Microsoft’s in-house AI push and the OpenAI deal rewrite point to less dependency risk

Microsoft’s relationship with OpenAI is shifting in ways that are likely to matter for how investors think about Microsoft’s long-term AI strategy and costs. In parallel with changes to the partnership’s commercial terms, Microsoft is also presenting an expanded stack of in-house AI models and agent capabilities, positioning its own Azure-centered platform as the backbone for future customer deployments.

On the partnership front, OpenAI and Microsoft announced an amended agreement on April 27, 2026, aimed at simplifying how they work together. The update keeps Microsoft as OpenAI’s primary cloud partner, and says OpenAI products will ship first on Azure unless Microsoft cannot and chooses not to provide the necessary capabilities. But it also explicitly allows OpenAI to serve its products to customers across any cloud provider.

The amendment also changes the economics between the companies. Microsoft will have a license to OpenAI intellectual property for models and products through 2032, with that license becoming non-exclusive, and Microsoft will no longer pay a revenue share to OpenAI. OpenAI to Microsoft revenue-share payments continue through 2030, at the same percentage, but subject to a total cap. The joint statement also reiterates that Microsoft remains a major shareholder in OpenAI.

At the same time, Microsoft has been emphasizing its own model and agent roadmap. In a Microsoft Build 2026 post, the company described work from its AI Superintelligence Team releasing a family of seven new in-house models, starting with MAI-Thinking-1, which Microsoft calls its first reasoning model. The company said MAI-Thinking-1 is trained from scratch, uses enterprise-grade clean and commercially licensed data, and is designed for efficient multi-step instructions, long-context reasoning, and code generation. It also cited performance comparisons and described the model’s scale and context window, while noting it is available via Foundry in private preview.

Microsoft used the same Build update to lay out a broader “agent” layer. The company introduced Microsoft Scout as a personal agent for work, built on OpenClaw and WorkIQ, designed to handle tasks like meeting preparation and scheduling conflicts by proactively using tools such as Teams and Outlook. Microsoft also described enterprise governance components for agents, including Agent 365 for local agents, as part of a control plane intended to observe, govern, and secure agents across an organization.

For infrastructure and deployment context, Microsoft has also been framing its Azure AI approach through Microsoft Foundry. A separate March 2026 Microsoft blog described Foundry as an enterprise-scale system for building, deploying, and operating AI agents, with Foundry Agent Service and observability in the control plane aimed at giving developers visibility into agent behavior. It also connected Foundry’s agent focus to inference-heavy, reasoning-based workloads that customers can deploy and operate consistently.

Still, the degree to which Microsoft’s in-house models reduce reliance on OpenAI in practice remains unclear from public disclosures. While Microsoft and OpenAI laid out revised partnership terms, neither the partnership statement nor the Build post breaks out how usage, pricing, and customer revenue will shift between OpenAI-powered experiences and Microsoft’s own models over time. Investors looking for a cleaner view of the risk reduction will likely have to wait for later product-level and financial detail.