Last updated: August 7, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
Microsoft's fiscal 2026 annual report discloses $24.1 billion in revenue from commercial arrangements with OpenAI, and Bloomberg's analysis of the filing puts that at roughly 70% of Microsoft's total AI business for the year. If you build products, workflows, or a dev team's tooling on top of Microsoft's AI stack — GitHub Copilot, Azure OpenAI Service, Microsoft 365 Copilot — that concentration is worth understanding before you lock in further.
Short answer: Microsoft's FY2026 10-K shows OpenAI accounts for about 70% of its AI revenue ($24.1B of an estimated $34B). For buyers, that means Copilot and Azure OpenAI Service still run mostly on OpenAI models day to day, even as Microsoft quietly adds Anthropic and its own MAI models as backups. Your practical risk is pricing and roadmap dependency, not an outage.
I test AI coding and productivity tools for this site every month, and when I went back through Microsoft's own Copilot and Azure OpenAI Service documentation after this filing came out, the buyer question was more interesting to me than the investor one everyone else was covering. Most of the "70% of AI revenue" coverage this week is written for people who own Microsoft stock. In my testing, the more useful question for a reader picking tools is simpler: if Microsoft's AI business runs mostly on someone else's models, what does that mean for the product you're actually paying for?
What the filing says, in plain terms
Microsoft's 10-K, filed the week of August 4, 2026, states that it recorded $24.1 billion in revenue from commercial arrangements with OpenAI, inclusive of revenue-sharing payments, for the fiscal year ended June 30, 2026. Accounts receivable from OpenAI sat at $6.0 billion at year-end, which is a large enough invoice balance to tell you the relationship is still growing, not winding down.
Microsoft doesn't publish a single "AI revenue" line, so the 70% figure is a calculation, not something Microsoft stated outright. CEO Satya Nadella told investors in the March 2026 quarter that Microsoft's AI business was tracking toward a $37 billion annualized run rate. Bloomberg's analysts applied the 123% year-over-year growth rate Microsoft reported that same quarter, which lands full-year AI revenue around $34 billion. Set the disclosed $24.1 billion from OpenAI against that estimate, and you get close to two-thirds to 70%, according to TheNextWeb’s coverage of the filing. Microsoft's total company revenue was $331.8 billion for the year, so OpenAI alone is a little over 7% of everything Microsoft sells — big for an "AI segment," small for the whole company. If you want the full investor-side math on that split, I broke it down separately in Microsoft’s AI sales mostly come from OpenAI, disclosures show.
Why this matters if you build on Microsoft's AI stack
Here's the part that gets skipped in most coverage of this filing: Microsoft is disclosing heavy OpenAI reliance in its financials at the same time it's telling developers, internally, to lean on OpenAI more, not less. According to TheNextWeb's reporting, Jay Parikh, who leads Microsoft's CoreAI engineering group, directed staff to default GitHub Copilot to OpenAI's newer model, saying that "shifting more workloads to OpenAI models helps us get greater value from our token investment." That's a cost-optimization call, not a diversification one, and it runs the opposite direction from Microsoft's public messaging about open-weight models and reduced vendor concentration.
KeyBanc analyst Jackson Ader summed up the investor read on this in a way that matters for buyers too: "The more of that revenue comes from services to OpenAI, the more favorably I'm going to look at it." Translated for a tool buyer: Wall Street wants Microsoft leaning into OpenAI, because that's the growth story right now. So don't expect Microsoft to quietly pull Copilot off OpenAI models anytime soon, even while it builds out Anthropic and MAI-model options elsewhere in the stack.
Step-by-step: how to check your own exposure
1. Find out which model actually answers your prompts
GitHub Copilot, Microsoft 365 Copilot, and Azure OpenAI Service each let you pick a model in settings or the admin console. Check whether you're on a default OpenAI model, an Anthropic option, or Microsoft's own MAI family — most teams have never looked.
2. Read your Azure OpenAI Service contract terms
If you're calling OpenAI models through Azure, your pricing and rate limits are governed by Microsoft's commercial terms with OpenAI, not a direct OpenAI account. Check your Azure OpenAI Service resource page for the specific model versions you're provisioned on and their deprecation dates.
3. Check whether multi-model support exists in your product tier
Copilot's enterprise tiers increasingly support model switching between OpenAI, Anthropic, and Microsoft's own models. Free and lower tiers often don't. Confirm which tier you're on before assuming you have a fallback.
4. Watch Microsoft's own diversification signals
Microsoft's 10-K and recent public statements reference its investment in Anthropic and continued work on in-house MAI models. That's the trend that will eventually loosen this dependency — track it in Microsoft's quarterly earnings calls, not just this one filing.
5. Price out a non-Microsoft alternative before you need one
Even if you stay on Copilot or Azure OpenAI Service, knowing what a direct OpenAI API plan, Anthropic's Claude, or a tool like Cursor would cost for the same workload gives you real leverage if pricing or model access changes.
