Grok falls flat in Washington, undercutting SpaceX’s AI growth story
SpaceX’s initial public offering is set to be the largest in history, partly fueled by its promise to grab a chunk of what it calls a multi-trillion-dollar market for artificial intelligence services through its AI startup, xAI.
But xAI’s Grok chatbot has been a flop with one of the world’s largest customers – the U.S. government, according to seven federal employees, three contracting experts and a Reuters review of government AI inventory documents.
The Reuters Iran Briefing newsletter keeps you informed with the latest developments and analysis of the Iran war. Sign up here.
The 2025 consolidated inventory records from federal agencies show more than 400 publicly identified examples of AI use in government that name a specific vendor. Of them, only three involve the use of xAI or Grok.
By contrast, 234 examples involved technology based on OpenAI’s models, including ChatGPT, Codex, and Microsoft Copilot; 33 involved Gemini or other Alphabet products; and 26 involved Anthropic’s Claude, which has since been blacklisted by the Trump administration.
The Office of Management and Budget (OMB), which collated the records, did not respond to requests for comment. xAI did not respond to detailed questions from Reuters about Grok’s use in government.
Most other AI companies, like Grok, that have been available to federal agencies through the General Services Administration (GSA), did not respond to requests for comment on the data. Google declined to comment and referred Reuters to blog posts, opening a new tab highlighting its government work.
Grok has been available to federal agencies for eight months at a cost of 42 cents per agency. That near-zero pricing, which is also used by xAI’s competitors, is a typical strategy tech giants use to entice government agencies to adopt their products, so they can lock them into higher-priced contracts later.
“The goal is to encourage adoption so that federal employees eventually can’t imagine doing their jobs without generative AI,” said Valerie Wirtschafter, a researcher at the Brookings Institution who studies AI adoption in the federal government.
The OMB data raise questions about whether Grok can take AI market share from leaders such as Claude or ChatGPT, and help justify SpaceX’s ambitious $1.75 trillion IPO valuation.
In a recent regulatory filing, SpaceX said it expects to make far more money building AI for large companies and other big organisations – a total market opportunity it values at $26.5 trillion – than from any of its other businesses.
The U.S. government’s lack of enthusiasm for Grok is a “canary in the coal mine,” casting doubt on SpaceX’s soaring ambitions for broad adoption, said Vineet Jain, co-founder and CEO of Egnyte, which makes AI-powered software for enterprise companies.
“It suggests the model lacks the security rigour required at the federal level, which will be a red flag” for some corporate buyers, Jain said. “Without government validation, the $1.75 trillion valuation looks less like a floor and more like a high ceiling.”
SpaceX chief executive Elon Musk has publicly touted Grok’s potential for federal work and lobbied for its widespread adoption.
In a September announcement of Grok’s deal with the GSA, he said his team wanted to work with President Donald Trump to “rapidly deploy AI throughout the government for the benefit of the country.”
Musk’s Department of Government Efficiency (DOGE) actively promoted Grok.
The now-defunct entity told Department of Homeland Security officials to use Grok, for example, even though it had not been approved for use at the sprawling agency, Reuters reported at the time.
GOVERNMENT GROK USAGE
The AI inventory data collected by OMB provides a window into how federal agencies deploy the technology.
The data typically describes how the tools are used and how many employees use them. Some of the uses are mundane, like categorising incoming emails or transcribing meetings.
Other, more sophisticated uses involve detecting fraud or space research. National security-related use cases are typically omitted.
The data has some inconsistencies. In many cases, the specific AI service used was left blank on forms. Wirtschafter, the Brookings researcher, cautioned that there were differences in how agencies defined an AI use case. Still, she said, the database was the “most comprehensive non-military, non-intelligence inventory of AI use cases we have.”
At the Office of Personnel Management and the Department of Health and Human Services, the chatbot was used for low-level tasks such as generating first drafts of documents or posting to social media, the data showed.
HHS didn’t return messages about its use of AI. A spokesperson for OPM said Microsoft Copilot is the most commonly used AI tool at the agency.
A second part of the AI inventory, which focuses on more ambitious applications which are used by fewer people, also shows little trace of Grok.
The only three references to Grok in that data, showed that Grok had been deployed “in a limited test or pilot capacity” at the Energy Department’s Lawrence Livermore National Laboratory and the Election Assistance Commission. By contrast, OpenAI and Microsoft together accounted for 140 use cases.
The Energy Department didn’t return messages. The EAC said in a statement that its evaluation was “ongoing.”
The inventory data excludes the Pentagon, which has a $200 million deal with xAI. Earlier this year, Defence Secretary Pete Hegseth announced the addition of Grok to GenAI.mil, the military’s unclassified hub for the use of AI models.
In May, xAI became one of seven companies to deploy on the Defence Department’s classified networks.
One Pentagon source with direct knowledge of the matter said many staffers preferred competitors’ AI tools over Grok.
At the Defence Advanced Research Projects Agency (DARPA), the Pentagon’s research and development arm, Google’s Gemini is used for engineering analysis, while Anthropic’s Claude is preferred for coding, writing and research, the source said. OpenAI was also used, the person said, but Grok was generally not.
Claude or Gemini are used within the more sophisticated engineering circles at DARPA, the person added, in part because Grok is “just not the best model out there,” he said.
The Pentagon and DARPA did not respond to requests for comment.
SIGN OF WEAKNESS WITH CORPORATE CUSTOMERS?
SpaceX is still fighting to make inroads. The company’s AI subsidiary, xAI, recently began pursuing FedRAMP High Authorisation – a kind of seal of approval for sensitive government work – with the help of the U.S. Department of Agriculture.
But three USDA information technology professionals said they were unaware that Grok was being used. The USDA said it was “proud to sponsor Grok” but didn’t respond to a question about how often the chatbot was used.
Last month, xAI lost a bid to build a Grok-powered product for the Department of Veterans Affairs, according to a person familiar with the matter. The person said the chatbot hadn’t met the department’s requirements.
Veterans Affairs didn’t directly address questions about its use of Grok.
The low usage within the federal government echoes data indicating that Grok is failing to break into the business world more broadly.
In a report published last year, the web traffic monitoring firm Netskope – which tracks how its thousands of corporate customers connect to AI models – said that Grok had “failed to gain significant traction” in corporate environments.
Updated figures that Netskope provided to Reuters showed that Grok enterprise usage had fallen even further, to 2 out of every 1,000 users, down from a peak of 5 out of every 1,000 users. Netskope executive Ray Canzanese said that even the employees who used Grok spent less time with the chatbot than its competitors – less than half the time that ChatGPT users spent with OpenAI’s model, for example.
Canzanese said the Grok usage data told him the chatbot “is just not going to enter the mainstream for corporate America.”
المصدر: MyJoyOnline (GH)