Why This Matters
Previously, Copilot was tied to OpenAI’s models almost entirely, but now Microsoft is opening the door to switching to models from other providers. This is risk diversification — no longer depending on a single party.
The self-built MAI team, combined with bringing in people from Inflection to lead the charge, means Microsoft wants a full-fledged AI stack of its own — not just remaining OpenAI’s largest shareholder as before.
For developers building on Azure or the Copilot API, this has a direct impact, since the number of model options will grow. But you’ll need to keep up with which model suits which task, because each provider excels in different areas.
Note: The spec data attached refers to the iPhone 17 Pro Max and is unrelated to this topic, so no figures from it were referenced in this section.
The Latest Moves That Say It All
The clearest turning point is that Microsoft has started letting Copilot switch between models from multiple providers, no longer locked into OpenAI the way it used to be.
Alongside this comes news that the in-house MAI team is moving forward with developing its own models in full — not just experimenting.
This signal matters because it’s essentially a public declaration that Microsoft no longer sees itself as just an investor or major partner of OpenAI, but is positioning itself as a competitor with its own offerings in the same arena.
Lately I’ve noticed Copilot’s responses keep changing — sometimes shorter, sometimes with new quirks I hadn’t seen before — without any clear update announcement.
Then I came across the news about the MAI team building their own models to this extent, and it finally clicked why — the Copilot we use every day might not be running on the same GPT it used to anymore.
This isn’t just boardroom-level politics — it directly affects users whose daily work is tied to Copilot. If the backend keeps switching models, the behavior we’ve gotten used to — like coding style or response tone — can shift at any time, without us even realizing it.
Microsoft Today Is No Longer Just an OpenAI Reseller for Enterprises
Microsoft used to play a simple role: wrap GPT models into Copilot and sell them through Azure to enterprises worldwide. That picture has now changed, because there’s now an in-house team, MAI (Microsoft AI), developing separate models — no longer relying 100% on OpenAI as before.
What’s interesting is that Copilot is being repositioned as a “brand,” not “the face of GPT.” In simple terms, Copilot can switch to whichever model runs behind the scenes — whether it’s OpenAI’s, MAI’s own, or from another provider — depending on which model suits which task.
This is a quiet but hugely impactful shift in strategy — from a partner dependent on OpenAI to a competitor with its own playing field.
From Closest Ally to Rival on the Same Board
Walking through the timeline, the change is more significant than it seems.
| Factor | Microsoft, Before | Microsoft, Now |
|---|---|---|
| AI Model | Used GPT from OpenAI exclusively | Has its own MAI model |
| Azure AI Foundry | Primarily tied to OpenAI | Open to choosing from multiple providers |
| Dependence on OpenAI | High, tens of billions invested | Reduced proportion |
From the table, you can see Microsoft hasn’t abandoned OpenAI entirely — it’s just no longer “exclusively locked in” the way it used to be.
Previously, Copilot was almost entirely the face of GPT, but now it has its own alternative option added in. This is the turning point that has moved Microsoft from standing on the sidelines to becoming a player on the very same field.
Four Moves That Show Microsoft Is Serious About This Game
The first move was launching its own model, MAI-1 — no longer relying on GPT alone.
The second move was bringing in Mustafa Suleyman, co-founder of DeepMind, along with the team from Inflection, to directly head the AI division. This means Microsoft now has its own AI leadership, no longer waiting to hear only from OpenAI.
The third move is letting Copilot switch models behind the scenes on its own — general document work might use Microsoft’s own model, while tasks requiring higher accuracy switch to another model.
The final move is Azure AI Foundry, which lets enterprises freely choose their model — whether GPT, Claude, or an open-source model — without being tied to a single provider.
Put together, this is a clear shift from being “OpenAI’s biggest customer” to being “a player with its own set of options.”
Positioning Microsoft Alongside OpenAI and Anthropic
Compared head-to-head, these three companies are genuinely playing different games. Microsoft isn’t competing to build the smartest model — it’s competing on “who can reach enterprises more broadly” through Azure, which businesses already use.
OpenAI and Anthropic remain focused purely on the models themselves, competing on reasoning and accuracy. Microsoft, meanwhile, uses its ecosystem strength (Office, Teams, Azure) as the draw for customers instead.
The clear distinction is speed of deployment to real users — Microsoft has the advantage because it already has enterprise channels in place, without having to start from zero like the other two.
| Factor | Microsoft (Azure AI) | OpenAI / Anthropic |
|---|---|---|
| Core Strength | Enterprise reach + integration with existing work tools | Pure model capability |
| Model Choice Flexibility | Choose from multiple providers via Azure AI Foundry | Locked into their own model |
| Development Focus | Integration + infrastructure | Research + model capability |
Put simply, Microsoft is playing a “risk diversification” game, while the other two are still betting everything on the model alone.
What’s Gained and Lost as Microsoft Stops Depending on One Side
For enterprises already using Microsoft 365 or Azure, having its own model means more negotiating power — no longer having to wait for OpenAI to adjust its roadmap to match enterprise needs. But it comes with risks too — resources once devoted to a single partnership must now be split to develop parallel stacks, which could slow down progress on both sides compared to when they combined forces.
For developers building on Azure AI Foundry, this is better news, since there are multiple model options to choose from depending on the task, no longer tied to a single provider as before. The long-term risk is that if the relationship with OpenAI keeps cooling, some features that used to be exclusive may disappear from the Microsoft ecosystem.
Pros
- +Enterprises gain more negotiating power, no longer tied to a single provider
- +Developers can choose from a variety of models via Azure AI Foundry based on the task at hand
Cons
- −Development resources are split across multiple fronts, potentially slowing overall progress
- −A cooling relationship with OpenAI may affect exclusive features in the future
The Price Enterprises Pay When Two Partners Start Drifting Apart
The headlines look sleek, but the IT teams who actually have to choose a stack have a much bigger headache.
Until now, enterprises locked into long-term Azure OpenAI Service contracts often planned their roadmaps on the assumption that Microsoft and OpenAI would move in the same direction indefinitely. Now that the two sides are starting to compete with each other, the old “sure thing” roadmap is no longer so sure.
If an enterprise has already committed to a particular model on Foundry, migrating the entire pipeline, prompts, and fine-tuning to the other side is no small effort — the bigger the system, the more painful it gets.
On top of that, exclusive features that used to come from OpenAI first may no longer arrive first in the future, because the relationship has changed. This is a risk that needs to be factored in when planning next year’s budget — not after the price hike notice arrives.
What to Watch Next in the AI War Between the Two Giants
This story doesn’t end with Microsoft simply talking to Anthropic more — keep an eye on which direction Copilot’s default model gradually shifts toward, and whether the next round of enterprise contracts locks into a single provider or opens up more choices than before.
For enterprises and developers, what you can do starting today is design your AI-calling layer so it can easily swap models, without being too tightly bound to any single provider’s API.
This split between Microsoft and OpenAI has only just begun — it’s not the end. The power equation in AI can still shift many more times. Whoever builds flexibility into their systems now will suffer less when the next wave hits.