What are Microsoft homegrown AI models? Microsoft homegrown AI models are foundation models that Microsoft designs and trains entirely in house, independent of OpenAI, and the company is preparing to expand that lineup at its Build 2026 developer conference in San Francisco on June 2 to 3, 2026. According to reporting from Reuters and The Information, Microsoft will unveil a new homegrown coding model built to power GitHub Copilot, along with a reasoning model, an in house agent, and a family of models offered in multiple sizes. For commercial real estate firms that run their entire back office on Microsoft 365, Teams, and Azure, the real question is no longer whether AI is embedded in your software, but which company controls the model doing the work. For the broader picture, see our AI tools for real estate investors guide.
Key Takeaways
- Microsoft is set to reveal new homegrown AI models at Build 2026, including a coding model for GitHub Copilot, a reasoning model, and an in house agent, deepening its independence from OpenAI.
- The MAI Superintelligence team, led by AI chief Mustafa Suleyman, builds these models from scratch after Microsoft renegotiated its OpenAI partnership in April 2026.
- CRE firms standardized on Microsoft 365 and Copilot will see the underlying AI model change, with direct effects on cost, data governance, and vendor concentration.
- Microsoft reportedly plans to phase out internal Claude Code usage by the end of June 2026 and push its own Copilot tools to control rising AI costs.
- Model independence can give CRE buyers more pricing leverage and tighter integration, but it also concentrates more of your AI stack inside a single vendor.
Microsoft Homegrown AI Models Explained
For most of the past decade, Microsoft's AI strategy was synonymous with OpenAI. Copilot, the AI layer woven through Word, Excel, Outlook, and Teams, ran largely on OpenAI's GPT models. That is changing fast. On April 2, 2026, Microsoft released its first three in house foundation models: MAI-Transcribe-1 for speech to text, MAI-Voice-1 for audio generation, and MAI-Image-2 for image creation. At Build 2026, the company is expected to extend that family with MAI-Voice-2, MAI-Image-2.5, and MAI-Transcribe-1.5, and, more importantly, to move into the categories that matter most for knowledge work: coding and reasoning. We covered the original launch in our breakdown of Microsoft's MAI models and what they mean for CRE; the Build 2026 news is the next chapter, and a far more consequential one.
The new homegrown coding model is built specifically to improve GitHub Copilot, the AI pair programmer that lost meaningful ground over the past year to Anthropic's Claude Code. The reasoning model targets the multi step analysis that underpins agentic workflows, the same capability CRE teams rely on when an AI assistant reads a lease, reconciles a rent roll, or drafts an investment memo.
Why Microsoft Is Building Its Own AI
Two forces are driving Microsoft toward homegrown AI models: control and cost. The control story begins with the partnership itself. After Microsoft and OpenAI renegotiated their agreement in April 2026, the terms that had kept Microsoft's internal team from training top tier foundation models were loosened, freeing Mustafa Suleyman's MAI Superintelligence team to build frontier models from scratch (Source: Reuters). These are not fine tuned versions of OpenAI systems; they are original models trained by Microsoft, reportedly by remarkably small teams.
The cost story is just as important. Enterprises spent 2025 and early 2026 discovering that third party AI licenses are expensive at scale. Microsoft itself allowed thousands of employees to use Claude Code internally, then reportedly moved to phase that usage out by the end of June 2026 to rein in spending, a dynamic we explored in our analysis of enterprise AI sticker shock. Owning the model removes a major recurring expense and gives Microsoft full control over its AI roadmap, its pricing, and the margin structure of every Copilot seat it sells.
What This Means for the CRE Technology Stack
Commercial real estate is one of the most Microsoft dependent industries in the economy. Offering memorandums live in PowerPoint, underwriting models live in Excel, investor updates go out through Outlook, and asset management teams coordinate in Teams. When Microsoft swaps the AI engine underneath Copilot, that change flows directly into the daily tools of acquisitions analysts, asset managers, and property teams, often without anyone noticing the model changed.
There are real upsides. Native models mean fewer third party integrations to vet, fewer separate subscriptions to manage, and AI features that arrive through standard Microsoft updates rather than custom plumbing. For firms that already run Copilot Studio and agentic workflows, this can lower the barrier to deploying AI across leasing, accounting, and reporting. See our guide to Microsoft Copilot Studio agents for CRE for how those building blocks fit together.
Vendor Concentration: The New Risk for CRE AI
The flip side of convenience is concentration. If your transcription, voice, image, coding, reasoning, and document analysis all run on Microsoft models inside Microsoft 365, then Microsoft becomes a single point of dependence for your entire AI operation. That matters for three reasons. First, pricing power shifts toward the vendor once switching costs are high. Second, a model change you did not choose can quietly alter the behavior of tools your team trusts for financial analysis. Third, governance and data residency questions become harder to answer when one provider touches every workflow.
Smart CRE operators treat this the way they treat lender concentration or tenant concentration: as a risk to be measured and managed, not ignored. Maintaining the ability to compare outputs across Microsoft, OpenAI's GPT-5.4, Anthropic's Claude, and Google's Gemini keeps leverage on your side. Our review of GPT-5.4 financial tools for CRE underwriting shows why side by side evaluation still matters even inside a single ecosystem.
How CRE Investors Should Respond
You do not need to predict which lab wins to position your firm well. A few practical moves:
- Inventory your AI surface area. List every Microsoft tool your team uses with AI features, and note which decisions those tools influence, especially underwriting, tenant screening, and valuation.
- Document model governance. Record which models touch sensitive data such as rent rolls, trailing 12 month statements, and investor information, and keep a written AI use policy.
- Preserve optionality. Avoid hard wiring critical workflows to a single model so you can switch if pricing or performance shifts.
- Pressure test outputs. Spot check AI generated financial work against your own models; honesty and accuracy vary by model and by version.
For personalized guidance on building a resilient AI stack, connect with The AI Consulting Network, which helps CRE firms evaluate platforms and put governance guardrails in place before problems appear.
The bigger context is a market racing ahead of its own readiness. AI in real estate is projected to reach $1.3 trillion by 2030 at a 33.9% compound annual growth rate, yet while 92% of corporate occupiers have launched AI programs, only 5% report achieving most of their goals (Source: JLL). Microsoft's homegrown AI models will put more capability into the hands of CRE professionals than ever before. Turning that capability into results still depends on governance, judgment, and disciplined implementation. If you want hands on help, reach out to Avi Hacker, J.D. at The AI Consulting Network.
Frequently Asked Questions
Q: What are Microsoft homegrown AI models?
A: They are foundation models Microsoft designs and trains in house, independent of OpenAI, through its MAI Superintelligence team. The lineup spans transcription, voice, and image, and, as of Build 2026, coding and reasoning models that power GitHub Copilot and Microsoft 365.
Q: How does this affect CRE firms that use Microsoft 365?
A: Most CRE firms run on Microsoft 365, Teams, and Azure, so when Microsoft changes the AI model behind Copilot, those changes reach your Excel, PowerPoint, and Outlook workflows automatically. The upside is tighter integration and potentially lower cost; the risk is greater dependence on one vendor.
Q: Is Microsoft dropping OpenAI and Anthropic?
A: Not entirely. Microsoft retains its OpenAI partnership and still offers third party models through Azure AI Foundry. But it is reducing reliance on outside models, including reportedly phasing out internal Claude Code usage by the end of June 2026 to control costs and own more of its roadmap.
Q: What should CRE investors do right now?
A: Inventory where AI touches your workflows, document which models handle sensitive data, keep the ability to compare models across vendors, and spot check AI generated financial analysis against your own underwriting before relying on it.