Skip to content
News & Analysis

Grok in Word is a subprocessor, not a muse

On 12 September Microsoft added Grok to Copilot in Word, Excel, and PowerPoint. SpaceXAI is on the subprocessor list. The admin switch is off by default.

GS

Published September 15, 2026· Updated Sep 18, 2026

On 12 September Microsoft said Copilot is adding Grok models from SpaceXAI. The post was updated 14 September. SpaceXAI is now on Microsoft’s Online Services Subprocessor List. Access is a dedicated admin setting, disabled by default. It is rolling out through the Frontier Program in Word, Excel, and PowerPoint. It is not available to Frontier customers in the EU, EFTA, or the United Kingdom during preview.

That is a procurement event. It is not a writing event. If your tenant already pays for Copilot, you now have another model family behind a switch. If your tenant does not pay for Copilot, this is not a reason to start. A new subprocessor does not change the tax.

Four honest ways to read it

  • The product read: model choice can be useful when one model is better at a particular spreadsheet, drafting, or presentation task.
  • The procurement read: enabling the preview adds another company to the service chain. That matters to vendor review, data maps, and contract owners even if the button looks small.
  • The competitive read: Microsoft is making Copilot a routing layer across model families, not insisting every Office task use one Microsoft-backed model.
  • The skeptical read: another model in a menu is not evidence of better work. Without task-level comparisons, it is optionality dressed as an outcome.

What is real

The preview has concrete limits: it is controlled by a dedicated administrator setting that starts disabled, it runs through the Frontier program, and it is unavailable in the EU, EFTA, and United Kingdom during preview. Microsoft says the models passed its safety and responsible-AI review. SpaceXAI also appears in Microsoft’s published subprocessor materials. Those are facts. A productivity gain is not yet one of them.

The skeptical read

A preview announcement rarely supplies blinded comparisons, failure rates, latency, or cost per accepted output. The most enthusiastic early users will also be the least representative. If a team cannot describe the Office task that its current model fails, “more choice” can become one more setting to govern and one more result to review.

What other evidence is saying

What to watch next

Watch general availability, regional expansion, exact model versions, retention and data-boundary documentation, admin reporting, and whether Microsoft publishes task-level quality or safety comparisons. Inside your tenant, test the same ten real documents against the current model and count accepted outputs, edits, latency, and escalations.

Skip this seat if

Skip enabling Grok if you cannot name the job Word-plus-OpenAI is already failing. Skip it if counsel has not read the subprocessor addition. Skip it if the team is in London or Frankfurt and you were about to demo a preview they cannot use. Skip buying Copilot because Grok landed.

Weekly Newsletter

AI Adoption Weekly

New research, field guides, training studies, and tool decisions for operators.

No spam. Unsubscribe anytime.

Related Comparisons

Calculator

AI seat cost calculator

List price × headcount. You enter the hours and the operating assumptions.

Open calculator