Skip to content

Multi-Model AI Strategy for Microsoft Shops: The Aggregator Advantage

By Eddy Gil (LinkedIn)Last updated:

Eddy Gil is the founder of EGil Marketing, a fractional marketing leadership practice for B2B tech and Microsoft partners, and a public speaker on marketing and AI, with talks in North America and Europe.

  • AI & Copilot
  • Microsoft Copilot
  • AI Strategy
  • Microsoft Partner
  • Consultant
  • Product Company
Headline 'Stop building it. Sell the governance.' beside a blueprint sheet with three items in a column: 'Governance + enablement' on top with a note reading 'Sell this', a crossed-out 'Middleware' in the middle, and 'Claude + GPT' at the bottom.
In this post
  1. The short version
  2. Most of your multi-model strategy is already built
  3. Multi-model is the default, not a clever hedge
  4. Microsoft is the aggregator
  5. So what is actually left to strategize
  6. What this means for you
  7. The play: market the multi-model map

The short version

  • A multi-model AI strategy sounds like a big build, and for some organizations it genuinely is one.
  • If you are a Microsoft shop, most of it is already built. Microsoft aggregates two of the three frontier labs for you.
  • Copilot gives users model choice; Microsoft Foundry gives developers Claude and GPT on one platform.
  • So your real strategy work is governance and enablement, not integration. This post covers where a gateway still earns its place, and where it does not.

Most of your multi-model strategy is already built

Search “multi-model AI strategy” and you will find plenty of advice to build a serious engineering project: abstraction layers, gateways, routing logic, a roadmap measured in quarters. For an organization with real non-Microsoft needs, that can be exactly right. But if you are a Microsoft shop, a good part of that is already in the platform, so the honest strategy is often smaller than the roadmap suggests, and your job is to know which parts your client actually needs to build.

Here is the claim, stated plainly: if you are on Microsoft, your multi-model strategy is largely made for you. Microsoft aggregates two of the three frontier labs, Anthropic and OpenAI. Your real work is governance and enablement, not building plumbing. Let me show you why.

Multi-model is the default, not a clever hedge

Start with reality. Using more than one model is not an advanced move anymore, it is the normal state. In a16z’s survey of 100 enterprise CIOs, 37% were already using five or more models, up from 29% the year before. They got there because the lead keeps moving: in 2025 Anthropic reached 40% of enterprise LLM spend while OpenAI fell to 27%, down from 50% in 2023, with Google third at 21%. And it has not settled since: in September OpenAI shipped GPT-6 Astra, and the top of the leaderboard moved again. In a market that reshuffles that fast, committing to one lab is the risky choice, not the safe one. So yes, you want access to more than one. Everyone does. Your clients are already there: alongside Copilot, 54% of organizations use ChatGPT Enterprise, 36% use Gemini and 32% use Claude for Work (ShareGate, Sep 2026; it sells Microsoft 365 governance tools).

Bar chart of organizations using other AI tools alongside Copilot: ChatGPT Enterprise 54%, Gemini 36%, Claude for Work 32%. Source: ShareGate, State of Microsoft 365, Sep 2026.

Microsoft is the aggregator

Here is the part the big-roadmap version misses: you do not have to assemble multi-model access yourself. Microsoft already did.

Your users get it inside Copilot, where model choice puts Claude and GPT in the apps they already work in. Your developers get it in Microsoft Foundry, which offers Claude and GPT frontier models on one platform. Two of the three frontier labs, on tap, with no build. And the roster keeps growing: Microsoft has started adding xAI’s Grok to Copilot as a preview an admin has to turn on. One thing to flag to a client, though: Google’s Gemini is not in Foundry, and Google holds about 21% of enterprise LLM spend, so if a client specifically needs Gemini, that is a real reason to look at a gateway. For everyone else, you are standing on the aggregator already.

Enterprise LLM spend share in 2025: Anthropic 40%, OpenAI 27%, Google 21%. Anthropic and OpenAI models are in Microsoft Foundry; Google’s are not. Sources: Menlo Ventures, Dec 2025; Microsoft Learn.

One planning note for the agent side. Everyday model choice in the apps is covered by the license, within fair-use limits. Copilot Cowork tasks are measured in Copilot Credits, Copilot Studio agents can draw on them too, and the model is one of the factors in how many a task uses. For the people doing the work, that shows up as a monthly limit set by the organization, not a bill. Reach it and the work waits until more credits are approved. So part of the multi-model strategy you scope for a client is deciding which teams get which limits, before a deadline finds out for you.

So what is actually left to strategize

If the plumbing is done, what is your strategy? Two things, and neither is middleware.

Governance: deciding which models are enabled, for whom, and under what rules. That is a real decision with security and compliance weight, and it is the part clients actually need help with.

Enablement: teaching people to use the choice. Which model for which task, how to switch, how to test on their own work instead of trusting a leaderboard. That is where the value gets realized or wasted.

That is the part that is genuine work. Notice that none of it is integration.

