Update — September 14, 2026: Elon Musk confirmed the integration directly on X, quote-posting Satya Nadella's original announcement with a two-word line: "Grok available in Microsoft Copilot." No new details beyond what's documented below — the significance is that both CEOs have now publicly put their names on the same integration, and Musk's post alone drew 6.8M views, well beyond Nadella's original announcement — a sign this crossed from an enterprise-IT story into mainstream tech-Twitter attention. Reaction under Musk's post ran skeptical, with replies questioning the rationale ("you couldn't possibly need the money") and Copilot's own reputation among developers.
Microsoft announced on September 12, 2026 that xAI's Grok models are now available as a selectable option inside Copilot for Word, Excel, and PowerPoint — rolled out through Microsoft's Frontier Program as a preview, opt-in, and explicitly not yet available to EU, EFTA, or UK customers. It's the clearest evidence yet that Microsoft is building Copilot as a genuinely multi-model product, not a single-vendor wrapper around OpenAI's models.
TL;DR
| Question | Answer |
|---|---|
| What changed? | Grok is now selectable as a model inside Copilot for Word, Excel, and PowerPoint |
| How is it rolled out? | Through Microsoft's Frontier Program, as a preview |
| Is it on by default? | No — disabled by default, requires a separate admin setting |
| Who can access it? | Eligible Frontier Program customers only |
| Where is it unavailable? | EU, EFTA, and UK, during this preview |
| What data controls exist? | Admins control what data is processed by Grok; xAI added to Microsoft's Online Services Subprocessor List |
| Which Grok model version? | Not specified in the announcement |
What actually shipped
The integration lets Frontier Program-enrolled customers select Grok as the underlying model answering a Copilot prompt inside Word, Excel, or PowerPoint, alongside the existing option to use GPT-family or Claude models within the same Copilot interface. This is a preview, gated behind a program that historically gives select Microsoft enterprise customers early access to features before general availability — not a rollout to every Microsoft 365 Copilot seat.
Two restrictions are worth flagging specifically because they shape who can actually try this today:
- It's disabled by default. An admin has to explicitly turn on Grok access as a separate setting; it does not appear automatically for eligible Frontier customers just by virtue of enrollment.
- It's geographically restricted. Microsoft explicitly excludes the EU, EFTA, and UK from this preview — a scope limitation likely tied to differing data-protection and AI-regulation requirements across those jurisdictions compared to where the preview is currently running.
On the data-governance side, Microsoft says admins retain control over what data is processed by xAI's models when Grok is selected, and xAI has been added to Microsoft's Online Services Subprocessor List — the standard mechanism Microsoft uses to disclose which third-party vendors may process customer data as part of its cloud services. This is the same category of disclosure Microsoft uses whenever it adds a new third-party AI provider into its enterprise product surface, and it's worth checking directly if your organization has specific data-residency or subprocessor-approval requirements before enabling Grok for any team.
Why this matters beyond "one more model option"
Microsoft's own framing describes this as part of a broader multi-model Copilot strategy — the same direction explainx.ai has tracked in GitHub Copilot's Hydrafusion multi-model orchestration, where Copilot increasingly treats "which model answers this" as a configurable choice rather than a fixed, Microsoft-controlled decision. Adding Grok specifically extends that same logic from GitHub Copilot's coding surface into the core Office productivity apps — Word, Excel, and PowerPoint — which is a much larger and more consequential surface in terms of daily active enterprise usage.
There's also a competitive angle worth naming plainly: Microsoft is a major investor in and infrastructure partner to OpenAI, and Copilot has historically been treated as, functionally, an OpenAI-model product wrapped in Microsoft's enterprise packaging. Making Grok — a direct competitor to OpenAI's models, built by Elon Musk's xAI — a first-class selectable option inside that same product is a meaningful signal that Microsoft's own commercial incentives now favor giving enterprise customers real model choice over exclusively channeling usage toward its most closely aligned model partner. Microsoft says it will use feedback from this specific preview to decide whether to expand Grok access to more Copilot surfaces beyond Word, Excel, and PowerPoint — meaning this could be the first of several product areas where Grok becomes a standing option, not just a Word/Excel/PowerPoint-specific pilot.
Don't confuse this with Grok's Outlook connectors
It's easy to conflate this with a separate, unrelated integration explainx.ai has already covered: Grok's Outlook Mail and Calendar connectors, which grant Grok — running as its own standalone assistant, not inside Copilot — direct read-write-send access to a user's Outlook mailbox and calendar. That integration is about Grok acting as an autonomous agent with mailbox permissions; this Copilot integration is about which underlying model generates a response when you ask Copilot to draft a paragraph in Word or build a formula in Excel. They're both xAI-Microsoft integrations announced within weeks of each other, but they involve completely different permission models, risk surfaces, and product teams — evaluate them separately before drawing conclusions about xAI's overall access footprint inside Microsoft's ecosystem.
