The broader strategy extends beyond models. Suleyman argued that enterprise data will become a key training frontier as publicly available web data becomes less central and more legally contested. Microsofts position in Microsoft 365, Teams, Dynamics 365, and Azure gives it access to the workflows where many companies already operate, although customer-specific tuning is framed as happening within secure enterprise boundaries.
Microsoft is also building infrastructure for autonomous agents. VentureBeat reported that Microsoft Scout, Windows 365 for Agents, hosted agents in Foundry, and a new Microsoft Agent Framework were part of the Build announcements. Suleyman described the industry as moving beyond conversation toward action, where agents can work across software tools with governed identities and auditable behavior.
Compute is another part of the plan. Suleyman said Microsoft is the largest buyer of graphics processing units on the planet and will continue buying Nvidia accelerators for years. At the same time, Microsoft is developing its own Maia AI chips, with Maia 200 already running in production across data centers in Iowa and Arizona, according to the VentureBeat report.
Why It Matters
The OpenAI contract change matters because it gives Microsoft more control over the model layer behind its AI products. Instead of relying only on external model providers, Microsoft is building a parallel internal stack that includes models, enterprise tuning, agent infrastructure, cloud deployment, and custom silicon.
That does not make OpenAI irrelevant to Microsoft. It does mean Microsoft is preparing for an AI market where owning core capabilities may matter as much as reselling partner models through cloud and productivity platforms.
Key Takeaways
- Mustafa Suleyman said Microsoft was freed by an OpenAI contract change to formally pursue superintelligence.
- Microsoft introduced seven MAI models developed by its in-house AI Superintelligence Team.
- The MAI family spans reasoning, coding, image generation, transcription, and voice synthesis.
- Microsoft says its reasoning models are trained from scratch on clean, commercially licensed data.
- The company is combining first-party models with enterprise tuning, AI agents, and custom Maia chips.
Source: VentureBeat
Frequently Asked Questions
What did Microsofts AI chief say about OpenAI?
Mustafa Suleyman said Microsoft was set free from its OpenAI contract roughly six months ago to formally pursue superintelligence internally.
Is Microsoft ending its OpenAI partnership?
No. Suleyman said Microsoft still has OpenAI, Anthropic, and thousands of models in Foundry, while also building its own frontier model capability.
What are Microsofts MAI models?
They are seven in-house AI models announced at Build 2026 for reasoning, code generation, image creation, transcription, and voice synthesis.
What is Frontier Tuning?
Frontier Tuning lets enterprise customers customize MAI models with their own data, workflows, and domain terminology inside a secure compliance boundary.
Why does Microsoft want its own AI models?
Suleyman said Microsoft wants the ability to build state-of-the-art frontier-scale models over the next five years, not only buy models from third parties.
The Bottom Line
Microsofts comments show a company trying to reduce strategic dependence on any single AI partner while preserving the OpenAI relationship that powered its early Copilot push. The Build 2026 announcements suggest Microsoft now sees first-party models, enterprise data, autonomous agents, and custom silicon as parts of the same long-term AI plan.
Continue reading related coverage in News or browse all stories on the articles page.