Example prompts you can copy
Paste these into any AI chat tool to speed up your own vendor-exposure check:
“ I use [Copilot / Azure OpenAI Service] for [coding / writing / support]. Based on public documentation, which underlying model provider does my current plan default to, and does my tier support switching providers? “
“ Compare the published pricing for running [X requests/month] of a GPT-class model through Azure OpenAI Service versus a direct OpenAI API account versus Anthropic's API. Flag any minimum commitments or overage fees. “
“ If [vendor]'s underlying AI model provider changed pricing or was deprecated, what would it cost in time and rework to switch my current setup to a different provider? “
These work because they force the AI to name a specific provider and tier instead of giving a generic answer about "AI vendor risk," which is where these questions usually go vague.
Common mistakes to avoid
In my testing, the most common mistake is assuming "Copilot" means one fixed thing — it doesn't, since Microsoft ships Copilot as a brand across at least four different products (GitHub Copilot, Microsoft 365 Copilot, Windows Copilot, and Security Copilot), each with its own model defaults and switching options. The second mistake is reading Microsoft's Anthropic investment as proof the OpenAI dependency is already shrinking; as of this filing, it's a hedge Microsoft is building, not a replacement that's live in most products yet. Third, teams often assume enterprise-tier model switching is available because they saw a headline about it, without checking their actual contract or admin console. Fourth, don't treat the 70% figure as a hard, Microsoft-disclosed number — it's Bloomberg's estimate built on a growth-rate assumption from a single earnings call, not a line Microsoft itself published.
Where the model dependency actually sits
| Microsoft AI surface | Default model source (Aug 2026) | Switch to another provider? |
|---|---|---|
| GitHub Copilot | OpenAI (directed to default to it per internal guidance) | Yes, on paid tiers — Anthropic and Google models selectable |
| Microsoft 365 Copilot | Mostly OpenAI, some MAI-model routing for cost/latency | Limited; mostly automatic, not user-selectable |
| Azure OpenAI Service | OpenAI models exclusively, hosted on Azure | No — by definition it's OpenAI's models |
| Windows Copilot | Mix of OpenAI and Microsoft's own on-device MAI models | No user-facing switch |
Sourced from current public product documentation and the model-routing behavior reported by TheNextWeb, August 6, 2026. Vendor defaults change fast — recheck your own admin console before making a switching decision.
Tools that make this easier
If GitHub Copilot's OpenAI dependency is the part that worries you most, my head-to-head on Cursor vs. GitHub Copilot and GitHub Copilot vs. ChatGPT both cover how each tool's underlying model choices affect day-to-day coding output, not just price. If you're weighing whether to diversify away from OpenAI-backed tools entirely, Claude vs. ChatGPT is the comparison I get asked about most, and it's a fair look at what you'd be trading off. My walkthrough on how to actually use Copilot is worth a skim if you haven't checked your own tier's model settings recently — most people haven't. And if you're choosing AI tools for a small team and want a vendor-risk lens baked into the decision, my AI tool ratings page and best AI tool for small business guide are the places I'd start.
My take
Microsoft's 70% OpenAI dependency is a real number, but the buyer risk it creates is narrower than the investor headlines suggest. You're not at risk of Copilot or Azure OpenAI Service disappearing — Microsoft's $13 billion funding commitment to OpenAI and its own compute ownership make that scenario unlikely for years. The actual risk is slower and less dramatic: pricing that tracks OpenAI's own cost structure, model defaults set by internal cost optimization rather than your preference, and a multi-model future that's further along in Microsoft's messaging than in most people's actual admin console. If you're building anything long-term on this stack, spend twenty minutes checking your tier's model settings and pricing terms this week — that's more useful than watching the stock-analyst debate over whether 70% is too high.
Frequently Asked Questions
Does Microsoft's 70% OpenAI dependency mean GitHub Copilot could shut down?
No. Microsoft owns the Azure infrastructure OpenAI runs on and has committed $13.0 billion in funding to OpenAI, with $11.9 billion funded as of June 30, 2026. The concentration is a business-strategy risk, not a product-continuity one.
Can I switch Copilot or Azure OpenAI Service away from OpenAI models?
It depends on the product and tier. GitHub Copilot's paid tiers increasingly support Anthropic and Google model options. Azure OpenAI Service, by design, only serves OpenAI's models — to use a different provider there, you'd need a separate Azure AI service or a different vendor entirely.
Is Microsoft actually diversifying away from OpenAI?
It's building toward that with its Anthropic investment and in-house MAI models, but as of this filing, internal guidance reported by TheNextWeb has GitHub Copilot defaulting more toward OpenAI models for cost reasons, not less. Diversification is a stated direction, not yet the default experience.
How much of Microsoft's AI revenue is OpenAI, exactly?
Microsoft's FY2026 10-K discloses $24.1 billion in revenue from OpenAI. Bloomberg estimates that's roughly 70% of Microsoft's total AI revenue for the year, based on Nadella's March 2026 run-rate comments — it's an analyst estimate, not a number Microsoft itself published as a percentage.
Should this change which AI tool I buy?
Not on its own. It's a reason to check which model your current plan defaults to and whether your tier supports switching, so you're not caught off guard by a pricing or model change later. It's not, by itself, a reason to cancel a Copilot or Azure OpenAI Service subscription that's working for you.