What builders and IT admins should actually check
If your organization uses Microsoft 365 Copilot and is enrolled (or considering enrollment) in the Frontier Program, a few concrete steps before this reaches your team:
- Confirm your organization's Frontier Program eligibility and current enrollment status — this isn't available outside that program during the preview.
- Check the Online Services Subprocessor List entry for xAI directly rather than assuming Microsoft's standard data-processing terms automatically extend identically to every model provider — subprocessor terms can vary by vendor.
- Decide deliberately whether to enable the admin setting, rather than leaving it in whatever state your Frontier enrollment defaults to — since it's off by default, someone in IT will need to make an active choice to turn it on for any user or team.
- If you're in the EU, EFTA, or UK, don't expect to see this yet — plan around the exclusion rather than assuming a delayed rollout is a bug.
For anyone evaluating Grok specifically for document and spreadsheet work versus Claude or GPT-family models already available in Copilot, the practical move is the same one explainx.ai recommends whenever a new model option lands inside an existing tool: run your own representative tasks (a specific document draft, a specific spreadsheet formula problem) across the available model choices before standardizing a team's default, rather than assuming any one vendor's marketing claims transfer directly to your specific use case.
The pattern this fits: enterprise software is becoming model-agnostic infrastructure
Step back from Grok specifically, and this announcement is one more data point in a trend that's been building across enterprise software all year: the model answering a prompt is increasingly a swappable component rather than a fixed part of the product a vendor sells. A year ago, "Copilot" implicitly meant "Copilot powered by OpenAI's models" — the two were functionally synonymous in how the product was marketed and understood. Today, the same product surface can route a request to GPT-family models, Claude, or Grok depending on admin configuration, with the underlying UI, permissions model, and enterprise agreement staying constant.
This has real consequences for how organizations should think about vendor lock-in and model selection going forward. If Copilot itself is becoming a thin, model-agnostic orchestration layer over whichever provider a customer prefers, the actual decision that matters shifts from "which productivity suite do we standardize on" to "which model do we configure that suite to call for which task" — a meaningfully different and more granular decision than the all-or-nothing platform choice organizations were making even a year ago. It also means switching costs for moving between model providers inside a tool like Copilot are dropping, since the interface and workflow a user experiences doesn't have to change even if the model underneath does.
For IT and platform teams, that argues for building internal evaluation processes around specific task types rather than around vendor relationships — the same "test with your own representative tasks" recommendation above, but now framed as an organizational capability worth building rather than a one-off check before enabling a single preview feature. Teams that develop a repeatable way to compare model outputs on their own real documents and spreadsheets will be positioned to take advantage of whichever model performs best for a given task, as more of these enterprise tools converge on the same model-agnostic pattern Copilot is now demonstrating with Grok.
What's still unclear
A few details Microsoft's announcement doesn't specify, worth checking directly before making any decisions based on this preview:
- Exact model version. Microsoft's announcement doesn't name which specific Grok model — Grok 4.6 or a newer variant — is powering this integration. Given xAI has reportedly delayed its next major release around the same week (with Musk saying it "needs a few more days to cook"), don't assume this integration necessarily reflects xAI's most current model.
- Pricing implications. Whether selecting Grok inside Copilot carries different token costs, consumption limits, or licensing implications compared to the default GPT-family or Claude options isn't detailed in the initial announcement — check your organization's specific Frontier Program terms before assuming cost parity across model choices.
- Expansion timeline. Microsoft says it will use preview feedback to decide on expanding Grok to more Copilot surfaces, but no specific timeline or criteria for that decision has been published.
Related reading
- Update — September 25, 2026: Microsoft relaunched Copilot around it — Word, Excel, and PowerPoint are now built directly into the Copilot app itself, not just accessible from it: Copilot's Home, Code, and Autopilot relaunch, explained →
- Grok Can Read, Write, and Send Outlook Mail on First Connect
- GitHub Copilot Hydrafusion: Multi-Model Orchestration
- SpaceXAI's Grok Bot Marketplace Launch
- Microsoft MAI Image 2.6 Flash Launch
This post reflects Microsoft's September 12, 2026 announcement of the Frontier Program preview. Availability, eligibility, and geographic restrictions may change as the preview progresses — check Microsoft's own Copilot documentation for the current rollout status before assuming access.