“Don’t we still need a gateway or abstraction layer for a real multi-model strategy?”

That is the question a client will put to you, and for most Microsoft shops the honest answer is no. The abstraction is the platform. You would add a gateway for specific reasons: heavy non-Microsoft vendors, a specialized model Foundry does not carry, or a hard portability requirement. Those are real cases, and they are the exception. Adding middleware you do not need is how a simple strategy turns into a maintenance project you will resent in a year.

What this means for you

If you are a consultant (SI, MSP, advisory): reposition the “multi-model strategy” engagement. It is not an integration build, it is a governance-and-enablement program layered on top of what Microsoft already gives the client. That is faster to sell, faster to deliver, and it is the honest version. Being the consultant who says “you already have most of this, here is the part that is real” builds more trust than selling a project the client does not need.

If you are a product company (ISV, SaaS): if you build on a catalog like Foundry’s, you are flexible by default, so say so. Make “we are not welded to one lab” part of your pitch. Your buyers are already asking the question, and answering it well is a differentiator while it lasts.

The play: market the multi-model map

This is a marketing play. The one-page map is the offer. The play is how you get in front of the accounts who need it.

  1. Listen for the phrase. Flag every account where “we need a multi-model AI strategy” is being said, in your pipeline and your existing book.
  2. Reach out with the shortcut, not a scope document: “You probably already own most of a multi-model strategy through Microsoft. Want a one-page read on what you have and the one part that is actually missing?” That is a far easier yes than a six-month project.
  3. Lead with the angles that disarm the big-build pitch: multi-model is already the norm, Microsoft is the aggregator, and the real work is governance, not middleware.
  4. Point it at the proposed item: a paid multi-model map, one page, what they have via Microsoft versus any real gap, delivered fast.

That is the kind of positioning I bring to Microsoft partners, turning an intimidating-sounding strategy request into a quick, honest engagement you can win. If “we need a multi-model strategy” is floating around one of your accounts, that is the campaign I help you run. Worth a conversation.


Sources: a16z, AI Enterprise 2025 (Jun 2025); Menlo Ventures, State of Generative AI in the Enterprise (Dec 2025); Claude models in Microsoft Foundry (Microsoft Learn); Foundry Models overview (Microsoft Learn); Microsoft Learn, Copilot in Microsoft 365 apps with Anthropic models; Microsoft, Expanding model choice in Copilot with Grok (Sep 2026); Microsoft, Evolution of the Copilot pricing model (Sep 2026); Credit usage for Microsoft Copilot Cowork tasks (Microsoft Learn); Managing AI experiences enabled by usage-based billing (Microsoft Learn); Microsoft Copilot Cowork adoption guide (Microsoft); SpaceXAI as a subprocessor in Microsoft Online Services (Microsoft Learn).; ShareGate, State of M365 press release (Sep 2026)

FAQ

What is a multi-model AI strategy?

It is the approach of using more than one AI model rather than committing to a single provider, so you can match the model to the task and are not exposed when the performance lead shifts. For most organizations it is now the default, not an advanced option.

Does Microsoft support multiple AI models?

Yes. Microsoft 365 Copilot offers model choice between Claude and GPT in the apps, has started adding xAI's Grok in preview, and Microsoft Foundry offers both frontier families to developers on one platform. You get multi-model access without leaving the Microsoft stack.

What is Microsoft Foundry?

Microsoft Foundry (formerly Azure AI Foundry) is Microsoft's platform for building AI applications and agents, with a catalog of models from multiple providers, including Claude and GPT, on one platform.

Do I need an AI gateway if I am on Microsoft?

Usually not. The Microsoft platform provides the multi-model access most shops need. A gateway makes sense for specific cases (significant non-Microsoft vendors, specialized models, or strict portability requirements), not as a default.

Is Microsoft Copilot multi-model?

Yes. Copilot now lets users choose between Anthropic's Claude and OpenAI's GPT models, administered through the Microsoft 365 admin center using the AI Administrator role. Microsoft has also started adding xAI's Grok as an admin-enabled preview. It launched on OpenAI models but is no longer tied to a single lab.

How do I govern which AI models my company uses?

Through the Microsoft 365 admin center for Copilot (enabling providers and scoping access to users or groups) and the Power Platform and Foundry controls for custom builds. Governance and enablement, not integration, are the real work of a Microsoft multi-model strategy.

Eddy's B2B Marketing Notes

One email a month: B2B tech and Microsoft marketing news, plus what to do with it.

Get the notes

Talk it through with Eddy

A free 30-minute Marketing Review on Teams. No pitch deck.

Book your free Marketing Review (30 minutes) (opens in a new tab)

Marketing notes worth reading.

Eddy's B2B Marketing Notes, once a month: campaign-worthy news from B2B tech and Microsoft marketing, plus a quick idea on what to do next, tailored to who you are. No fluff, no spam.

By subscribing, you agree to receive marketing emails from EGil Marketing. You can unsubscribe anytime.

Want help with this?

Book a free Marketing Review and we'll talk through what's next for your marketing.