Reuters: Grok barely used in Washington despite near-free pricing

Reuters has surfaced a number that matters more than another model benchmark: the U.S. government is barely using Grok.

According to Reuters' review of 2025 federal AI inventories, more than 400 public AI use cases identify a named vendor. Only three involve xAI or Grok. By comparison, 234 involve OpenAI technology, including ChatGPT, Codex and Microsoft Copilot. 33 involve Gemini or other Alphabet products. 26 involve Anthropic's Claude.

This is not a pricing problem. Reuters reports that Grok has been available to federal agencies for eight months at 42 cents per agency. Near-zero pricing is a familiar enterprise and public-sector AI tactic: get the tool in, create habits, then convert usage into larger contracts.

The problem is that the habit does not seem to be forming.

Reuters says the story is based on seven federal employees, three contracting experts and a review of AI inventory documents collated by the Office of Management and Budget. xAI did not respond to detailed Reuters questions. OMB did not respond either. Google declined to comment and pointed Reuters to public-sector blog posts.

Why this matters

For executives outside the United States, the point is not simply that Elon Musk has hit resistance in Washington. The larger lesson is that the AI market is not decided by hype, price or capital alone.

In regulated and large organizations, model choice quickly becomes a question of trust, documentation, security, integration and actual workflow value. A model can be cheap, visible and politically promoted, and still lose when users and procurement teams assess risk and utility.

That is a governance signal. Many companies are now standardizing on one or a small set of AI platforms. Choosing that platform based on demos, investor narratives or a chief executive's personal preference is dangerous. Boards and CIOs need to look at usage data, approvals, security posture, logging, data handling, adoption and how well the tool fits into existing work.

Reuters also reports that Grok's weakness is not limited to civilian federal agencies. A Pentagon source with direct knowledge told Reuters that many staff preferred competing AI tools. At DARPA, Google's Gemini is used for engineering analysis, while Claude is preferred for coding, writing and research, according to the same source. Grok was described as less attractive in the more sophisticated technical circles.

That does not mean xAI is out of government. Reuters notes that Grok has been added to GenAI.mil, that xAI has a $200 million Pentagon deal, and that the company became one of several providers selected in May to deploy on classified defense networks. But access is not adoption. A vendor can be listed without becoming the practical default.

The IPO story gets a reality check

The story also lands directly on SpaceX. As part of its planned IPO, SpaceX has pointed to AI as a much larger business opportunity than rockets. Reuters cites a regulatory filing in which SpaceX describes AI for large companies and organizations as a $26.5 trillion market opportunity.

That makes public-sector and enterprise adoption of Grok more than a product metric. It becomes evidence for the entire growth narrative.

Reuters quotes Vineet Jain of Egnyte calling low government usage a “canary in the coal mine.” His argument is that lack of government validation could become a red flag for some corporate buyers, especially where security and auditability matter.

For CIOs and boards, the lesson is blunt: AI vendors often sell future platform power. Buyers must measure current operational value.

Three questions before standardizing

The first question is not which model looks smartest in a closed benchmark. It is which processes the model is actually used in after the pilot phase.

The second is whether the vendor has mature enough security and governance for the environment it wants to enter. FedRAMP, data retention, identity, audit logs and incident handling sound boring. That is exactly why they often decide who gets embedded.

The third is whether the organization can switch models without tearing up its workflow. Once an AI tool is tightly connected to documents, code, identity, collaboration surfaces and approval flows, switching cost and lock-in risk become real long before the contract becomes large.

The Grok case shows a more mature phase of the AI market. Vendors are no longer judged only by what they can demonstrate. They are judged by who actually uses them, for which work, and under what control regime.

That is less spectacular than a new model video. It is also much more important for companies trying to turn AI into production, not theater.

Sources and media

• Primary source: Reuters, “Exclusive: Grok falls flat in Washington, undercutting SpaceX's AI growth story”, published May 21, 2026 at 10:02 UTC and updated at 10:19 UTC: https://www.reuters.com/world/grok-falls-flat-washington-undercutting-spacexs-ai-growth-story-2026-05-21/
• Background data: Office of Management and Budget, 2025 Federal Agency AI Use Case Inventory: https://github.com/ombegov/2025-Federal-Agency-AI-Use-Case-Inventory/tree/main
• Reuters says the article is based on seven federal employees, three contracting experts and a review of U.S. AI inventory documents. xAI and OMB did not respond to Reuters' detailed questions.
• Reuters uses an illustration of xAI and Grok logos by Reuters/Dado Ruvic. The image is not reused by hogby.ai.
• Thumbnail: OpenAI Image 2 / hogby.ai.