Microsoft Adds Grok to Copilot — But the Default Setting Tells the Real Story

SpaceXAI’s Grok is now selectable inside Word, Excel, and PowerPoint. Anthropic’s models were already there, on by default. Those two sentences are not the same thing.

What Happened

On 12 September 2026, Microsoft announced it is adding Grok models from SpaceXAI to Copilot in Word, Excel, and PowerPoint, delivered through its Microsoft Frontier program as a preview for eligible Frontier customers. The announcement is straightforward. The conditions attached to it are where the analysis begins.

The capability sits behind a separate administrator setting that is disabled by default. SpaceXAI has been added to Microsoft’s Online Services subprocessor list, and administrators retain control over data processed by SpaceXAI models in supported Copilot experiences. The preview is not offered in the European Union, the European Free Trade Association, or the United Kingdom. Those five facts, taken together, define what SpaceXAI has and has not acquired.

Now set that against the third-party model already living in the same three applications. Microsoft’s admin documentation for the Anthropic rollout — posted in April 2026 and effective 4 May for Excel and PowerPoint, with Word added over the summer — states that “this setting is enabled by default for your tenant” and that “Anthropic models will be available by default for Copilot in Word, Excel, and PowerPoint.” Administrators “can review or change this setting at any time.” Both vendors are Microsoft subprocessors under the same Product Terms and Data Protection Addendum. The contractual category is identical. The default setting is not.

The Structural Read

“Model choice” sounds like a level playing field — several vendors on one menu, the customer selects whichever suits the task. That framing obscures what actually drives outcomes in enterprise software. As a general characteristic of the category, the great majority of tenants never change a default administrator setting. Nobody audits the model picker in a working spreadsheet. Nobody files a change request for a configuration that is causing no problems.

So the variable that carries economic weight is not whether a provider appears on the menu. It is whether the provider is the pre-selected option. Anthropic entered this surface opt-out. Grok enters it opt-in. Both statements are accurate: both models are available in Copilot. The two positions are not remotely the same distribution asset. Opt-out reaches every tenant that does nothing. Opt-in reaches only those tenants where an administrator makes an affirmative choice for a product that has not yet been evaluated, on a surface where the existing configuration is functioning. That is a high bar in enterprise IT, where the reward for changing a working configuration is modest and the blame for breaking one is not.

The same checkbox is also doing a second job, which is why it deserves more attention than the announcement warrants on its own. Off by default means Microsoft requires an affirmative act from the customer before a newly added subprocessor touches tenant data: the tenant makes the decision and owns it. On by default means Microsoft made the decision, and an administrator must act in order to undo it. Distribution lever and liability posture, expressed through one setting.

The regional split is the sharpest contrast in the announcement. Anthropic’s models are offered to European, EFTA, and UK tenants with the disclosure that processing occurs “outside of the European Union Data Boundary (EUDB)” when the setting is enabled. Grok is not offered in those regions at all during the preview. Microsoft has shipped one third-party model into Europe with a caveat and withheld another entirely, and it has published no reason for the difference.

A plausible reading — and this should be read as an interpretation, not a rationale anyone has stated — is that onboarding a new subprocessor for European tenants carries compliance and data-residency work that a preview does not yet justify, and that excluding the region during preview defers that work cheaply. The EU/EFTA/UK exclusion is stated as applying during the preview; it is not presented as permanent, and regional availability can change as compliance work is completed. Separately: note that Anthropic’s arrangement was documented in April and effective in May, with a further application added months later. A default written in May is not a judgement about a vendor onboarded in September. The two rollouts are four months apart, each arrived under the conditions available at the time, and timelines alone do not explain the differences in default position.

Three Implications

The Bottom Line

SpaceXAI is inside the most widely used productivity software on earth, which is not nothing — it is precisely the kind of distribution most AI companies never reach. But it is inside that software as an opt-in, in a preview program, unavailable to European tenants during that preview, while Anthropic’s models run opt-out across the same three applications in most of the world. The announcement and the analysis are both true at once: this is a meaningful foothold, and the foothold is not yet distribution. The platform owner writes the default. Until that setting changes, the distinction is the entire story.

Microsoft (Nasdaq: MSFT) is publicly listed. Anthropic is a private company. This article is business analysis only and does not constitute investment advice. No view is expressed on any security. Nothing in this piece predicts general availability, European availability, or any change to a default setting for any vendor.

Sources: Microsoft Tech Community — Expanding Model Choice in Copilot with Grok (12 Sept 2026) · Microsoft Admin Documentation, Anthropic in Copilot (April 2026, effective 4 May 2026)